Skip to content

P
pytensor
提交 ef279e19 authored 作者: Brandon T. Willard's avatar Brandon T. Willard
浏览文件
操作

Replace theano.tensor alias T with tt in tests.tensor

上级 ac7b5225
隐藏空白字符变更
内嵌 并排
正在显示 包含 892 行增加 和 912 行删除
tests/tensor/test_basic.py
浏览文件 @ ef279e19
......@@ -11,6 +11,7 @@ import pytest
import numpy as np
import theano
import theano.tensor as tt
from tempfile import mkstemp
from copy import copy, deepcopy
......@@ -23,7 +24,7 @@ from numpy.testing import assert_array_equal, assert_allclose, assert_almost_equ
from theano import change_flags
from theano.compat import exc_message, operator_div
from theano import compile, config, function, gof, tensor, shared
from theano import compile, config, function, gof, shared
from theano.compile import DeepCopyOp
from theano.compile.mode import get_default_mode
from theano.scalar import autocast_float_as, autocast_float
......@@ -659,7 +660,7 @@ def makeTester(
out_grad_vars = []
for out in expecteds:
if str(out.dtype) in tensor.discrete_dtypes:
if str(out.dtype) in tt.discrete_dtypes:
dtype = floatX
else:
dtype = str(out.dtype)
......@@ -736,11 +737,11 @@ def randc128_ranged(min, max, shape):
def rand_of_dtype(shape, dtype):
if dtype in tensor.discrete_dtypes:
if dtype in tt.discrete_dtypes:
return randint(*shape).astype(dtype)
elif dtype in tensor.float_dtypes:
elif dtype in tt.float_dtypes:
return rand(*shape).astype(dtype)
elif dtype in tensor.complex_dtypes:
elif dtype in tt.complex_dtypes:
return randcomplex(*shape).astype(dtype)
else:
raise TypeError()
......@@ -947,8 +948,8 @@ TestMaximumInplaceBroadcast = makeBroadcastTester(
def test_maximum_minimum_grad():
# Test the discontinuity point.
# We decided that we only pass the gradient to the first input in that case.
x, y = tensor.vectors("xy")
for op in [tensor.maximum, tensor.minimum]:
x, y = tt.vectors("xy")
for op in [tt.maximum, tt.minimum]:
o = op(x, y)
g = theano.grad(o.sum(), [x, y])
......@@ -1076,13 +1077,13 @@ def _numpy_true_div(x, y):
# because simply calling np.true_divide could cause a dtype mismatch.
out = np.true_divide(x, y)
# Use floatX as the result of int / int
if x.dtype in tensor.discrete_dtypes and y.dtype in tensor.discrete_dtypes:
if x.dtype in tt.discrete_dtypes and y.dtype in tt.discrete_dtypes:
out = theano._asarray(out, dtype=config.floatX)
return out
TestTrueDivBroadcast = makeBroadcastTester(
op=tensor.true_div,
op=tt.true_div,
expected=_numpy_true_div,
good=_good_broadcast_div_mod_normal_float_no_complex,
grad=_grad_broadcast_div_mod_normal,
......@@ -1128,7 +1129,7 @@ _bad_runtime_inv = dict(
TestInvBroadcast = makeBroadcastTester(
op=tensor.inv,
op=tt.inv,
expected=lambda x: upcast_int8_nfunc(np.true_divide)(np.int8(1), x),
good=_good_inv,
bad_runtime=_bad_runtime_inv,
......@@ -1147,7 +1148,7 @@ TestInvInplaceBroadcast = makeBroadcastTester(
TestCeilIntDivBroadcast = makeBroadcastTester(
op=tensor.ceil_intdiv,
op=tt.ceil_intdiv,
expected=lambda x, y: check_floatX((x, y), (x // y) + ((x % y) != 0)),
good=_good_broadcast_div_mod_normal_float_no_complex,
name="CeilIntDiv",
......@@ -1157,7 +1158,7 @@ TestCeilIntDivBroadcast = makeBroadcastTester(
)
TestModBroadcast = makeBroadcastTester(
op=tensor.mod,
op=tt.mod,
expected=lambda x, y: np.asarray(
x % y, dtype=theano.scalar.basic.upcast(x.dtype, y.dtype)
),
......@@ -1331,7 +1332,7 @@ _grad_broadcast_unary_0_2_no_complex = dict(
# don't have a well defined behavior. We don't test that case.
TestAbsBroadcast = makeBroadcastTester(
op=tensor.abs_,
op=tt.abs_,
expected=lambda x: abs(x),
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -1347,7 +1348,7 @@ TestAbsInplaceBroadcast = makeBroadcastTester(
)
TestNegBroadcast = makeBroadcastTester(
op=tensor.neg,
op=tt.neg,
expected=lambda x: -x,
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -1360,7 +1361,7 @@ TestNegInplaceBroadcast = makeBroadcastTester(
)
TestSgnBroadcast = makeBroadcastTester(
op=tensor.sgn,
op=tt.sgn,
expected=np.sign,
good=_good_broadcast_unary_normal_no_complex,
grad=_grad_broadcast_unary_normal,
......@@ -1373,7 +1374,7 @@ TestSgnInplaceBroadcast = makeBroadcastTester(
)
TestIntDivBroadcast = makeBroadcastTester(
op=tensor.int_div,
op=tt.int_div,
expected=lambda x, y: check_floatX((x, y), x // y),
good=_good_broadcast_div_mod_normal_float,
# I don't test the grad as the output is always an integer
......@@ -1393,7 +1394,7 @@ TestIntDivInplaceBroadcast = makeBroadcastTester(
TestCeilBroadcast = makeBroadcastTester(
op=tensor.ceil,
op=tt.ceil,
expected=upcast_float16_ufunc(np.ceil),
good=_good_broadcast_unary_normal_no_complex,
grad=copymod(
......@@ -1415,7 +1416,7 @@ TestCeilInplaceBroadcast = makeBroadcastTester(
)
TestFloorBroadcast = makeBroadcastTester(
op=tensor.floor,
op=tt.floor,
expected=upcast_float16_ufunc(np.floor),
good=_good_broadcast_unary_normal_no_complex,
grad=_grad_broadcast_unary_normal_noint,
......@@ -1439,13 +1440,13 @@ TestTruncInplaceBroadcast = makeBroadcastTester(
)
TestTruncBroadcast = makeBroadcastTester(
op=tensor.trunc,
op=tt.trunc,
expected=upcast_float16_ufunc(np.trunc),
good=_good_broadcast_unary_normal_no_complex,
)
TestRoundHalfToEvenBroadcast = makeBroadcastTester(
op=tensor.round_half_to_even,
op=tt.round_half_to_even,
expected=np.round,
good=_good_broadcast_unary_normal_float_no_complex,
grad=_grad_broadcast_unary_normal_no_complex_no_corner_case,
......@@ -1462,7 +1463,7 @@ TestRoundHalfToEvenInplaceBroadcast = makeBroadcastTester(
# see in their file numpy/lib/function_base.py in class vectorize.__call__
# This happen in float32 mode.
TestRoundHalfAwayFromZeroBroadcast = makeBroadcastTester(
op=tensor.round_half_away_from_zero,
op=tt.round_half_away_from_zero,
expected=lambda a: theano.scalar.basic.round_half_away_from_zero_vec(a),
good=_good_broadcast_unary_normal_float_no_empty_no_complex,
grad=_grad_broadcast_unary_normal_no_complex_no_corner_case,
......@@ -1476,7 +1477,7 @@ TestRoundHalfAwayFromZeroInplaceBroadcast = makeBroadcastTester(
)
TestSqrBroadcast = makeBroadcastTester(
op=tensor.sqr,
op=tt.sqr,
expected=np.square,
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -1490,7 +1491,7 @@ TestSqrInplaceBroadcast = makeBroadcastTester(
)
TestExpBroadcast = makeBroadcastTester(
op=tensor.exp,
op=tt.exp,
expected=upcast_float16_ufunc(np.exp),
good=dict(
_good_broadcast_unary_normal,
......@@ -1508,7 +1509,7 @@ TestExpInplaceBroadcast = makeBroadcastTester(
)
TestExp2Broadcast = makeBroadcastTester(
op=tensor.exp2,
op=tt.exp2,
expected=upcast_float16_ufunc(np.exp2),
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -1522,7 +1523,7 @@ TestExp2InplaceBroadcast = makeBroadcastTester(
TestExpm1Broadcast = makeBroadcastTester(
op=tensor.expm1,
op=tt.expm1,
expected=upcast_float16_ufunc(np.expm1),
good=dict(
_good_broadcast_unary_normal,
......@@ -1557,7 +1558,7 @@ _grad_broadcast_unary_positive = dict(
)
TestLogBroadcast = makeBroadcastTester(
op=tensor.log,
op=tt.log,
expected=upcast_float16_ufunc(np.log),
good=_good_broadcast_unary_positive,
grad=_grad_broadcast_unary_positive,
......@@ -1570,7 +1571,7 @@ TestLogInplaceBroadcast = makeBroadcastTester(
)
TestLog2Broadcast = makeBroadcastTester(
op=tensor.log2,
op=tt.log2,
expected=upcast_float16_ufunc(np.log2),
good=_good_broadcast_unary_positive,
grad=_grad_broadcast_unary_positive,
......@@ -1583,7 +1584,7 @@ TestLog2InplaceBroadcast = makeBroadcastTester(
)
TestLog10Broadcast = makeBroadcastTester(
op=tensor.log10,
op=tt.log10,
expected=upcast_float16_ufunc(np.log10),
good=_good_broadcast_unary_positive,
grad=_grad_broadcast_unary_positive,
......@@ -1596,7 +1597,7 @@ TestLog10InplaceBroadcast = makeBroadcastTester(
)
TestLog1pBroadcast = makeBroadcastTester(
op=tensor.log1p,
op=tt.log1p,
expected=upcast_float16_ufunc(np.log1p),
good=_good_broadcast_unary_positive,
grad=_grad_broadcast_unary_positive,
......@@ -1609,7 +1610,7 @@ TestLog1pInplaceBroadcast = makeBroadcastTester(
)
TestSqrtBroadcast = makeBroadcastTester(
op=tensor.sqrt,
op=tt.sqrt,
expected=upcast_float16_ufunc(np.sqrt),
good=_good_broadcast_unary_positive,
grad=_grad_broadcast_unary_positive,
......@@ -1643,7 +1644,7 @@ else:
angle_eps = 1e-10
TestDeg2radBroadcast = makeBroadcastTester(
op=tensor.deg2rad,
op=tt.deg2rad,
expected=upcast_float16_ufunc(np.deg2rad),
good=_good_broadcast_unary_normal_no_complex,
grad=_grad_broadcast_unary_normal_no_complex,
......@@ -1658,7 +1659,7 @@ TestDeg2radInplaceBroadcast = makeBroadcastTester(
)
TestRad2degBroadcast = makeBroadcastTester(
op=tensor.rad2deg,
op=tt.rad2deg,
expected=upcast_float16_ufunc(np.rad2deg),
good=_good_broadcast_unary_normal_no_complex,
grad=_grad_broadcast_unary_normal_no_complex,
......@@ -1673,7 +1674,7 @@ TestRad2degInplaceBroadcast = makeBroadcastTester(
)
TestSinBroadcast = makeBroadcastTester(
op=tensor.sin,
op=tt.sin,
expected=upcast_float16_ufunc(np.sin),
good=_good_broadcast_unary_wide,
grad=_grad_broadcast_unary_wide,
......@@ -1706,7 +1707,7 @@ _grad_broadcast_unary_arcsin = dict(
)
TestArcsinBroadcast = makeBroadcastTester(
op=tensor.arcsin,
op=tt.arcsin,
expected=upcast_float16_ufunc(np.arcsin),
good=_good_broadcast_unary_arcsin,
grad=_grad_broadcast_unary_arcsin,
......@@ -1719,7 +1720,7 @@ TestArcsinInplaceBroadcast = makeBroadcastTester(
)
TestCosBroadcast = makeBroadcastTester(
op=tensor.cos,
op=tt.cos,
expected=upcast_float16_ufunc(np.cos),
good=_good_broadcast_unary_wide,
grad=_grad_broadcast_unary_wide,
......@@ -1733,14 +1734,14 @@ TestCosInplaceBroadcast = makeBroadcastTester(
def test_py_c_match():
a = tensor.TensorType(dtype="int8", broadcastable=(False,))()
f = theano.function([a], tensor.arccos(a), mode="DebugMode")
a = tt.TensorType(dtype="int8", broadcastable=(False,))()
f = theano.function([a], tt.arccos(a), mode="DebugMode")
# This can fail in DebugMode
f(np.asarray([1, 0, -1], dtype="int8"))
TestArccosBroadcast = makeBroadcastTester(
op=tensor.arccos,
op=tt.arccos,
expected=upcast_float16_ufunc(np.arccos),
good=_good_broadcast_unary_arcsin,
grad=_grad_broadcast_unary_arcsin,
......@@ -1768,7 +1769,7 @@ _grad_broadcast_unary_tan = dict(
)
TestTanBroadcast = makeBroadcastTester(
op=tensor.tan,
op=tt.tan,
expected=upcast_float16_ufunc(np.tan),
good=_good_broadcast_unary_tan,
grad=_grad_broadcast_unary_tan,
......@@ -1784,7 +1785,7 @@ TestTanInplaceBroadcast = makeBroadcastTester(
)
TestArctanBroadcast = makeBroadcastTester(
op=tensor.arctan,
op=tt.arctan,
expected=upcast_float16_ufunc(np.arctan),
good=_good_broadcast_unary_wide,
grad=_grad_broadcast_unary_wide,
......@@ -1828,7 +1829,7 @@ _grad_broadcast_binary_arctan2 = dict(
)
TestArctan2Broadcast = makeBroadcastTester(
op=tensor.arctan2,
op=tt.arctan2,
expected=upcast_float16_ufunc(np.arctan2),
good=_good_broadcast_binary_arctan2,
grad=_grad_broadcast_binary_arctan2,
......@@ -1845,7 +1846,7 @@ TestArctan2InplaceBroadcast = makeBroadcastTester(
)
TestCoshBroadcast = makeBroadcastTester(
op=tensor.cosh,
op=tt.cosh,
expected=upcast_float16_ufunc(np.cosh),
good=dict(
_good_broadcast_unary_normal,
......@@ -1874,7 +1875,7 @@ _grad_broadcast_unary_arccosh = dict(
)
TestArccoshBroadcast = makeBroadcastTester(
op=tensor.arccosh,
op=tt.arccosh,
expected=upcast_float16_ufunc(np.arccosh),
good=_good_broadcast_unary_arccosh,
grad=_grad_broadcast_unary_arccosh,
......@@ -1887,7 +1888,7 @@ TestArccoshInplaceBroadcast = makeBroadcastTester(
)
TestSinhBroadcast = makeBroadcastTester(
op=tensor.sinh,
op=tt.sinh,
expected=upcast_float16_ufunc(np.sinh),
good=dict(
_good_broadcast_unary_normal,
......@@ -1905,7 +1906,7 @@ TestSinhInplaceBroadcast = makeBroadcastTester(
)
TestArcsinhBroadcast = makeBroadcastTester(
op=tensor.arcsinh,
op=tt.arcsinh,
expected=upcast_float16_ufunc(np.arcsinh),
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -1918,7 +1919,7 @@ TestArcsinhInplaceBroadcast = makeBroadcastTester(
)
TestTanhBroadcast = makeBroadcastTester(
op=tensor.tanh,
op=tt.tanh,
expected=upcast_float16_ufunc(np.tanh),
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -1944,7 +1945,7 @@ _grad_broadcast_unary_arctanh = dict(
)
TestArctanhBroadcast = makeBroadcastTester(
op=tensor.arctanh,
op=tt.arctanh,
expected=upcast_float16_ufunc(np.arctanh),
good=_good_broadcast_unary_arctanh,
grad=_grad_broadcast_unary_arctanh,
......@@ -2001,7 +2002,7 @@ else:
skip_scipy = "scipy is not present"
TestErfBroadcast = makeBroadcastTester(
op=tensor.erf,
op=tt.erf,
expected=expected_erf,
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -2020,7 +2021,7 @@ TestErfInplaceBroadcast = makeBroadcastTester(
)
TestErfcBroadcast = makeBroadcastTester(
op=tensor.erfc,
op=tt.erfc,
expected=expected_erfc,
good=_good_broadcast_unary_normal_float_no_complex,
grad=_grad_broadcast_unary_normal,
......@@ -2039,7 +2040,7 @@ TestErfcInplaceBroadcast = makeBroadcastTester(
)
TestErfcxBroadcast = makeBroadcastTester(
op=tensor.erfcx,
op=tt.erfcx,
expected=expected_erfcx,
good=_good_broadcast_unary_normal_float_no_complex_small_neg_range,
grad=_grad_broadcast_unary_normal_small_neg_range,
......@@ -2058,7 +2059,7 @@ TestErfcxInplaceBroadcast = makeBroadcastTester(
)
TestErfinvBroadcast = makeBroadcastTester(
op=tensor.erfinv,
op=tt.erfinv,
expected=expected_erfinv,
good={
"normal": [rand_ranged(-0.9, 0.9, (2, 3))],
......@@ -2071,7 +2072,7 @@ TestErfinvBroadcast = makeBroadcastTester(
)
TestErfcinvBroadcast = makeBroadcastTester(
op=tensor.erfcinv,
op=tt.erfcinv,
expected=expected_erfcinv,
good={
"normal": [rand_ranged(0.001, 1.9, (2, 3))],
......@@ -2097,7 +2098,7 @@ _grad_broadcast_unary_gammaln = dict(
)
TestGammaBroadcast = makeBroadcastTester(
op=tensor.gamma,
op=tt.gamma,
expected=expected_gamma,
good=_good_broadcast_unary_gammaln,
grad=_grad_broadcast_unary_gammaln,
......@@ -2116,7 +2117,7 @@ TestGammaInplaceBroadcast = makeBroadcastTester(
)
TestGammalnBroadcast = makeBroadcastTester(
op=tensor.gammaln,
op=tt.gammaln,
expected=expected_gammaln,
good=_good_broadcast_unary_gammaln,
grad=_grad_broadcast_unary_gammaln,
......@@ -2143,7 +2144,7 @@ _good_broadcast_unary_psi = dict(
)
TestPsiBroadcast = makeBroadcastTester(
op=tensor.psi,
op=tt.psi,
expected=expected_psi,
good=_good_broadcast_unary_psi,
eps=2e-10,
......@@ -2163,7 +2164,7 @@ TestPsiInplaceBroadcast = makeBroadcastTester(
_good_broadcast_unary_tri_gamma = _good_broadcast_unary_psi
TestTriGammaBroadcast = makeBroadcastTester(
op=tensor.tri_gamma,
op=tt.tri_gamma,
expected=expected_tri_gamma,
good=_good_broadcast_unary_psi,
eps=2e-8,
......@@ -2198,7 +2199,7 @@ _good_broadcast_unary_chi2sf = dict(
)
TestChi2SFBroadcast = makeBroadcastTester(
op=tensor.chi2sf,
op=tt.chi2sf,
expected=expected_chi2sf,
good=_good_broadcast_unary_chi2sf,
eps=2e-10,
......@@ -2249,7 +2250,7 @@ _grad_broadcast_binary_bessel = dict(
)
TestJ0Broadcast = makeBroadcastTester(
op=tensor.j0,
op=tt.j0,
expected=expected_j0,
good=_good_broadcast_unary_bessel,
grad=_grad_broadcast_unary_bessel,
......@@ -2269,7 +2270,7 @@ TestJ0InplaceBroadcast = makeBroadcastTester(
)
TestJ1Broadcast = makeBroadcastTester(
op=tensor.j1,
op=tt.j1,
expected=expected_j1,
good=_good_broadcast_unary_bessel,
grad=_grad_broadcast_unary_bessel,
......@@ -2289,7 +2290,7 @@ TestJ1InplaceBroadcast = makeBroadcastTester(
)
TestJvBroadcast = makeBroadcastTester(
op=tensor.jv,
op=tt.jv,
expected=expected_jv,
good=_good_broadcast_binary_bessel,
eps=2e-10,
......@@ -2316,13 +2317,13 @@ def test_verify_jv_grad():
v_val, x_val = _grad_broadcast_binary_bessel["normal"]
def fixed_first_input_jv(x):
return tensor.jv(v_val, x)
return tt.jv(v_val, x)
utt.verify_grad(fixed_first_input_jv, [x_val])
TestI0Broadcast = makeBroadcastTester(
op=tensor.i0,
op=tt.i0,
expected=expected_i0,
good=_good_broadcast_unary_bessel,
grad=_grad_broadcast_unary_bessel,
......@@ -2342,7 +2343,7 @@ TestI0InplaceBroadcast = makeBroadcastTester(
)
TestI1Broadcast = makeBroadcastTester(
op=tensor.i1,
op=tt.i1,
expected=expected_i1,
good=_good_broadcast_unary_bessel,
grad=_grad_broadcast_unary_bessel,
......@@ -2362,7 +2363,7 @@ TestI1InplaceBroadcast = makeBroadcastTester(
)
TestIvBroadcast = makeBroadcastTester(
op=tensor.iv,
op=tt.iv,
expected=expected_iv,
good=_good_broadcast_binary_bessel,
eps=2e-10,
......@@ -2389,13 +2390,13 @@ def test_verify_iv_grad():
v_val, x_val = _grad_broadcast_binary_bessel["normal"]
def fixed_first_input_iv(x):
return tensor.iv(v_val, x)
return tt.iv(v_val, x)
utt.verify_grad(fixed_first_input_iv, [x_val])
TestZerosLikeBroadcast = makeBroadcastTester(
op=tensor.zeros_like,
op=tt.zeros_like,
expected=np.zeros_like,
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -2403,7 +2404,7 @@ TestZerosLikeBroadcast = makeBroadcastTester(
)
TestOnesLikeBroadcast = makeBroadcastTester(
op=tensor.ones_like,
op=tt.ones_like,
expected=np.ones_like,
good=_good_broadcast_unary_normal,
grad=_grad_broadcast_unary_normal,
......@@ -2428,13 +2429,13 @@ _grad_complex_from_polar = dict(
)
TestComplexFromPolarBroadcast = makeBroadcastTester(
op=tensor.complex_from_polar,
op=tt.complex_from_polar,
expected=lambda r, theta: r * np.cos(theta) + 1j * r * np.sin(theta),
good=_good_complex_from_polar,
)
TestConjBroadcast = makeBroadcastTester(
op=tensor.conj, expected=np.conj, good=_good_broadcast_unary_normal
op=tt.conj, expected=np.conj, good=_good_broadcast_unary_normal
)
TestConjInplaceBroadcast = makeBroadcastTester(
op=inplace.conj_inplace,
......@@ -2751,23 +2752,23 @@ class ApplyDefaultTestOp(theano.Op):
self.default_output = id
def make_node(self, x):
x = theano.tensor.as_tensor_variable(x)
x = tt.as_tensor_variable(x)
return theano.Apply(self, [x], [x.type()])
def test_constant():
int8_vector_type = tensor.TensorType(dtype="int8", broadcastable=(False,))
int8_vector_type = tt.TensorType(dtype="int8", broadcastable=(False,))
# Make sure we return a `TensorConstant` unchanged
x = tensor.TensorConstant(int8_vector_type, [1, 2])
x = tt.TensorConstant(int8_vector_type, [1, 2])
y = constant(x)
assert y is x
# Make sure we can add and remove broadcastable dimensions
int8_scalar_type = tensor.TensorType(dtype="int8", broadcastable=())
int8_scalar_type = tt.TensorType(dtype="int8", broadcastable=())
x_data = np.array(2, dtype="int8")
x = tensor.TensorConstant(int8_scalar_type, x_data)
x = tt.TensorConstant(int8_scalar_type, x_data)
y = constant(x, ndim=1)
assert y.ndim == 1
assert np.array_equal(y.data, np.expand_dims(x_data, 0))
......@@ -2788,7 +2789,7 @@ class TestAsTensorVariable:
"""
def setup_method(self):
self.x = tensor.scalar("x")
self.x = tt.scalar("x")
def test_tensor_from_scalar(self):
y = as_tensor_variable(scal.int8())
......@@ -2811,7 +2812,7 @@ class TestAsTensorVariable:
# mask exceptions caused by unrelated logic (e.g. computing test
# values)
with change_flags(compute_test_value="raise"), pytest.raises(ValueError):
a = tensor.lscalar("a")
a = tt.lscalar("a")
y = (a, a, 1)
_ = as_tensor_variable(y)
......@@ -2820,11 +2821,11 @@ class TestAsTensorVariable:
as_tensor_variable(bad_apply_var)
def test_strip_leading_broadcastable(self):
x = tensor.TensorType(config.floatX, (True, False))("x")
x = tt.TensorType(config.floatX, (True, False))("x")
x = as_tensor_variable(x, ndim=1)
assert x.ndim == 1
x = tensor.matrix("x", dtype=config.floatX)
x = tt.matrix("x", dtype=config.floatX)
with pytest.raises(ValueError):
as_tensor_variable(x, ndim=1)
......@@ -2843,7 +2844,7 @@ class TestAsTensorVariable:
new_inp = np.memmap(fname, dtype=inp.dtype, mode="w+", shape=inp.shape)
new_inp[...] = inp
res = as_tensor_variable(new_inp)
assert isinstance(res, tensor.TensorConstant)
assert isinstance(res, tt.TensorConstant)
assert res.data is new_inp
@pytest.mark.parametrize(
......@@ -2863,26 +2864,24 @@ class TestAsTensorVariable:
("x", "y"),
[
([1, 2], [1, 2]),
([tensor.as_tensor(1), tensor.as_tensor(2)], [1, 2]),
([tt.as_tensor(1), tt.as_tensor(2)], [1, 2]),
([theano.scalar.constant(1), theano.scalar.constant(2)], [1, 2]),
],
)
def test_constant_consistency(self, x, y):
a = as_tensor_variable(x)
assert isinstance(a, tensor.TensorConstant)
assert isinstance(a, tt.TensorConstant)
assert np.array_equal(a.data, y)
def test_constant_identity(self):
# Values that are already `TensorType`s shouldn't be recreated by
# `as_tensor_variable`
x_scalar = tensor.TensorConstant(
tensor.TensorType(dtype="int8", broadcastable=()), 2
)
x_scalar = tt.TensorConstant(tt.TensorType(dtype="int8", broadcastable=()), 2)
a_scalar = as_tensor_variable(x_scalar)
assert x_scalar is a_scalar
x_vector = tensor.TensorConstant(
tensor.TensorType(dtype="int8", broadcastable=(False,)),
x_vector = tt.TensorConstant(
tt.TensorType(dtype="int8", broadcastable=(False,)),
np.array([1, 2], dtype="int8"),
)
a_vector = as_tensor_variable(x_vector)
......@@ -2893,7 +2892,7 @@ class TestAlloc:
dtype = config.floatX
mode = mode_opt
shared = staticmethod(theano.shared)
allocs = [tensor.Alloc()] * 3
allocs = [tt.Alloc()] * 3
def setup_method(self):
self.rng = np.random.RandomState(seed=utt.fetch_seed())
......@@ -2904,7 +2903,7 @@ class TestAlloc:
some_vector = vector("some_vector", dtype=self.dtype)
some_matrix = some_vector.reshape((60, 50))
variables = self.shared(np.ones((50,), dtype=self.dtype))
idx = tensor.constant(np.arange(50))
idx = tt.constant(np.arange(50))
for alloc_, (subtensor, n_alloc) in zip(
self.allocs,
......@@ -2935,7 +2934,7 @@ class TestAlloc:
fgrad(test_params)
def test_alloc_output(self):
val = tensor.constant(self.rng.randn(1, 1), dtype=self.dtype)
val = tt.constant(self.rng.randn(1, 1), dtype=self.dtype)
for alloc_ in self.allocs:
# The output is the result of the alloc operation,
# we do not want it to be constant-folded
......@@ -2948,35 +2947,35 @@ class TestAlloc:
def test_ones(self):
for shp in [[], 1, [1], [1, 2], [1, 2, 3], np.r_[1, 2, 3]]:
ones = theano.function([], [tensor.ones(shp)], mode=self.mode)
ones = theano.function([], [tt.ones(shp)], mode=self.mode)
assert np.allclose(ones(), np.ones(shp))
# scalar doesn't have to be provided as input
x = scalar()
shp = []
ones_scalar = theano.function([], [tensor.ones(x.shape)], mode=self.mode)
ones_scalar = theano.function([], [tt.ones(x.shape)], mode=self.mode)
assert np.allclose(ones_scalar(), np.ones(shp))
for (typ, shp) in [(vector, [3]), (matrix, [3, 4])]:
x = typ()
ones_tensor = theano.function([x], [tensor.ones(x.shape)], mode=self.mode)
ones_tensor = theano.function([x], [tt.ones(x.shape)], mode=self.mode)
inp = np.zeros(shp, dtype=config.floatX)
assert np.allclose(ones_tensor(inp), np.ones(shp))
def test_zeros(self):
for shp in [[], 1, [1], [1, 2], [1, 2, 3], np.r_[1, 2, 3]]:
zeros = theano.function([], [tensor.zeros(shp)], mode=self.mode)
zeros = theano.function([], [tt.zeros(shp)], mode=self.mode)
assert np.allclose(zeros(), np.zeros(shp))
# scalar doesn't have to be provided as input
x = scalar()
shp = []
zeros_scalar = theano.function([], [tensor.zeros(x.shape)], mode=self.mode)
zeros_scalar = theano.function([], [tt.zeros(x.shape)], mode=self.mode)
assert np.allclose(zeros_scalar(), np.zeros(shp))
for (typ, shp) in [(vector, [3]), (matrix, [3, 4])]:
x = typ()
zeros_tensor = theano.function([x], [tensor.zeros(x.shape)], mode=self.mode)
zeros_tensor = theano.function([x], [tt.zeros(x.shape)], mode=self.mode)
inp = np.zeros(shp, dtype=config.floatX)
assert np.allclose(zeros_tensor(inp), np.zeros(shp))
......@@ -2991,9 +2990,9 @@ def test_eye():
# allowed.
if M is None and theano.config.mode in ["DebugMode", "DEBUG_MODE"]:
M = N
N_symb = tensor.iscalar()
M_symb = tensor.iscalar()
k_symb = tensor.iscalar()
N_symb = tt.iscalar()
M_symb = tt.iscalar()
k_symb = tt.iscalar()
f = function([N_symb, M_symb, k_symb], eye(N_symb, M_symb, k_symb, dtype=dtype))
result = f(N, M, k)
assert np.allclose(result, np.eye(N, M_, k, dtype=dtype))
......@@ -3025,9 +3024,9 @@ class TestTriangle:
# allowed.
if M is None and theano.config.mode in ["DebugMode", "DEBUG_MODE"]:
M = N
N_symb = tensor.iscalar()
M_symb = tensor.iscalar()
k_symb = tensor.iscalar()
N_symb = tt.iscalar()
M_symb = tt.iscalar()
k_symb = tt.iscalar()
f = function(
[N_symb, M_symb, k_symb], tri(N_symb, M_symb, k_symb, dtype=dtype)
)
......@@ -3091,9 +3090,7 @@ class TestNonzero:
@change_flags(compute_test_value="raise")
def test_nonzero(self):
def check(m):
m_symb = theano.tensor.tensor(
dtype=m.dtype, broadcastable=(False,) * m.ndim
)
m_symb = tt.tensor(dtype=m.dtype, broadcastable=(False,) * m.ndim)
m_symb.tag.test_value = m
res_tuple_tt = nonzero(m_symb, return_matrix=False)
......@@ -3122,9 +3119,7 @@ class TestNonzero:
@change_flags(compute_test_value="raise")
def test_flatnonzero(self):
def check(m):
m_symb = theano.tensor.tensor(
dtype=m.dtype, broadcastable=(False,) * m.ndim
)
m_symb = tt.tensor(dtype=m.dtype, broadcastable=(False,) * m.ndim)
m_symb.tag.test_value = m
res_tt = flatnonzero(m_symb)
......@@ -3147,9 +3142,7 @@ class TestNonzero:
@change_flags(compute_test_value="raise")
def test_nonzero_values(self):
def check(m):
m_symb = theano.tensor.tensor(
dtype=m.dtype, broadcastable=(False,) * m.ndim
)
m_symb = tt.tensor(dtype=m.dtype, broadcastable=(False,) * m.ndim)
m_symb.tag.test_value = m
res_tt = nonzero_values(m_symb)
......@@ -3173,7 +3166,7 @@ class TestNonzero:
def test_identity():
def check(dtype):
obj = rand_of_dtype((2,), dtype)
sym = tensor.vector(dtype=dtype)
sym = tt.vector(dtype=dtype)
f = function([sym], tensor_copy(sym))
assert np.all(obj == f(obj))
assert obj.dtype == f(obj).dtype
......@@ -3200,8 +3193,8 @@ class TestCast:
],
)
for testname, (obj, dtype) in good:
inp = tensor.vector(dtype=obj.dtype)
out = tensor.cast(inp, dtype=dtype)
inp = tt.vector(dtype=obj.dtype)
out = tt.cast(inp, dtype=dtype)
f = function([inp], out)
assert f(obj).dtype == np.dtype(dtype)
......@@ -3212,8 +3205,8 @@ class TestCast:
def test_cast_from_real_to_complex(self):
for real_dtype in REAL_DTYPES:
for complex_dtype in COMPLEX_DTYPES:
inp = tensor.vector(dtype=real_dtype)
out = tensor.cast(inp, dtype=complex_dtype)
inp = tt.vector(dtype=real_dtype)
out = tt.cast(inp, dtype=complex_dtype)
f = function([inp], out)
obj = rand_of_dtype((2,), real_dtype)
assert f(obj).dtype == np.dtype(complex_dtype)
......@@ -3221,9 +3214,9 @@ class TestCast:
def test_cast_from_complex_to_real_raises_error(self):
for real_dtype in REAL_DTYPES:
for complex_dtype in COMPLEX_DTYPES:
inp = tensor.vector(dtype=real_dtype)
inp = tt.vector(dtype=real_dtype)
with pytest.raises(TypeError):
tensor(tensor.cast(inp, dtype=complex_dtype))
tt.tensor(tt.cast(inp, dtype=complex_dtype))
TestClip = makeTester(
......@@ -3296,15 +3289,15 @@ TestBackwardsClip = makeTester(
class TestClip:
def test_complex_value(self):
for dtype in ["complex64", "complex128"]:
a = tensor.vector(dtype=dtype)
b = tensor.scalar()
c = tensor.scalar()
a = tt.vector(dtype=dtype)
b = tt.scalar()
c = tt.scalar()
with pytest.raises(TypeError):
clip(a, b, c)
def test_clip_repeat_grad(self):
# This is testing for the issue #633
x, y = tensor.vectors("xy")
x, y = tt.vectors("xy")
a = clip(x, y, x)
g = theano.gradient.grad(a.sum(), x)
fn = theano.function([x, y], [g])
......@@ -3315,7 +3308,7 @@ class TestClip:
fn2 = theano.function([x, y], [g2])
# Test for the equal case too
a3 = theano.tensor.clip(x, x, x)
a3 = tt.clip(x, x, x)
g3 = theano.gradient.grad(a3.sum(), x)
fn3 = theano.function([x], [g3])
......@@ -3353,9 +3346,9 @@ class TestClip:
def test_batched_dot():
first = theano.tensor.tensor3("first")
second = theano.tensor.tensor3("second")
output = theano.tensor.basic.batched_dot(first, second)
first = tt.tensor3("first")
second = tt.tensor3("second")
output = tt.basic.batched_dot(first, second)
first_val = np.random.rand(10, 10, 20).astype(config.floatX)
second_val = np.random.rand(10, 20, 5).astype(config.floatX)
result_fn = theano.function([first, second], output)
......@@ -3364,9 +3357,9 @@ def test_batched_dot():
assert result.shape[1] == first_val.shape[1]
assert result.shape[2] == second_val.shape[2]
first_mat = theano.tensor.dmatrix("first")
second_mat = theano.tensor.dmatrix("second")
output = theano.tensor.basic.batched_dot(first_mat, second_mat)
first_mat = tt.dmatrix("first")
second_mat = tt.dmatrix("second")
output = tt.basic.batched_dot(first_mat, second_mat)
first_mat_val = np.random.rand(10, 10).astype(config.floatX)
second_mat_val = np.random.rand(10, 10).astype(config.floatX)
result_fn = theano.function([first_mat, second_mat], output)
......@@ -3406,10 +3399,10 @@ def test_batched_dot_not_contiguous():
def test_batched_tensordot():
first = theano.tensor.tensor4("first")
second = theano.tensor.tensor4("second")
first = tt.tensor4("first")
second = tt.tensor4("second")
axes = [[1, 2], [3, 1]]
output = theano.tensor.basic.batched_tensordot(first, second, axes)
output = tt.batched_tensordot(first, second, axes)
first_val = np.random.rand(8, 10, 20, 3).astype(config.floatX)
second_val = np.random.rand(8, 20, 5, 10).astype(config.floatX)
result_fn = theano.function([first, second], output)
......@@ -3418,10 +3411,10 @@ def test_batched_tensordot():
assert result.shape[1] == first_val.shape[3]
assert result.shape[2] == second_val.shape[2]
first_mat = theano.tensor.dmatrix("first")
second_mat = theano.tensor.dmatrix("second")
first_mat = tt.dmatrix("first")
second_mat = tt.dmatrix("second")
axes = 1
output = theano.tensor.basic.batched_tensordot(first_mat, second_mat, axes)
output = tt.batched_tensordot(first_mat, second_mat, axes)
first_mat_val = np.random.rand(10, 4).astype(config.floatX)
second_mat_val = np.random.rand(10, 4).astype(config.floatX)
result_fn = theano.function([first_mat, second_mat], output)
......@@ -3481,7 +3474,7 @@ def test_nan_inf_constant_signature():
assert (x.signature() == y.signature()) == (i == j)
# Also test that nan !=0 and nan != nan.
x = tensor.scalar()
x = tt.scalar()
mode = get_default_mode()
if isinstance(mode, theano.compile.debugmode.DebugMode):
# Disable the check preventing usage of NaN / Inf values.
......@@ -3495,16 +3488,16 @@ def test_nan_inf_constant_signature():
def test_isnan():
for x in [tensor.matrix(), tensor.imatrix(), tensor.matrix(dtype="bool")]:
y = tensor.isnan(x)
assert isinstance(y.owner.op, tensor.Elemwise) == (
x.dtype not in tensor.discrete_dtypes
for x in [tt.matrix(), tt.imatrix(), tt.matrix(dtype="bool")]:
y = tt.isnan(x)
assert isinstance(y.owner.op, tt.Elemwise) == (
x.dtype not in tt.discrete_dtypes
)
assert y.dtype == "bool"
# Test c code generator even for int type.
y = tensor.isnan_(x)
assert isinstance(y.owner.op, tensor.Elemwise)
y = tt.isnan_(x)
assert isinstance(y.owner.op, tt.Elemwise)
assert y.dtype == "bool"
f = theano.function([x], y, allow_input_downcast=True)
f([[0, 1, 2]])
......@@ -3661,7 +3654,7 @@ class TestMaxAndArgmax:
x = matrix()
cost = argmax(x, axis=0).sum()
gx = grad(cost, x)
val = tensor.get_scalar_constant_value(gx)
val = tt.get_scalar_constant_value(gx)
assert val == 0.0
def test_grad(self):
......@@ -3743,7 +3736,7 @@ class TestMaxAndArgmax:
def test_preserve_broadcastable(self):
# Ensure the original broadcastable flags are preserved by Max/Argmax.
x = tensor.matrix().dimshuffle("x", 0, "x", 1, "x")
x = tt.matrix().dimshuffle("x", 0, "x", 1, "x")
y = x.max(axis=1)
assert y.type.broadcastable == (True, True, False, True)
......@@ -3758,7 +3751,7 @@ class TestMaxAndArgmax:
assert tuple(v) == np.max(data, (1, -1)).shape
def test_zero_shape(self):
x = tensor.matrix()
x = tt.matrix()
m, i = max_and_argmax(x, axis=1)
f = theano.function([x], [m, i])
xv = np.zeros((0, 4), dtype=floatX)
......@@ -4177,20 +4170,20 @@ class TestMinMax:
def test_basic_allclose():
# This was raised by a user in https://github.com/Theano/Theano/issues/2975
assert tensor.basic._allclose(-0.311023883434, -0.311022856884)
assert tt.basic._allclose(-0.311023883434, -0.311022856884)
class TestOuter:
def test_outer(self):
for m in range(4):
for n in range(4):
x = tensor.tensor(dtype="floatX", broadcastable=(False,) * m)
y = tensor.tensor(dtype="floatX", broadcastable=(False,) * n)
x = tt.tensor(dtype="floatX", broadcastable=(False,) * m)
y = tt.tensor(dtype="floatX", broadcastable=(False,) * n)
s1 = np.random.randint(1, 10, m)
s2 = np.random.randint(1, 10, n)
v1 = np.asarray(np.random.rand(*s1)).astype(floatX)
v2 = np.asarray(np.random.rand(*s2)).astype(floatX)
o = tensor.outer(x, y).eval({x: v1, y: v2})
o = tt.outer(x, y).eval({x: v1, y: v2})
assert_allclose(o, np.outer(v1, v2))
def test_grad(self):
......@@ -4212,7 +4205,7 @@ class TestOuter:
]:
data0 = np.random.rand(*shp0).astype(floatX)
data1 = np.random.rand(*shp1).astype(floatX)
utt.verify_grad(tensor.outer, [data0, data1])
utt.verify_grad(tt.outer, [data0, data1])
class TestGetVectorLength:
......@@ -4282,7 +4275,7 @@ class TestJoinAndSplit:
# tested only on cpu as gpu support only float32
a = as_tensor_variable(1)
b = as_tensor_variable(2.0)
c = tensor._shared(np.asarray(3.0, dtype=self.floatX))
c = tt._shared(np.asarray(3.0, dtype=self.floatX))
s = stack([a, b, c])
want = np.array([1, 2, 3])
out = self.eval_outputs_and_check_vector([s], opt.MakeVector())
......@@ -4302,8 +4295,8 @@ class TestJoinAndSplit:
# Test that calling stack() on scalars instantiates MakeVector,
# not Join. Test that the floatX dtype stay floatX, not downcasted
# to int64
a = tensor.scalar("a", dtype=self.floatX)
b = tensor.scalar("b", dtype=self.floatX)
a = tt.scalar("a", dtype=self.floatX)
b = tt.scalar("b", dtype=self.floatX)
s = stack([a, b, a, b])
f = function([a, b], s, mode=self.mode)
val = f(1, 2)
......@@ -4317,8 +4310,8 @@ class TestJoinAndSplit:
def test_stack_scalar_make_vector_dtype(self):
# Test that calling stack() on scalars instantiates MakeVector,
# event when the scalar don't have the same dtype.
a = tensor.iscalar("a")
b = tensor.lscalar("b")
a = tt.iscalar("a")
b = tt.lscalar("b")
s = stack([a, b, a, b])
f = function([a, b], s, mode=self.mode)
val = f(1, 2)
......@@ -4331,8 +4324,8 @@ class TestJoinAndSplit:
def test_stack_scalar_make_vector_constant(self):
# Test that calling stack() on scalars instantiates MakeVector,
# event when the scalar are simple int type.
a = tensor.iscalar("a")
b = tensor.lscalar("b")
a = tt.iscalar("a")
b = tt.lscalar("b")
# test when the constant is the first element.
# The first element is used in a special way
s = stack([10, a, b, np.int8(3)])
......@@ -4349,8 +4342,8 @@ class TestJoinAndSplit:
# Testing against old interface
warnings.simplefilter("always", DeprecationWarning)
a = tensor.imatrix("a")
b = tensor.imatrix("b")
a = tt.imatrix("a")
b = tt.imatrix("b")
s1 = stack(a, b)
s2 = stack([a, b])
f = function([a, b], [s1, s2], mode=self.mode)
......@@ -4403,8 +4396,8 @@ class TestJoinAndSplit:
def test_stack_hessian(self):
# Test the gradient of stack when used in hessian, see gh-1589
a = tensor.dvector("a")
b = tensor.dvector("b")
a = tt.dvector("a")
b = tt.dvector("b")
A = stack([a, b])
B = A.T.dot(A)
Ha, Hb = hessian(B.sum(), [a, b])
......@@ -4423,8 +4416,8 @@ class TestJoinAndSplit:
def test_stack_hessian2(self):
# Test the hessian macro when the gradient itself does not depend
# on the input (but the cost does)
a = tensor.dvector("a")
b = tensor.dvector("b")
a = tt.dvector("a")
b = tt.dvector("b")
A = stack([a, b])
Ha, Hb = hessian(A.sum(), [a, b])
......@@ -4442,8 +4435,8 @@ class TestJoinAndSplit:
def test_join_concatenate_one_element(self):
# Fast test of concatenate as this is an alias for join.
# also test that we remove the Join op if there is only 1 input
m = tensor.fmatrix()
c = tensor.concatenate([m])
m = tt.fmatrix()
c = tt.concatenate([m])
f = theano.function(
inputs=[m], outputs=[c], mode=self.mode.including("local_join_1")
)
......@@ -4741,14 +4734,12 @@ class TestJoinAndSplit:
assert not c.type.broadcastable[1]
# Opt can remplace the int by a Theano constant
c = self.join_op(theano.tensor.constant(1), a, b)
c = self.join_op(tt.constant(1), a, b)
assert c.type.broadcastable[0] and c.type.broadcastable[2]
assert not c.type.broadcastable[1]
# In case futur opt insert other useless stuff
c = self.join_op(
theano.tensor.cast(theano.tensor.constant(1), dtype="int32"), a, b
)
c = self.join_op(tt.cast(tt.constant(1), dtype="int32"), a, b)
assert c.type.broadcastable[0] and c.type.broadcastable[2]
assert not c.type.broadcastable[1]
......@@ -4958,10 +4949,10 @@ class TestJoinAndSplit:
def test_rebroadcast(self):
# Regression test for a crash that used to happen when rebroadcasting.
x = tensor.TensorType(self.floatX, [False, False, True])()
u = tensor.TensorType(self.floatX, [False, False, True])()
x = tt.TensorType(self.floatX, [False, False, True])()
u = tt.TensorType(self.floatX, [False, False, True])()
# This line used to crash.
tensor.concatenate([x, -u], axis=2)
tt.concatenate([x, -u], axis=2)
def test_concatenate_same(self):
# Test that we can concatenate the same tensor multiple time.
......@@ -4969,7 +4960,7 @@ class TestJoinAndSplit:
# In the past it was broken on the GPU.
rng = np.random.RandomState(seed=utt.fetch_seed())
T_shared = self.shared(rng.rand(3, 4).astype(self.floatX))
Tout = tensor.concatenate([T_shared, T_shared])
Tout = tt.concatenate([T_shared, T_shared])
f = function([], Tout, mode=self.mode)
out = f()
if theano.config.mode != "FAST_COMPILE":
......@@ -5027,9 +5018,9 @@ def test_join_inplace():
# join function but all except one of them are empty. In this case join
# should work inplace and the output should be the view of the non-empty
# element.
s = tensor.lscalar()
x = tensor.vector("x")
z = tensor.zeros((s,))
s = tt.lscalar()
x = tt.vector("x")
z = tt.zeros((s,))
join = Join(view=0)
c = join(0, x, z, z)
......@@ -5051,13 +5042,11 @@ def test_join_oneInput():
# on an array of tensors but the array has only one element.
# In this case, we would like to avoid the computational
# overhead of concatenation of one element.
x_0 = theano.tensor.fmatrix()
x_1 = theano.tensor.fmatrix()
x_2 = theano.tensor.fvector()
join_0 = theano.tensor.concatenate([x_0], axis=1)
join_1 = theano.tensor.concatenate(
[x_0, x_1, theano.tensor.shape_padright(x_2)], axis=1
)
x_0 = tt.fmatrix()
x_1 = tt.fmatrix()
x_2 = tt.fvector()
join_0 = tt.concatenate([x_0], axis=1)
join_1 = tt.concatenate([x_0, x_1, tt.shape_padright(x_2)], axis=1)
assert join_0 is x_0
assert join_1 is not x_0
......@@ -5093,9 +5082,9 @@ class TestComparison:
for x, y, err in [
(self.shared(l.astype(dtype)), self.shared(r.astype(dtype)), False),
(l, self.shared(r.astype(dtype)), True),
(tensor.constant(l), self.shared(r.astype(dtype)), False),
(tt.constant(l), self.shared(r.astype(dtype)), False),
(self.shared(l.astype(dtype)), r, False),
(self.shared(l.astype(dtype)), tensor.constant(r), False),
(self.shared(l.astype(dtype)), tt.constant(r), False),
]:
try:
fn = self.inplace_func([], x > y)
......@@ -5111,9 +5100,9 @@ class TestComparison:
for x, y, err in [
(self.shared(l.astype(dtype)), self.shared(r.astype(dtype)), False),
(l, self.shared(r.astype(dtype)), True),
(tensor.constant(l), self.shared(r.astype(dtype)), False),
(tt.constant(l), self.shared(r.astype(dtype)), False),
(self.shared(l.astype(dtype)), r, False),
(self.shared(l.astype(dtype)), tensor.constant(r), False),
(self.shared(l.astype(dtype)), tt.constant(r), False),
]:
try:
fn = self.inplace_func([], x < y)
......@@ -5129,9 +5118,9 @@ class TestComparison:
for x, y, err in [
(self.shared(l.astype(dtype)), self.shared(r.astype(dtype)), False),
(l, self.shared(r.astype(dtype)), True),
(tensor.constant(l), self.shared(r.astype(dtype)), False),
(tt.constant(l), self.shared(r.astype(dtype)), False),
(self.shared(l.astype(dtype)), r, False),
(self.shared(l.astype(dtype)), tensor.constant(r), False),
(self.shared(l.astype(dtype)), tt.constant(r), False),
]:
try:
fn = self.inplace_func([], x <= y)
......@@ -5147,9 +5136,9 @@ class TestComparison:
for x, y, err in [
(self.shared(l.astype(dtype)), self.shared(r.astype(dtype)), False),
(l, self.shared(r.astype(dtype)), True),
(tensor.constant(l), self.shared(r.astype(dtype)), False),
(tt.constant(l), self.shared(r.astype(dtype)), False),
(self.shared(l.astype(dtype)), r, False),
(self.shared(l.astype(dtype)), tensor.constant(r), False),
(self.shared(l.astype(dtype)), tt.constant(r), False),
]:
try:
fn = self.inplace_func([], x >= y)
......@@ -5165,9 +5154,9 @@ class TestComparison:
for x, y, err in [
(self.shared(l.astype(dtype)), self.shared(r.astype(dtype)), False),
(l, self.shared(r.astype(dtype)), True),
(tensor.constant(l), self.shared(r.astype(dtype)), False),
(tt.constant(l), self.shared(r.astype(dtype)), False),
(self.shared(l.astype(dtype)), r, False),
(self.shared(l.astype(dtype)), tensor.constant(r), False),
(self.shared(l.astype(dtype)), tt.constant(r), False),
]:
try:
fn = self.inplace_func([], eq(x, y))
......@@ -5183,9 +5172,9 @@ class TestComparison:
for x, y, err in [
(self.shared(l.astype(dtype)), self.shared(r.astype(dtype)), False),
(l, self.shared(r.astype(dtype)), True),
(tensor.constant(l), self.shared(r.astype(dtype)), False),
(tt.constant(l), self.shared(r.astype(dtype)), False),
(self.shared(l.astype(dtype)), r, False),
(self.shared(l.astype(dtype)), tensor.constant(r), False),
(self.shared(l.astype(dtype)), tt.constant(r), False),
]:
try:
fn = self.inplace_func([], neq(x, y))
......@@ -5367,7 +5356,7 @@ class TestAdd:
class TestCeil:
def test_complex(self):
with pytest.raises(TypeError):
tensor.ceil(tensor.zvector())
tt.ceil(tt.zvector())
class TestExp:
......@@ -5442,24 +5431,24 @@ class TestDivimpl:
class TestMean:
def test_mean_single_element(self):
res = tensor.mean(np.zeros(1))
res = tt.mean(np.zeros(1))
assert res.eval() == 0.0
def test_mean_f16(self):
x = tensor.vector(dtype="float16")
x = tt.vector(dtype="float16")
y = x.mean()
f = theano.function([x], y)
utt.assert_allclose(f(np.ones((100000,), dtype="float16")), 1.0)
def test_basic(self):
x = tensor.vector()
f = theano.function([x], tensor.mean(x))
x = tt.vector()
f = theano.function([x], tt.mean(x))
data = rand(50)
assert np.allclose(f(data), np.mean(data))
def test_list(self):
ll = [theano.shared(0.0), theano.shared(2.0)]
assert tensor.mean(ll).eval() == 1
assert tt.mean(ll).eval() == 1
class TestMatinv:
......@@ -5540,7 +5529,7 @@ class TestDot:
def test_Op_dims(self):
# _dot is a Dot op instance
_dot = theano.tensor.basic._dot
_dot = tt.basic._dot
d0 = scalar()
d1 = vector()
d2 = matrix()
......@@ -5881,7 +5870,7 @@ class TestGrad:
# grad: Test passing a single variable param
o = TestGrad.Obj1()
a1 = o.make_node()
assert o.gval0 is tensor.grad(a1.outputs[0], a1.inputs[0])
assert o.gval0 is tt.grad(a1.outputs[0], a1.inputs[0])
def test_Nparam(self):
# grad: Test passing multiple variable params
......@@ -5900,14 +5889,14 @@ class TestGrad:
# requires changing this test or making it fail you are almost certainly
# making a common mistake, NOT fixing something.
X = tensor.matrix()
X = tt.matrix()
y = X.sum()
G = tensor.grad(y, [X])
G = tt.grad(y, [X])
assert isinstance(G, list)
G = tensor.grad(y, X)
G = tt.grad(y, X)
assert not isinstance(G, list)
......@@ -5970,7 +5959,7 @@ class TestOpCache:
class TestReshape(utt.InferShapeTester, utt.OptimizationTestMixin):
def setup_method(self):
self.shared = tensor._shared
self.shared = tt._shared
self.op = Reshape
# The tag canonicalize is needed for the shape test in FAST_COMPILE
self.mode = None
......@@ -5979,7 +5968,7 @@ class TestReshape(utt.InferShapeTester, utt.OptimizationTestMixin):
opt.MakeVector,
opt.Shape_i,
DimShuffle,
theano.tensor.Elemwise,
tt.Elemwise,
)
super().setup_method()
......@@ -6148,7 +6137,7 @@ class TestReshape(utt.InferShapeTester, utt.OptimizationTestMixin):
assert f(np.ndarray((0,), dtype="float32")).shape == (0, 100)
def test_empty_shp(self):
const = theano.tensor.constant([1]).reshape(())
const = tt.constant([1]).reshape(())
f = function([], const)
assert f().shape == ()
......@@ -6156,7 +6145,7 @@ class TestReshape(utt.InferShapeTester, utt.OptimizationTestMixin):
def test_make_column_matrix_broadcastable():
# The goal of the operation made by `b` is to ensure the second dimension
# of the column matrix is broadcastable.
a = tensor.dmatrix()
a = tt.dmatrix()
b = a.reshape((a.shape[0],)).dimshuffle(0, "x")
f = function([a], b)
assert (f(np.zeros((3, 1))) + np.ones(2) == np.ones((3, 2))).all()
......@@ -6270,26 +6259,26 @@ def test_is_flat():
# given outdim
# Constant variable
assert tensor.is_flat(tensor.as_tensor_variable(np.zeros((10))))
assert tensor.is_flat(tensor.as_tensor_variable(np.zeros((10, 10, 10))), ndim=3)
assert not tensor.is_flat(tensor.as_tensor_variable(np.zeros((10, 10, 10))))
assert tt.is_flat(tt.as_tensor_variable(np.zeros((10))))
assert tt.is_flat(tt.as_tensor_variable(np.zeros((10, 10, 10))), ndim=3)
assert not tt.is_flat(tt.as_tensor_variable(np.zeros((10, 10, 10))))
# Symbolic variable
assert tensor.is_flat(tensor.vector())
assert tensor.is_flat(tensor.tensor3(), ndim=3)
assert not tensor.is_flat(tensor.tensor3())
assert tt.is_flat(tt.vector())
assert tt.is_flat(tt.tensor3(), ndim=3)
assert not tt.is_flat(tt.tensor3())
# Reshape with constant shape
X = tensor.tensor4()
assert tensor.is_flat(X.reshape((-1,)))
assert tensor.is_flat(X.reshape((10, 10, -1)), ndim=3)
assert not tensor.is_flat(X.reshape((10, 10, -1)))
X = tt.tensor4()
assert tt.is_flat(X.reshape((-1,)))
assert tt.is_flat(X.reshape((10, 10, -1)), ndim=3)
assert not tt.is_flat(X.reshape((10, 10, -1)))
# Reshape with symbolic shape
X = tensor.tensor4()
assert tensor.is_flat(X.reshape((tensor.iscalar(),)))
assert tensor.is_flat(X.reshape((tensor.iscalar(),) * 3), ndim=3)
assert not tensor.is_flat(X.reshape((tensor.iscalar(),) * 3))
X = tt.tensor4()
assert tt.is_flat(X.reshape((tt.iscalar(),)))
assert tt.is_flat(X.reshape((tt.iscalar(),) * 3), ndim=3)
assert not tt.is_flat(X.reshape((tt.iscalar(),) * 3))
def test_tile():
......@@ -6330,7 +6319,7 @@ def test_tile():
== np.tile(x_, (2, 3, 4, 6))
)
# Test when reps is integer, tensor.scalar or tensor.vector.
# Test when reps is integer, tt.scalar or tt.vector.
# Test 1,2,3,4-dimensional cases.
# Test input x has the shape [2], [2, 4], [2, 4, 3], [2, 4, 3, 5].
test_shape = [2, 4, 3, 5]
......@@ -6345,13 +6334,13 @@ def test_tile():
f = function([x], tile(x, reps_))
assert np.all(f(x_) == np.tile(x_, reps_))
# tensor.scalar:
# tt.scalar:
reps = iscalar()
reps_ = 2
f = function([x, reps], tile(x, reps))
assert np.all(f(x_, reps_) == np.tile(x_, reps_))
# tensor.vector:
# tt.vector:
reps = ivector()
reps_ = [2] if k == 1 or k == 2 else [2, 3]
ndim_ = k
......@@ -6363,7 +6352,7 @@ def test_tile():
f = function([x], tile(x, reps_))
assert np.all(f(x_) == np.tile(x_, reps_))
# list of integers and tensor.scalars:
# list of integers and tt.scalars:
d = iscalar()
reps = [2, d, 4]
f = function([x, d], tile(x, reps))
......@@ -6409,7 +6398,7 @@ def test_tile():
tile(x, reps)
# error raising test: ndim is not None, ndim < x.ndim
# 3 cases below (reps is list/tensor.scalar/tensor.vector):
# 3 cases below (reps is list/tt.scalar/tt.vector):
for reps in [[2, 3, 4], iscalar(), ivector()]:
if k > 1:
ndim = k - 1
......@@ -6424,7 +6413,7 @@ def test_tile():
tile(x, reps, ndim)
# error raising test:
# reps is tensor.vector and len(reps_value) > ndim,
# reps is tt.vector and len(reps_value) > ndim,
# reps_value is the real value when excuting the function.
reps = ivector()
r = [2, 3, 4, 5, 6, 7]
......@@ -7286,7 +7275,7 @@ def test_var():
assert np.allclose(v, f(a_val))
# Test that we don't upcast float16 computation
assert theano.tensor.vector(dtype="float16").var().dtype == "float16"
assert tt.vector(dtype="float16").var().dtype == "float16"
class TestSum:
......@@ -7298,7 +7287,7 @@ class TestSum:
def test_list(self):
ll = [theano.shared(0.0), theano.shared(2.0)]
tensor.sum(ll).eval() == 2
tt.sum(ll).eval() == 2
@pytest.mark.skipif(
......@@ -7420,7 +7409,7 @@ def _test_autocast_numpy():
# Go through some typical scalar values.
def ok(z):
assert tensor.constant(z).dtype == np.asarray(z).dtype
assert tt.constant(z).dtype == np.asarray(z).dtype
for x in (
[2 ** i for i in range(63)] + [0, 0, 1, 2 ** 63 - 1] + [0.0, 1.0, 1.1, 1.5]
......@@ -7442,9 +7431,9 @@ def _test_autocast_numpy_floatX():
def ok(z, floatX):
if isinstance(z, float) and floatX == "float32" and not hasattr(z, "dtype"):
# Special case where we use 'float32' instead of 'float64'.
assert tensor.constant(z).dtype == "float32"
assert tt.constant(z).dtype == "float32"
else:
assert tensor.constant(z).dtype == np.asarray(z).dtype
assert tt.constant(z).dtype == np.asarray(z).dtype
try:
# Test with various values of `config.floatX`.
......@@ -7484,13 +7473,13 @@ class TestArithmeticCast:
# array == 1d array
# i_scalar == scalar type used internally by Theano
def theano_scalar(dtype):
return tensor.scalar(dtype=str(dtype))
return tt.scalar(dtype=str(dtype))
def numpy_scalar(dtype):
return np.array(1, dtype=dtype)
def theano_array(dtype):
return tensor.vector(dtype=str(dtype))
return tt.vector(dtype=str(dtype))
def numpy_array(dtype):
return np.array([1], dtype=dtype)
......@@ -7524,8 +7513,8 @@ class TestArithmeticCast:
# special way (depending on `config.int_division`).
is_int_division = (
op is operator_div
and a_type in tensor.discrete_dtypes
and b_type in tensor.discrete_dtypes
and a_type in tt.discrete_dtypes
and b_type in tt.discrete_dtypes
)
# We will test all meaningful combinations of
# scalar and array operations.
......@@ -7655,7 +7644,7 @@ class TestLongTensor:
val = 2 ** exponent - 1
scalar_ct = constant(val)
assert scalar_ct.dtype in tensor.int_dtypes, (
assert scalar_ct.dtype in tt.int_dtypes, (
exponent,
val,
scalar_ct.dtype,
......@@ -7745,7 +7734,7 @@ class TestBroadcast:
def test_patternbroadcast(self):
# Test that patternbroadcast with an empty broadcasting pattern works
x = scalar("x")
m = tensor.matrix("m")
m = tt.matrix("m")
s = patternbroadcast(m, x.broadcastable)
assert s is m
x2 = patternbroadcast(x, x.broadcastable)
......@@ -7786,7 +7775,7 @@ class TestBroadcast:
def test_len():
for shape_ in [(5,), (3, 4), (7, 4, 6)]:
x = tensor.tensor(dtype="floatX", broadcastable=(False,) * len(shape_))
x = tt.tensor(dtype="floatX", broadcastable=(False,) * len(shape_))
with pytest.raises(TypeError):
len(x)
......@@ -7853,9 +7842,9 @@ def test_mod_compile():
#
# The c_code generated is not compiling as of 30 June 2010. I fix the
# compilation in the same commit.
x = tensor.vector()
y = tensor.vector()
out = tensor.switch(tensor.eq(3 % x.shape[0], 0), y, y[:-1])
x = tt.vector()
y = tt.vector()
out = tt.switch(tt.eq(3 % x.shape[0], 0), y, y[:-1])
theano.function([x, y], out)
......@@ -7874,7 +7863,7 @@ def test_unalign():
b[:] = rand(len(b))
out_numpy = 2 * a + 3 * b
av, bv = tensor.vectors("ab")
av, bv = tt.vectors("ab")
f = theano.function([av, bv], 2 * av + 3 * bv)
f.maker.fgraph.toposort()
......@@ -7893,7 +7882,7 @@ def test_unalign():
assert not b.flags.aligned
out_numpy = 2 * a + 3 * b
av, bv = tensor.scalars("ab")
av, bv = tt.scalars("ab")
f = theano.function([av, bv], 2 * av + 3 * bv)
f.maker.fgraph.toposort()
try:
......@@ -7907,12 +7896,12 @@ def test_unalign():
def test_dimshuffle_duplicate():
x = tensor.vector()
x = tt.vector()
success = False
try:
tensor.DimShuffle((False,), (0, 0))(x)
tt.DimShuffle((False,), (0, 0))(x)
except ValueError as e:
assert str(e).find("may not appear twice") != -1
success = True
......@@ -7922,32 +7911,32 @@ def test_dimshuffle_duplicate():
class TestGetScalarConstantValue:
def test_get_scalar_constant_value(self):
a = tensor.stack([1, 2, 3])
a = tt.stack([1, 2, 3])
assert get_scalar_constant_value(a[0]) == 1
assert get_scalar_constant_value(a[1]) == 2
assert get_scalar_constant_value(a[2]) == 3
b = tensor.iscalar()
a = tensor.stack([b, 2, 3])
with pytest.raises(tensor.basic.NotScalarConstantError):
b = tt.iscalar()
a = tt.stack([b, 2, 3])
with pytest.raises(tt.basic.NotScalarConstantError):
get_scalar_constant_value(a[0])
assert get_scalar_constant_value(a[1]) == 2
assert get_scalar_constant_value(a[2]) == 3
# For now get_scalar_constant_value goes through only MakeVector and Join of
# scalars.
v = tensor.ivector()
a = tensor.stack([v, [2], [3]])
with pytest.raises(tensor.NotScalarConstantError):
v = tt.ivector()
a = tt.stack([v, [2], [3]])
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(a[0])
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(a[1])
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(a[2])
# Test the case SubTensor(Shape(v)) when the dimensions
# is broadcastable.
v = tensor.row()
v = tt.row()
assert get_scalar_constant_value(v.shape[0]) == 1
def test_subtensor_of_constant(self):
......@@ -7962,14 +7951,14 @@ class TestGetScalarConstantValue:
def test_numpy_array(self):
# Regression test for crash when called on a numpy array.
assert get_scalar_constant_value(np.array(3)) == 3
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(np.array([0, 1]))
with pytest.raises(tensor.EmptyConstantError):
with pytest.raises(tt.EmptyConstantError):
get_scalar_constant_value(np.array([]))
def test_make_vector(self):
mv = opt.make_vector(1, 2, 3)
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(mv)
assert get_scalar_constant_value(mv[0]) == 1
assert get_scalar_constant_value(mv[1]) == 2
......@@ -7978,23 +7967,23 @@ class TestGetScalarConstantValue:
assert get_scalar_constant_value(mv[np.int64(1)]) == 2
assert get_scalar_constant_value(mv[np.uint(2)]) == 3
t = theano.scalar.Scalar("int64")
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(mv[t()])
def test_shape_i(self):
c = theano.tensor.constant(np.random.rand(3, 4))
c = tt.constant(np.random.rand(3, 4))
s = opt.Shape_i(0)(c)
assert get_scalar_constant_value(s) == 3
s = opt.Shape_i(1)(c)
assert get_scalar_constant_value(s) == 4
d = theano.shared(np.random.randn(1, 1), broadcastable=(True, True))
f = theano.tensor.basic.ScalarFromTensor()(opt.Shape_i(0)(d))
f = tt.ScalarFromTensor()(opt.Shape_i(0)(d))
assert get_scalar_constant_value(f) == 1
def test_elemwise(self):
# We test only for a few elemwise, the list of all supported
# elemwise are in the fct.
c = theano.tensor.constant(np.random.rand())
c = tt.constant(np.random.rand())
s = c + 1
assert np.allclose(get_scalar_constant_value(s), c.data + 1)
s = c - 1
......@@ -8003,14 +7992,14 @@ class TestGetScalarConstantValue:
assert np.allclose(get_scalar_constant_value(s), c.data * 1.2)
s = c < 0.5
assert np.allclose(get_scalar_constant_value(s), int(c.data < 0.5))
s = tensor.second(c, 0.4)
s = tt.second(c, 0.4)
assert np.allclose(get_scalar_constant_value(s), 0.4)
def test_assert(self):
# Make sure we still get the constant value if it is wrapped in
# an Assert.
c = theano.tensor.constant(2)
x = theano.tensor.scalar()
c = tt.constant(2)
x = tt.scalar()
# condition is always True
a = opt.Assert()(c, c > 1)
......@@ -8019,25 +8008,25 @@ class TestGetScalarConstantValue:
with change_flags(compute_test_value="off"):
# condition is always False
a = opt.Assert()(c, c > 2)
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(a)
# condition is not constant
a = opt.Assert()(c, c > x)
with pytest.raises(tensor.NotScalarConstantError):
with pytest.raises(tt.NotScalarConstantError):
get_scalar_constant_value(a)
def test_second(self):
# Second should apply when the value is constant but not the shape
c = theano.tensor.constant(np.random.rand())
shp = theano.tensor.vector()
s = theano.tensor.second(shp, c)
c = tt.constant(np.random.rand())
shp = tt.vector()
s = tt.second(shp, c)
assert get_scalar_constant_value(s) == c.data
def test_copy(self):
# Make sure we do not return the internal storage of a constant,
# so we cannot change the value of a constant by mistake.
c = theano.tensor.constant(3)
c = tt.constant(3)
d = extract_constant(c)
d += 1
e = extract_constant(c)
......@@ -8058,17 +8047,17 @@ class TestComplexMod:
class TestSize:
# Ensure the `size` attribute of tensors behaves as in numpy.
def test_matrix(self):
x = tensor.matrix()
x = tt.matrix()
y = np.zeros((5, 7), dtype=config.floatX)
assert y.size == function([x], x.size)(y)
def test_vector(self):
x = tensor.vector()
x = tt.vector()
y = np.zeros(7, dtype=config.floatX)
assert y.size == function([x], x.size)(y)
def test_scalar(self):
x = tensor.scalar()
x = tt.scalar()
y = np.array(7, dtype=config.floatX)
assert y.size == function([x], x.size)(y)
......@@ -8080,7 +8069,7 @@ class TestSize:
class TestDiag:
# Test that tensor.diag has the same behavior as np.diag.
# Test that tt.diag has the same behavior as np.diag.
# np.diag has two behaviors:
#
# (1) when given a vector, it returns a matrix with that vector as the
......@@ -8095,15 +8084,15 @@ class TestDiag:
# the right op based on the dimension of the input.
def setup_method(self):
self.mode = None
self.shared = tensor._shared
self.shared = tt._shared
self.floatX = config.floatX
self.type = tensor.TensorType
self.type = tt.TensorType
def test_diag(self):
rng = np.random.RandomState(utt.fetch_seed())
# test vector input
x = theano.tensor.vector()
x = tt.vector()
g = diag(x)
assert isinstance(g.owner.op, AllocDiag)
f = theano.function([x], g)
......@@ -8128,14 +8117,14 @@ class TestDiag:
assert (r == v).all()
# Test scalar input
xx = theano.tensor.scalar()
xx = tt.scalar()
with pytest.raises(ValueError):
diag(xx)
def test_infer_shape(self):
rng = np.random.RandomState(utt.fetch_seed())
x = theano.tensor.vector()
x = tt.vector()
g = diag(x)
f = theano.function([x], g.shape)
topo = f.maker.fgraph.toposort()
......@@ -8145,7 +8134,7 @@ class TestDiag:
m = rng.rand(shp).astype(self.floatX)
assert (f(m) == np.diag(m).shape).all()
x = theano.tensor.matrix()
x = tt.matrix()
g = diag(x)
f = theano.function([x], g.shape)
topo = f.maker.fgraph.toposort()
......@@ -8158,9 +8147,9 @@ class TestDiag:
def test_diag_grad(self):
rng = np.random.RandomState(utt.fetch_seed())
x = rng.rand(5)
tensor.verify_grad(diag, [x], rng=rng)
tt.verify_grad(diag, [x], rng=rng)
x = rng.rand(5, 3)
tensor.verify_grad(diag, [x], rng=rng)
tt.verify_grad(diag, [x], rng=rng)
class TestAllocDiag:
......@@ -8223,9 +8212,9 @@ class TestAllocDiag:
assert np.all(rediag_shape == test_val.shape)
diag_x = adiag_op(x)
sum_diag_x = tensor.sum(diag_x)
grad_x = tensor.grad(sum_diag_x, x)
grad_diag_x = tensor.grad(sum_diag_x, diag_x)
sum_diag_x = tt.sum(diag_x)
grad_x = tt.grad(sum_diag_x, x)
grad_diag_x = tt.grad(sum_diag_x, diag_x)
f_grad_x = theano.function([x], grad_x, mode=self.mode)
f_grad_diag_x = theano.function([x], grad_diag_x, mode=self.mode)
grad_input = f_grad_x(test_val)
......@@ -8262,9 +8251,9 @@ class TestNumpyAssumptions:
def test_transpose():
x1 = tensor.dvector("x1")
x2 = tensor.dmatrix("x2")
x3 = tensor.dtensor3("x3")
x1 = tt.dvector("x1")
x2 = tt.dmatrix("x2")
x3 = tt.dtensor3("x3")
x1v = np.arange(24)
x2v = np.arange(24).reshape(2, 12)
......@@ -8273,16 +8262,16 @@ def test_transpose():
f = theano.function(
[x1, x2, x3],
[
tensor.transpose(x1),
tensor.transpose(x2),
tensor.transpose(x3),
tt.transpose(x1),
tt.transpose(x2),
tt.transpose(x3),
x1.transpose(),
x2.transpose(),
x3.transpose(),
x2.transpose(0, 1),
x3.transpose((0, 2, 1)),
tensor.transpose(x2, [0, 1]),
tensor.transpose(x3, [0, 2, 1]),
tt.transpose(x2, [0, 1]),
tt.transpose(x3, [0, 2, 1]),
],
)
......@@ -8306,10 +8295,10 @@ def test_transpose():
assert np.all(t3d == np.transpose(x3v, [0, 2, 1]))
# Check that we create a name.
assert tensor.transpose(x1).name == "x1.T"
assert tensor.transpose(x2).name == "x2.T"
assert tensor.transpose(x3).name == "x3.T"
assert tensor.transpose(tensor.dmatrix()).name is None
assert tt.transpose(x1).name == "x1.T"
assert tt.transpose(x2).name == "x2.T"
assert tt.transpose(x3).name == "x3.T"
assert tt.transpose(tt.dmatrix()).name is None
def test_stacklists():
......@@ -8536,7 +8525,7 @@ class TestInferShape(utt.InferShapeTester):
[advec, bdvec],
[Dot()(advec, bdvec)],
[advec_val, bdvec_val],
(Dot, tensor.blas.Dot22, tensor.blas.Gemv, tensor.blas_c.CGemv),
(Dot, tt.blas.Dot22, tt.blas.Gemv, tt.blas_c.CGemv),
)
# mat/mat
......@@ -8548,7 +8537,7 @@ class TestInferShape(utt.InferShapeTester):
[admat, bdmat],
[Dot()(admat, bdmat)],
[admat_val, bdmat_val],
(Dot, tensor.blas.Dot22),
(Dot, tt.blas.Dot22),
)
# vec/mat
......@@ -8557,7 +8546,7 @@ class TestInferShape(utt.InferShapeTester):
[advec, bdmat],
[Dot()(advec, bdmat)],
[advec_val, bdmat_val],
(Dot, tensor.blas.Dot22, tensor.blas.Gemv, tensor.blas_c.CGemv),
(Dot, tt.blas.Dot22, tt.blas.Gemv, tt.blas_c.CGemv),
)
# mat/vec
......@@ -8566,7 +8555,7 @@ class TestInferShape(utt.InferShapeTester):
[admat, bdvec],
[Dot()(admat, bdvec)],
[admat_val, bdvec_val],
(Dot, tensor.blas.Dot22, tensor.blas.Gemv, tensor.blas_c.CGemv),
(Dot, tt.blas.Dot22, tt.blas.Gemv, tt.blas_c.CGemv),
)
# Split
......@@ -9034,7 +9023,7 @@ class TestTensorInstanceMethods:
def test_norm():
x = theano.tensor.vector("x")
x = tt.vector("x")
n = x.norm(2)
f = theano.function([x], n)
assert np.allclose(f([1, 1]), np.sqrt(2))
......@@ -9042,8 +9031,8 @@ def test_norm():
class TestCov:
def test_core(self):
x = theano.tensor.matrix("x")
c = theano.tensor.cov(x)
x = tt.matrix("x")
c = tt.cov(x)
f = theano.function([x], c)
# basic cov function
......@@ -9064,8 +9053,8 @@ class TestCov:
def test_rowvar(self):
for rowvar in [True, False]:
x = theano.tensor.matrix("x")
c = theano.tensor.cov(x, rowvar=rowvar)
x = tt.matrix("x")
c = tt.cov(x, rowvar=rowvar)
f = theano.function([x], c)
data = np.asarray(np.random.rand(3, 5), dtype=config.floatX)
......@@ -9081,17 +9070,17 @@ class TestCov:
assert np.allclose(f(data), np.cov(data, rowvar=rowvar))
# check when variables are along the first axis
x = theano.tensor.matrix("x")
c = theano.tensor.cov(x, rowvar=False)
x = tt.matrix("x")
c = tt.cov(x, rowvar=False)
f = theano.function([x], c)
data = np.asarray(np.random.rand(2, 1), dtype=config.floatX)
assert np.allclose(f(data), np.cov(data, rowvar=False))
def test_y(self):
# test y
x = theano.tensor.matrix("x")
y = theano.tensor.matrix("y")
c = theano.tensor.cov(x, y=y)
x = tt.matrix("x")
y = tt.matrix("y")
c = tt.cov(x, y=y)
f = theano.function([x, y], c)
data = np.asarray(np.random.rand(3, 5), dtype=config.floatX)
......@@ -9113,8 +9102,8 @@ class TestCov:
def test_ddof(self):
for ddof in range(0, 5):
x = theano.tensor.matrix("x")
c = theano.tensor.cov(x, ddof=ddof)
x = tt.matrix("x")
c = tt.cov(x, ddof=ddof)
f = theano.function([x], c)
data = np.asarray(np.random.rand(3, 5), dtype=config.floatX)
......@@ -9123,8 +9112,8 @@ class TestCov:
def test_bias(self):
for bias in [True, False]:
x = theano.tensor.matrix("x")
c = theano.tensor.cov(x, bias=bias)
x = tt.matrix("x")
c = tt.cov(x, bias=bias)
f = theano.function([x], c)
data = np.asarray(np.random.rand(3, 5), dtype=config.floatX)
......@@ -9132,8 +9121,8 @@ class TestCov:
for ddof in range(0, 5):
for bias in [True, False]:
x = theano.tensor.matrix("x")
c = theano.tensor.cov(x, ddof=ddof, bias=bias)
x = tt.matrix("x")
c = tt.cov(x, ddof=ddof, bias=bias)
f = theano.function([x], c)
data = np.asarray(np.random.rand(3, 5), dtype=config.floatX)
......@@ -9243,12 +9232,12 @@ class TestSwapaxes:
assert np.array_equal(testMatrix, f(f(testMatrix)))
def test_interface(self):
x = theano.tensor.matrix()
x = tt.matrix()
x.swapaxes(0, 1)
def test_numpy_compare(self):
rng = np.random.RandomState(utt.fetch_seed())
A = tensor.matrix("A", dtype=theano.config.floatX)
A = tt.matrix("A", dtype=theano.config.floatX)
Q = swapaxes(A, 0, 1)
fn = function([A], [Q])
a = rng.rand(4, 4).astype(theano.config.floatX)
......@@ -9261,7 +9250,7 @@ class TestSwapaxes:
class TestPower:
def test_numpy_compare(self):
rng = np.random.RandomState(utt.fetch_seed())
A = tensor.matrix("A", dtype=theano.config.floatX)
A = tt.matrix("A", dtype=theano.config.floatX)
Q = power(A, 3)
fn = function([A], [Q])
a = rng.rand(4, 4).astype(theano.config.floatX)
......@@ -9271,14 +9260,14 @@ class TestPower:
assert np.allclose(n_p, t_p)
def test_multiple_power(self):
x = tensor.vector()
x = tt.vector()
y = [1, 2, 3]
z = power(x, y)
f = function([x], z)
assert np.allclose(f([1, 2, 3]), [1, 4, 27])
def test_wrong_shape(self):
x = tensor.vector()
x = tt.vector()
y = [1, 2, 3]
z = power(x, y)
f = function([x], z)
......@@ -9293,8 +9282,8 @@ class TestChoose(utt.InferShapeTester):
def test_numpy_compare(self):
a = tensor.vector(dtype="int32")
b = tensor.matrix(dtype="float32")
a = tt.vector(dtype="int32")
b = tt.matrix(dtype="float32")
A = np.random.randint(0, 4, 4).astype("int32")
B = np.asarray(np.random.rand(4, 4), dtype="float32")
......@@ -9306,8 +9295,8 @@ class TestChoose(utt.InferShapeTester):
assert np.allclose(t_c, n_c)
def test_method(self):
a = tensor.vector(dtype="int32")
b = tensor.matrix(dtype="float32")
a = tt.vector(dtype="int32")
b = tt.matrix(dtype="float32")
A = np.random.randint(0, 4, 4).astype("int32")
B = np.asarray(np.random.rand(4, 4), dtype="float32")
......@@ -9319,8 +9308,8 @@ class TestChoose(utt.InferShapeTester):
assert np.allclose(t_c, n_c)
def test_broadcasted(self):
a = tensor.scalar(dtype="int32")
b = tensor.matrix(dtype="float32")
a = tt.scalar(dtype="int32")
b = tt.matrix(dtype="float32")
# Test when a is broadcastable
A = 3
......@@ -9333,7 +9322,7 @@ class TestChoose(utt.InferShapeTester):
assert np.allclose(t_c, n_c)
# Test when the result should be broadcastable
b = theano.tensor.col(dtype="float32")
b = tt.col(dtype="float32")
B = np.asarray(np.random.rand(4, 1), dtype="float32")
for m in self.modes:
f = function([a, b], choose(a, b, mode=m))
......@@ -9343,17 +9332,17 @@ class TestChoose(utt.InferShapeTester):
assert np.allclose(t_c, n_c)
def test_dtype_error(self):
a = tensor.scalar(dtype="float32")
b = tensor.matrix(dtype="float32")
a = tt.scalar(dtype="float32")
b = tt.matrix(dtype="float32")
with pytest.raises(TypeError):
choose(a, b)
def test_numpy_compare_tuple(self):
a = tensor.tensor3(dtype="int32")
b = tensor.tensor3(dtype="float32")
c = tensor.tensor3(dtype="float32")
a = tt.tensor3(dtype="int32")
b = tt.tensor3(dtype="float32")
c = tt.tensor3(dtype="float32")
A = np.random.randint(0, 2, (2, 1, 1)).astype("int32")
B = np.asarray(np.random.rand(1, 6, 1), dtype="float32")
......@@ -9385,8 +9374,8 @@ class TestChoose(utt.InferShapeTester):
((4,), (1,)),
((1,), (1,)),
]:
a = tensor.tensor(dtype="int32", broadcastable=[n == 1 for n in shp1])
c = tensor.tensor(dtype="float32", broadcastable=[n == 1 for n in shp2])
a = tt.tensor(dtype="int32", broadcastable=[n == 1 for n in shp1])
c = tt.tensor(dtype="float32", broadcastable=[n == 1 for n in shp2])
A = np.asarray(np.random.rand(*shp1) * shp2[0], dtype="int32")
C = np.asarray(np.random.rand(*shp2) * shp2[0], dtype="float32")
self._compile_and_check(
......@@ -9402,9 +9391,9 @@ class TestChoose(utt.InferShapeTester):
@pytest.mark.skip(reason="Not implemented")
def test_infer_shape_tuple(self):
a = tensor.tensor3(dtype="int32")
b = tensor.tensor3(dtype="int32")
c = tensor.tensor3(dtype="int32")
a = tt.tensor3(dtype="int32")
b = tt.tensor3(dtype="int32")
c = tt.tensor3(dtype="int32")
A = np.asarray([1, 0], dtype="int32").reshape((2, 1, 1))
B = np.asarray(np.random.rand(1, 4, 1), dtype="int32")
......@@ -9436,7 +9425,7 @@ def test_allocempty():
def test_symbolic_slice():
x = theano.tensor.tensor4("x")
x = tt.tensor4("x")
a, b = x.shape[:2]
output = a.eval({x: np.zeros((5, 4, 3, 2), dtype=theano.config.floatX)})
assert output == np.array(5)
tests/tensor/test_blas.py
浏览文件 @ ef279e19
......@@ -3,7 +3,7 @@ import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
import theano.tensor.blas_scipy
from copy import copy
......@@ -25,7 +25,7 @@ from numpy import (
)
from numpy.testing import assert_array_almost_equal
from theano import tensor, In, shared, config
from theano import In, shared, config
from theano.tensor.blas import (
_dot22,
_dot22scalar,
......@@ -42,7 +42,17 @@ from theano.tensor.blas import (
Ger,
ger,
ger_destructive,
Dot22,
Dot22Scalar,
gemm,
local_dot22_to_dot22scalar,
gemv_no_inplace,
gemv,
gemv_inplace,
local_gemm_to_ger,
)
from theano.tensor.nnet import sigmoid
from theano.tensor.opt import in2out
from tests import unittest_tools
from tests.tensor.test_basic import as_tensor_variable, inplace_func, compile, inplace
......@@ -60,7 +70,7 @@ mode_blas_opt = mode_blas_opt.excluding("c_blas")
def test_dot_eq():
assert T.Dot() == T.Dot()
assert tt.Dot() == tt.Dot()
def sharedX(x, name):
......@@ -174,18 +184,18 @@ class TestGemm:
self.cmp(self.rand(0, 0), -1.0, self.rand(0, 0), self.rand(0, 0), -1.0)
def test_factorised_scalar(self):
a = T.matrix()
b = T.matrix()
a = tt.matrix()
b = tt.matrix()
s = theano.shared(np.zeros((5, 5)).astype(config.floatX))
lr1 = T.constant(0.01).astype(config.floatX)
lr2 = T.constant(2).astype(config.floatX)
l2_reg = T.constant(0.0001).astype(config.floatX)
lr1 = tt.constant(0.01).astype(config.floatX)
lr2 = tt.constant(2).astype(config.floatX)
l2_reg = tt.constant(0.0001).astype(config.floatX)
# test constant merge with gemm
f = theano.function(
[a, b],
updates=[(s, lr1 * T.dot(a, b) + l2_reg * lr2 * s)],
updates=[(s, lr1 * tt.dot(a, b) + l2_reg * lr2 * s)],
mode=mode_not_fast_compile,
).maker.fgraph.toposort()
# [Gemm{inplace}(<TensorType(float64, matrix)>, 0.01,
......@@ -197,7 +207,7 @@ class TestGemm:
# test factored scalar with merge
f = theano.function(
[a, b],
updates=[(s, lr1 * (T.dot(a, b) - l2_reg * s))],
updates=[(s, lr1 * (tt.dot(a, b) - l2_reg * s))],
mode=mode_not_fast_compile,
).maker.fgraph.toposort()
# [Gemm{inplace}(<TensorType(float64, matrix)>, 0.01,
......@@ -209,7 +219,7 @@ class TestGemm:
# test factored scalar with merge and neg
f = theano.function(
[a, b],
updates=[(s, s - lr1 * (s * 0.0002 + T.dot(a, b)))],
updates=[(s, s - lr1 * (s * 0.0002 + tt.dot(a, b)))],
mode=mode_not_fast_compile,
).maker.fgraph.toposort()
# [Gemm{inplace}(<TensorType(float64, matrix)>, -0.01,
......@@ -249,7 +259,7 @@ class TestGemm:
# test that dot args can be aliased
Z = shared(self.rand(2, 2), name="Z")
A = shared(self.rand(2, 2), name="A")
one = T.constant(1.0).astype(Z.dtype)
one = tt.constant(1.0).astype(Z.dtype)
f = inplace_func([], gemm_inplace(Z, one, A, A, one))
f()
f = inplace_func([], gemm_inplace(Z, one, A, A.T, one))
......@@ -349,7 +359,7 @@ class TestGemm:
g_i = theano.function(
[],
tz_i,
updates=[(tz, T.set_subtensor(tz[:, :, i], tz_i))],
updates=[(tz, tt.set_subtensor(tz[:, :, i], tz_i))],
mode=compile.Mode(optimizer=None, linker=l),
)
for j in range(3):
......@@ -400,11 +410,11 @@ class TestGemmNoFlags(object):
slice_B=False,
slice_C=False,
):
alpha = theano.tensor.scalar(dtype=dtype, name="alpha")
beta = theano.tensor.scalar(dtype=dtype, name="beta")
A = theano.tensor.matrix(dtype=dtype, name="A")
B = theano.tensor.matrix(dtype=dtype, name="B")
C = theano.tensor.matrix(dtype=dtype, name="C")
alpha = tt.scalar(dtype=dtype, name="alpha")
beta = tt.scalar(dtype=dtype, name="beta")
A = tt.matrix(dtype=dtype, name="A")
B = tt.matrix(dtype=dtype, name="B")
C = tt.matrix(dtype=dtype, name="C")
A1 = self.get_variable(A, transpose_A, slice_A)
B1 = self.get_variable(B, transpose_B, slice_B)
......@@ -522,9 +532,9 @@ class TestGemmNoFlags(object):
def test_res_is_a():
X, Y, Z, a, b = XYZab()
assert not res_is_a(a, T.sqrt)
assert not res_is_a(a + a, T.sqrt)
assert res_is_a(T.sqrt(a + a), T.sqrt)
assert not res_is_a(a, tt.sqrt)
assert not res_is_a(a + a, tt.sqrt)
assert res_is_a(tt.sqrt(a + a), tt.sqrt)
# leave the maxclients stuff untested because it requires being in an fgraph.
......@@ -532,13 +542,13 @@ def test_res_is_a():
class TestAsScalar:
def test_basic(self):
# Test that it works on scalar constants
a = T.constant(2.5)
b = T.constant(np.asarray([[[0.5]]]))
a = tt.constant(2.5)
b = tt.constant(np.asarray([[[0.5]]]))
b2 = b.dimshuffle()
assert b2.ndim == 0
d_a = T.DimShuffle([], [])(a)
d_b = T.DimShuffle([True, True, True], [0, 2, 1])(b)
d_a2 = T.DimShuffle([], ["x", "x", "x"])(a)
d_a = tt.DimShuffle([], [])(a)
d_b = tt.DimShuffle([True, True, True], [0, 2, 1])(b)
d_a2 = tt.DimShuffle([], ["x", "x", "x"])(a)
assert _as_scalar(a) == a
assert _as_scalar(b) != b
......@@ -548,15 +558,15 @@ class TestAsScalar:
def test_basic_1(self):
# Test that it fails on nonscalar constants
a = T.constant(np.ones(5))
a = tt.constant(np.ones(5))
assert _as_scalar(a) is None
assert _as_scalar(T.DimShuffle([False], [0, "x"])(a)) is None
assert _as_scalar(tt.DimShuffle([False], [0, "x"])(a)) is None
def test_basic_2(self):
# Test that it works on scalar variables
a = T.dscalar()
d_a = T.DimShuffle([], [])(a)
d_a2 = T.DimShuffle([], ["x", "x"])(a)
a = tt.dscalar()
d_a = tt.DimShuffle([], [])(a)
d_a2 = tt.DimShuffle([], ["x", "x"])(a)
assert _as_scalar(a) is a
assert _as_scalar(d_a) is a
......@@ -564,15 +574,15 @@ class TestAsScalar:
def test_basic_3(self):
# Test that it fails on nonscalar variables
a = T.matrix()
a = tt.matrix()
assert _as_scalar(a) is None
assert _as_scalar(T.DimShuffle([False, False], [0, "x", 1])(a)) is None
assert _as_scalar(tt.DimShuffle([False, False], [0, "x", 1])(a)) is None
class TestRealMatrix:
def test_basic(self):
assert _is_real_matrix(T.DimShuffle([False, False], [1, 0])(T.matrix()))
assert not _is_real_matrix(T.DimShuffle([False], ["x", 0])(T.dvector()))
assert _is_real_matrix(tt.DimShuffle([False, False], [1, 0])(tt.matrix()))
assert not _is_real_matrix(tt.DimShuffle([False], ["x", 0])(tt.dvector()))
def fail(msg):
......@@ -587,7 +597,7 @@ that the resulting functions compute the same things as the originals.
def XYZab():
return T.matrix(), T.matrix(), T.matrix(), T.scalar(), T.scalar()
return tt.matrix(), tt.matrix(), tt.matrix(), tt.scalar(), tt.scalar()
def just_gemm(
......@@ -602,7 +612,7 @@ def just_gemm(
nb_gemm = 0
for node in f.maker.fgraph.apply_nodes:
assert not isinstance(
node.op, T.Dot
node.op, tt.Dot
), "_dot22 not changed to gemm_inplace in graph"
assert node.op != _dot22
if node.op == gemm_inplace:
......@@ -640,42 +650,42 @@ def test_gemm_opt0():
# Many subgraphs whose dots can be eliminated
X, Y, Z, a, b = XYZab()
just_gemm([X, Y, Z, a, b], [T.dot(X, Y) * a + Z * b])
just_gemm([X, Y, Z, a, b], [a * T.dot(X, Y) + b * Z])
just_gemm([X, Y, Z, a, b], [b * Z + a * T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [T.dot(X, Y) * a - Z * b])
just_gemm([X, Y, Z, a, b], [a * T.dot(X, Y) - b * Z])
just_gemm([X, Y, Z, a, b], [b * Z - a * T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [tt.dot(X, Y) * a + Z * b])
just_gemm([X, Y, Z, a, b], [a * tt.dot(X, Y) + b * Z])
just_gemm([X, Y, Z, a, b], [b * Z + a * tt.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [tt.dot(X, Y) * a - Z * b])
just_gemm([X, Y, Z, a, b], [a * tt.dot(X, Y) - b * Z])
just_gemm([X, Y, Z, a, b], [b * Z - a * tt.dot(X, Y)])
# with transposes (transposes should be pushed through dot in canonicalize)
just_gemm([X, Y, Z, a, b], [b * Z.T - a * T.dot(Y.T, X.T)])
just_gemm([X, Y, Z, a, b], [b * Z.T + a * b * T.dot(X, Y).T])
just_gemm([X, Y, Z, a, b], [b * Z.T - a * tt.dot(Y.T, X.T)])
just_gemm([X, Y, Z, a, b], [b * Z.T + a * b * tt.dot(X, Y).T])
just_gemm(
[X, Y, Z, a, b],
[b * Z + a * T.dot(X, Y).T],
[b * Z + a * tt.dot(X, Y).T],
ishapes=[(5, 3), (3, 4), (4, 5), (), ()],
)
# with N multiplications instead of just one
just_gemm([X, Y, Z, a, b], [(b * b) * Z * a + (a * a) * T.dot(X, Y) * b])
just_gemm([X, Y, Z, a, b], [Z + T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z * b + T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z + a * b * a * T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [(b * b) * Z * a - (a * a) * T.dot(X, Y) * b])
just_gemm([X, Y, Z, a, b], [Z - T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z * b - T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z - a * b * a * T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [(b * b) * Z * a + (a * a) * tt.dot(X, Y) * b])
just_gemm([X, Y, Z, a, b], [Z + tt.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z * b + tt.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z + a * b * a * tt.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [(b * b) * Z * a - (a * a) * tt.dot(X, Y) * b])
just_gemm([X, Y, Z, a, b], [Z - tt.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z * b - tt.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z - a * b * a * tt.dot(X, Y)])
@unittest_tools.assertFailure_fast
def test_gemm_opt_double_gemm():
# This is the pattern that shows up in the autoencoder
X, Y, Z, a, b = T.matrix(), T.matrix(), T.matrix(), T.scalar(), T.scalar()
R, S, c = T.matrix(), T.matrix(), T.scalar()
X, Y, Z, a, b = tt.matrix(), tt.matrix(), tt.matrix(), tt.scalar(), tt.scalar()
R, S, c = tt.matrix(), tt.matrix(), tt.scalar()
just_gemm(
[X, Y, Z, a, b, R, S, c],
[Z * c + a * T.dot(X, Y) + b * T.dot(R, S).T],
[Z * c + a * tt.dot(X, Y) + b * tt.dot(R, S).T],
ishapes=[(4, 3), (3, 5), (4, 5), (), (), (5, 9), (9, 4), ()],
expected_nb_gemm=2,
)
......@@ -684,8 +694,8 @@ def test_gemm_opt_double_gemm():
i = [X, Y, Z, a, b, R, S, c]
o = [
(
a * T.dot(X, Y)
+ gemm_inplace(Z, b, S.T, R.T, T.constant(1.0).astype(config.floatX))
a * tt.dot(X, Y)
+ gemm_inplace(Z, b, S.T, R.T, tt.constant(1.0).astype(config.floatX))
)
]
f = inplace_func(
......@@ -695,7 +705,7 @@ def test_gemm_opt_double_gemm():
on_unused_input="ignore",
)
for node in f.maker.fgraph.apply_nodes:
assert not isinstance(node.op, T.Dot)
assert not isinstance(node.op, tt.Dot)
assert node.op != _dot22
g = inplace_func(
i,
......@@ -717,16 +727,16 @@ def test_gemm_opt_double_gemm():
def test_gemm_canonicalize():
X, Y, Z, a, b = (
T.matrix("X"),
T.matrix("Y"),
T.matrix("Z"),
T.scalar("a"),
T.scalar("b"),
tt.matrix("X"),
tt.matrix("Y"),
tt.matrix("Z"),
tt.scalar("a"),
tt.scalar("b"),
)
c, d = T.scalar("c"), T.scalar("d")
u = T.row("u")
v = T.vector("v")
w = T.col("w")
c, d = tt.scalar("c"), tt.scalar("d")
u = tt.row("u")
v = tt.vector("v")
w = tt.col("w")
can = []
_gemm_canonicalize(X + Y + Z, 1.0, can, 0)
......@@ -744,7 +754,7 @@ def test_gemm_canonicalize():
assert len(can[2]) == 2
assert can[2][0] == 1.0
assert can[2][1].owner
assert isinstance(can[2][1].owner.op, T.DimShuffle)
assert isinstance(can[2][1].owner.op, tt.DimShuffle)
assert can[2][1].owner.inputs == [v]
can = []
......@@ -755,26 +765,26 @@ def test_gemm_canonicalize():
_gemm_canonicalize(a * X + Y - b * Z * c, 1.0, can, 0)
assert can[0] == (a, X)
assert can[1] == (1.0, Y)
assert can[2][0].owner.op == T.mul
assert can[2][0].owner.inputs[0].owner.op == T.neg
assert can[2][0].owner.op == tt.mul
assert can[2][0].owner.inputs[0].owner.op == tt.neg
assert can[2][0].owner.inputs[0].owner.inputs[0] == c
assert can[2][0].owner.inputs[1] == b
can = []
_gemm_canonicalize((-d) * X - (a * X + Y - b * Z * c), 1.0, can, 0)
# print can
assert can[0][0].owner.op == T.neg
assert can[0][0].owner.op == tt.neg
assert can[0][0].owner.inputs[0] == d
assert can[0][1] == X
assert can[1][0].owner.op == T.neg
assert can[1][0].owner.op == tt.neg
assert can[1][0].owner.inputs[0] == a
assert can[2] == (-1.0, Y)
assert can[3][0].owner.op == T.mul
assert can[3][0].owner.op == tt.mul
assert can[3][0].owner.inputs == [c, b]
def test_gemm_factor():
X, Y = T.matrix("X"), T.matrix("Y")
X, Y = tt.matrix("X"), tt.matrix("Y")
assert [(1.0, X), (1.0, Y)] == _factor_canonicalized([(1.0, X), (1.0, Y)])
assert [(2.0, X)] == _factor_canonicalized([(1.0, X), (1.0, X)])
......@@ -783,27 +793,27 @@ def test_gemm_factor():
def test_upcasting_scalar_nogemm():
# Test that the optimization does not crash when the scale has an incorrect
# dtype, and forces upcasting of the result
v = T.fmatrix("v")
w = T.fmatrix("w")
t = T.fmatrix("t")
alpha = T.dscalar("a")
v = tt.fmatrix("v")
w = tt.fmatrix("w")
t = tt.fmatrix("t")
alpha = tt.dscalar("a")
rval = T.dot(w, v) * alpha + t
rval = tt.dot(w, v) * alpha + t
f = theano.function([w, v, t, alpha], rval)
t = f.maker.fgraph.toposort()
assert np.sum([isinstance(n.op, Gemm) for n in t]) == 0
# theano.printing.debugprint(f, print_type=True)
v = T.fmatrix("v")
w = T.fmatrix("w")
t = T.fmatrix("t")
alpha = T.cscalar("a")
v = tt.fmatrix("v")
w = tt.fmatrix("w")
t = tt.fmatrix("t")
alpha = tt.cscalar("a")
on_opt_error = config.on_opt_error
try:
config.on_opt_error = "raise"
rval = T.dot(w, v) * alpha + t
rval = tt.dot(w, v) * alpha + t
f = theano.function([w, v, t, alpha], rval)
finally:
config.on_opt_error = on_opt_error
......@@ -815,51 +825,51 @@ def test_upcasting_scalar_nogemm():
def test_gemm_nested():
X, Y, Z, a, b = (
T.matrix("X"),
T.matrix("Y"),
T.matrix("Z"),
T.scalar("a"),
T.scalar("b"),
tt.matrix("X"),
tt.matrix("Y"),
tt.matrix("Z"),
tt.scalar("a"),
tt.scalar("b"),
)
R, S, U, c, d = (
T.matrix("R"),
T.matrix("S"),
T.matrix("U"),
T.scalar("c"),
T.scalar("d"),
tt.matrix("R"),
tt.matrix("S"),
tt.matrix("U"),
tt.scalar("c"),
tt.scalar("d"),
)
just_gemm(
[X, Y, Z, R, S, U, a, b, c, d],
[a * Z - b * (c * T.dot(X, Y) + d * Z)],
[a * Z - b * (c * tt.dot(X, Y) + d * Z)],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4), (2, 4), (), (), (), ()],
max_graphlen=1,
)
# print "---------------------"
just_gemm(
[X, Y, Z, R, S, U, a, b, c, d],
[a * Z - b * (c * T.dot(X, Y) + d * Z + c * Z)],
[a * Z - b * (c * tt.dot(X, Y) + d * Z + c * Z)],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4), (2, 4), (), (), (), ()],
max_graphlen=1,
)
# print "---------------------"
just_gemm(
[X, Y, Z, R, S, U, a, b, c, d],
[a * Z - b * (c * T.dot(X, Y) + d * Z + c * U)],
[a * Z - b * (c * tt.dot(X, Y) + d * Z + c * U)],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4), (2, 4), (), (), (), ()],
max_graphlen=3,
)
def test_gemm_opt_wishlist():
X, Y, Z, a, b = T.matrix(), T.matrix(), T.matrix(), T.scalar(), T.scalar()
X, Y, Z, a, b = tt.matrix(), tt.matrix(), tt.matrix(), tt.scalar(), tt.scalar()
# with >2 additions of the same T.dot(X,Y term
just_gemm(
[X, Y, Z, a, b], [(b * b) * Z * a + (a * a) * T.dot(X, Y) + b * T.dot(X, Y)]
[X, Y, Z, a, b], [(b * b) * Z * a + (a * a) * tt.dot(X, Y) + b * tt.dot(X, Y)]
)
just_gemm([X, Y, Z, a, b], [Z + T.dot(X, Y) + T.dot(X, Y)])
just_gemm([X, Y, Z, a, b], [Z + tt.dot(X, Y) + tt.dot(X, Y)])
def test_gemm_with_vector():
......@@ -868,39 +878,39 @@ def test_gemm_with_vector():
# bug.
X, Y, Z, a, b = XYZab()
v = T.vector()
v = tt.vector()
def my_just_gemm(o):
i = [X, Y, Z, a, b, v]
ishapes = [(4, 3), (3, 5), (4, 5), (), (), (5,)]
just_gemm(i, o, ishapes=ishapes)
my_just_gemm([v + T.dot(X, Y) * a + Z * b])
my_just_gemm([v + a * T.dot(X, Y) + b * Z])
my_just_gemm([v + b * Z + a * T.dot(X, Y)])
my_just_gemm([v + T.dot(X, Y) * a - Z * b])
my_just_gemm([v + a * T.dot(X, Y) - b * Z])
my_just_gemm([v + b * Z - a * T.dot(X, Y)])
my_just_gemm([v + tt.dot(X, Y) * a + Z * b])
my_just_gemm([v + a * tt.dot(X, Y) + b * Z])
my_just_gemm([v + b * Z + a * tt.dot(X, Y)])
my_just_gemm([v + tt.dot(X, Y) * a - Z * b])
my_just_gemm([v + a * tt.dot(X, Y) - b * Z])
my_just_gemm([v + b * Z - a * tt.dot(X, Y)])
# with N multiplications instead of just one
my_just_gemm([v + (b * b) * Z * a + (a * a) * T.dot(X, Y) * b])
my_just_gemm([v + Z + T.dot(X, Y)])
my_just_gemm([v + Z * b + T.dot(X, Y)])
my_just_gemm([v + Z + a * b * a * T.dot(X, Y)])
my_just_gemm([v + (b * b) * Z * a - (a * a) * T.dot(X, Y) * b])
my_just_gemm([Z - T.dot(X, Y) + v])
my_just_gemm([Z * b - T.dot(X, Y) + v])
my_just_gemm([Z - a * b * a * T.dot(X, Y) + v])
my_just_gemm([v + (b * b) * Z * a + (a * a) * tt.dot(X, Y) * b])
my_just_gemm([v + Z + tt.dot(X, Y)])
my_just_gemm([v + Z * b + tt.dot(X, Y)])
my_just_gemm([v + Z + a * b * a * tt.dot(X, Y)])
my_just_gemm([v + (b * b) * Z * a - (a * a) * tt.dot(X, Y) * b])
my_just_gemm([Z - tt.dot(X, Y) + v])
my_just_gemm([Z * b - tt.dot(X, Y) + v])
my_just_gemm([Z - a * b * a * tt.dot(X, Y) + v])
def test_gemm_opt_vector_stuff():
X, Y, a = T.matrix(), T.matrix(), T.scalar()
u, v = T.vector(), T.vector()
X, Y, a = tt.matrix(), tt.matrix(), tt.scalar()
u, v = tt.vector(), tt.vector()
f = inplace_func([a, u, v], a + T.dot(u, v), mode="FAST_RUN")
f = inplace_func([a, u, v], a + tt.dot(u, v), mode="FAST_RUN")
assert gemm_inplace not in [n.op for n in f.maker.fgraph.apply_nodes]
f = inplace_func([a, u, X, Y], a * u + T.dot(X, Y), mode="FAST_RUN")
f = inplace_func([a, u, X, Y], a * u + tt.dot(X, Y), mode="FAST_RUN")
assert gemm_inplace not in [n.op for n in f.maker.fgraph.apply_nodes]
......@@ -925,10 +935,10 @@ def test_gemm_unrolled():
cur_H = H
def update_V(cur_H):
return T.nnet.sigmoid(T.dot(cur_H, W.T))
return sigmoid(tt.dot(cur_H, W.T))
def update_H(cur_V):
return T.nnet.sigmoid(T.dot(cur_V, W) + T.dot(G, W.T))
return sigmoid(tt.dot(cur_V, W) + tt.dot(G, W.T))
for i in range(num_rounds):
cur_V = update_V(cur_H)
......@@ -944,9 +954,9 @@ def test_gemm_unrolled():
if isinstance(
node.op,
(
theano.tensor.Dot,
theano.tensor.blas.Dot22,
theano.tensor.blas.Gemm,
tt.Dot,
Dot22,
Gemm,
),
)
]
......@@ -962,22 +972,22 @@ def test_inplace0():
# should fail to insert gemm_inplace because gemm_inplace would
# create cycles
X, Y, Z, a, b = (
T.matrix("X"),
T.matrix("Y"),
T.matrix("Z"),
T.scalar("a"),
T.scalar("b"),
tt.matrix("X"),
tt.matrix("Y"),
tt.matrix("Z"),
tt.scalar("a"),
tt.scalar("b"),
)
R, S, c = T.matrix("R"), T.matrix("S"), T.scalar("c")
R, S, c = tt.matrix("R"), tt.matrix("S"), tt.scalar("c")
f = inplace_func([Z, b, R, S], [Z * (Z + b * T.dot(R, S).T)], mode="FAST_RUN")
f = inplace_func([Z, b, R, S], [Z * (Z + b * tt.dot(R, S).T)], mode="FAST_RUN")
assert gemm_inplace not in [n.op for n in f.maker.fgraph.apply_nodes]
assert gemm_no_inplace in [n.op for n in f.maker.fgraph.apply_nodes]
# gemm_inplace should be inserted here, to work in-place on Z*c
f = inplace_func(
[X, Y, Z, a, b, R, S, c],
[Z * (c * Z + a * T.dot(X, Y) + b * T.dot(R, S).T)],
[Z * (c * Z + a * tt.dot(X, Y) + b * tt.dot(R, S).T)],
mode="FAST_RUN",
)
assert gemm_inplace in [n.op for n in f.maker.fgraph.apply_nodes]
......@@ -986,7 +996,7 @@ def test_inplace0():
def test_inplace1():
X, Y, Z, a, b = XYZab()
# with > 2 terms in the overall addition
f = inplace_func([X, Y, Z], [Z + Z + T.dot(X, Y)], mode="FAST_RUN")
f = inplace_func([X, Y, Z], [Z + Z + tt.dot(X, Y)], mode="FAST_RUN")
# theano.printing.debugprint(f)
# it doesn't work inplace because we didn't mark Z as mutable input
assert [n.op for n in f.maker.fgraph.apply_nodes] == [gemm_no_inplace]
......@@ -994,15 +1004,15 @@ def test_inplace1():
def test_dot22():
for dtype1 in ["float32", "float64", "complex64", "complex128"]:
a = T.matrix(dtype=dtype1)
a = tt.matrix(dtype=dtype1)
for dtype2 in ["float32", "float64", "complex64", "complex128"]:
b = T.matrix(dtype=dtype2)
f = theano.function([a, b], T.dot(a, b), mode=mode_blas_opt)
b = tt.matrix(dtype=dtype2)
f = theano.function([a, b], tt.dot(a, b), mode=mode_blas_opt)
topo = f.maker.fgraph.toposort()
if dtype1 == dtype2:
assert _dot22 in [x.op for x in topo], (dtype1, dtype2)
else:
check = [isinstance(x.op, T.Dot) for x in topo]
check = [isinstance(x.op, tt.Dot) for x in topo]
assert any(check), (dtype1, dtype2)
rng = np.random.RandomState(unittest_tools.fetch_seed())
......@@ -1029,14 +1039,14 @@ def test_dot22scalar():
# m = theano.compile.get_default_mode().including('BlasOpt', 'specialize')
rng = np.random.RandomState(unittest_tools.fetch_seed())
for dtype1 in ["complex64", "complex128"]:
a = T.matrix("a", dtype=dtype1)
a = tt.matrix("a", dtype=dtype1)
for dtype2 in ["complex64", "complex128"]:
b = T.matrix("b", dtype=dtype2)
b = tt.matrix("b", dtype=dtype2)
for dtype3 in ["complex64", "complex128"]:
c = T.matrix("c", dtype=dtype3)
c = tt.matrix("c", dtype=dtype3)
for dtype4 in ["complex64", "complex128"]:
cst = theano.tensor.basic.constant(0.2, dtype=dtype4)
cst2 = theano.tensor.basic.constant(0.1, dtype=dtype4)
cst = tt.constant(0.2, dtype=dtype4)
cst2 = tt.constant(0.1, dtype=dtype4)
def check_dot22scalar(func, len_topo_scalar=-1):
topo = func.maker.fgraph.toposort()
......@@ -1072,7 +1082,7 @@ def test_dot22scalar():
elif dtype1 == dtype2:
assert _dot22 in ops, (dtype1, dtype2, dtype3, dtype4)
else:
check = [isinstance(o, T.Dot) for o in ops]
check = [isinstance(o, tt.Dot) for o in ops]
assert any(check), (dtype1, dtype2, dtype3, dtype4)
def cmp(a_shp, b_shp, c_shp, sqr_shp=(5, 5)):
......@@ -1083,7 +1093,7 @@ def test_dot22scalar():
if False:
f = theano.function(
[a, b], cst * T.dot(a, b), mode=mode_blas_opt
[a, b], cst * tt.dot(a, b), mode=mode_blas_opt
)
f.maker.fgraph.toposort()
check_dot22scalar(f, 1)
......@@ -1092,7 +1102,7 @@ def test_dot22scalar():
if True:
f = theano.function(
[a, b, c], cst * c * T.dot(a, b), mode=mode_blas_opt
[a, b, c], cst * c * tt.dot(a, b), mode=mode_blas_opt
)
f.maker.fgraph.toposort()
check_dot22scalar(f, 2)
......@@ -1100,7 +1110,7 @@ def test_dot22scalar():
f(av, bv, cv)
f = theano.function(
[a, b, c], c * cst * T.dot(a, b), mode=mode_blas_opt
[a, b, c], c * cst * tt.dot(a, b), mode=mode_blas_opt
)
f.maker.fgraph.toposort()
check_dot22scalar(f, 2)
......@@ -1110,7 +1120,7 @@ def test_dot22scalar():
# TODO: add only the optimizations needed?
m2 = mode_blas_opt.including("canonicalize")
f = theano.function(
[a, b, c], cst2 * c * cst * T.dot(a, b), mode=m2
[a, b, c], cst2 * c * cst * tt.dot(a, b), mode=m2
)
f.maker.fgraph.toposort()
check_dot22scalar(f, 2)
......@@ -1118,14 +1128,16 @@ def test_dot22scalar():
if dtype1 == dtype2 == dtype3:
f = theano.function(
[a, b, c], c * cst * a * T.dot(a, b), mode=m2
[a, b, c], c * cst * a * tt.dot(a, b), mode=m2
)
f.maker.fgraph.toposort()
check_dot22scalar(f, 2)
f(sv, sv, sv)
f = theano.function(
[a, b, c], cst * c * a * T.dot(a, b), mode=mode_blas_opt
[a, b, c],
cst * c * a * tt.dot(a, b),
mode=mode_blas_opt,
)
f.maker.fgraph.toposort()
# currently the canonizer don't always
......@@ -1141,7 +1153,7 @@ def test_dot22scalar():
f(sv, sv, sv)
f = theano.function(
[a, b, c], c * a * cst * T.dot(a, b), mode=m2
[a, b, c], c * a * cst * tt.dot(a, b), mode=m2
)
f.maker.fgraph.toposort()
check_dot22scalar(f, 2)
......@@ -1158,15 +1170,15 @@ def test_dot22scalar():
def test_dot22scalar_cast():
# Test that in `dot22_to_dot22scalar` we properly cast integers to floats.
# Note that this test was failing before d5ff6904.
A = T.dmatrix()
for scalar_int_type in T.int_dtypes:
y = T.scalar(dtype=scalar_int_type)
f = theano.function([A, y], T.dot(A, A) * y, mode=mode_blas_opt)
A = tt.dmatrix()
for scalar_int_type in tt.int_dtypes:
y = tt.scalar(dtype=scalar_int_type)
f = theano.function([A, y], tt.dot(A, A) * y, mode=mode_blas_opt)
assert _dot22scalar in [x.op for x in f.maker.fgraph.toposort()]
A = T.fmatrix()
for scalar_int_type in T.int_dtypes:
y = T.scalar(dtype=scalar_int_type)
f = theano.function([A, y], T.dot(A, A) * y, mode=mode_blas_opt)
A = tt.fmatrix()
for scalar_int_type in tt.int_dtypes:
y = tt.scalar(dtype=scalar_int_type)
f = theano.function([A, y], tt.dot(A, A) * y, mode=mode_blas_opt)
if scalar_int_type in ["int32", "int64"]:
assert _dot22 in [x.op for x in f.maker.fgraph.toposort()]
else:
......@@ -1175,46 +1187,46 @@ def test_dot22scalar_cast():
def test_local_dot22_to_dot22scalar():
# This test that the bug in gh-1507 is really fixed
A = T.dmatrix()
A = tt.dmatrix()
mode = theano.compile.mode.get_default_mode()
opt = theano.tensor.opt.in2out(theano.tensor.blas.local_dot22_to_dot22scalar)
opt = in2out(local_dot22_to_dot22scalar)
mode = mode.__class__(optimizer=opt)
x = T.dscalar()
y = T.dscalar()
z = T.dscalar()
x = tt.dscalar()
y = tt.dscalar()
z = tt.dscalar()
# make sure to don't have dimshuffle as we don't opt those cases
m = T.dmatrix()
r = T.drow()
m = tt.dmatrix()
r = tt.drow()
for idx, node in enumerate(
[
# Old working cases
T.mul(_dot22(A, A), x),
T.mul(_dot22(A, A), x, y),
T.mul(_dot22(A, A), x, r),
T.mul(_dot22(A, A), m, x),
T.mul(_dot22(A, A), x, m),
T.mul(_dot22(A, A), x, (m * y)),
T.mul(_dot22(A, A), (m * y), x),
T.mul(_dot22(A, A), x, (r * y)),
T.mul(_dot22(A, A), (r * y), x),
T.mul(_dot22(A, A), (x * y), (m * x)),
T.mul(_dot22(A, A), (r * y), (y * x)),
tt.mul(_dot22(A, A), x),
tt.mul(_dot22(A, A), x, y),
tt.mul(_dot22(A, A), x, r),
tt.mul(_dot22(A, A), m, x),
tt.mul(_dot22(A, A), x, m),
tt.mul(_dot22(A, A), x, (m * y)),
tt.mul(_dot22(A, A), (m * y), x),
tt.mul(_dot22(A, A), x, (r * y)),
tt.mul(_dot22(A, A), (r * y), x),
tt.mul(_dot22(A, A), (x * y), (m * x)),
tt.mul(_dot22(A, A), (r * y), (y * x)),
# Case that was raising an assert that is fixed in gh-1507
T.mul(_dot22(A, A), (m * y), m),
T.mul(_dot22(A, A), m, (m * y)),
T.mul(_dot22(A, A), (r * y), (m * x)),
tt.mul(_dot22(A, A), (m * y), m),
tt.mul(_dot22(A, A), m, (m * y)),
tt.mul(_dot22(A, A), (r * y), (m * x)),
# assert fixed in gh-1507 and opt case added in gh-1515
T.mul(_dot22(A, A), (m * y * z), m),
T.mul(_dot22(A, A), m, (m * y * z)),
tt.mul(_dot22(A, A), (m * y * z), m),
tt.mul(_dot22(A, A), m, (m * y * z)),
# Opt case added in gh-1515
T.mul(_dot22(A, A), T.mul(m, y, z), m),
T.mul(_dot22(A, A), m, T.mul(m, y, z)),
tt.mul(_dot22(A, A), tt.mul(m, y, z), m),
tt.mul(_dot22(A, A), m, tt.mul(m, y, z)),
# Case that opt later in gh-1515
T.mul(_dot22(A, A), (r * m), (m * x)),
tt.mul(_dot22(A, A), (r * m), (m * x)),
]
):
node2 = theano.tensor.blas.local_dot22_to_dot22scalar.transform(node.owner)
node2 = local_dot22_to_dot22scalar.transform(node.owner)
assert node2
f = theano.function(
[x, y, z, m, r, A], node, mode=mode, on_unused_input="ignore"
......@@ -1228,11 +1240,11 @@ def test_dot_w_self():
# one of the inputs.
A = shared(value=np.ones((2, 2)))
B = T.matrix()
B = tt.matrix()
p = T.dot(A, A) * B
p = tt.dot(A, A) * B
grad = T.grad(T.mean(p), A)
grad = tt.grad(tt.mean(p), A)
f = theano.function([B], p, updates=[(A, A - grad)])
# tests correctness in debugmode
......@@ -1253,7 +1265,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
f = theano.function([], theano.dot(v, w), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
self.assertFunctionContains0(f, T.dot)
self.assertFunctionContains0(f, tt.dot)
self.assertFunctionContains1(f, Gemv(True))
# Assert they produce the same output
......@@ -1267,7 +1279,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
f = theano.function([], theano.dot(v, m), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
self.assertFunctionContains0(f, T.dot)
self.assertFunctionContains0(f, tt.dot)
self.assertFunctionContains1(f, Gemv(True))
# Assert they produce the same output
......@@ -1284,7 +1296,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
f = theano.function([], theano.dot(m, v), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
self.assertFunctionContains0(f, T.dot)
self.assertFunctionContains0(f, tt.dot)
self.assertFunctionContains1(f, Gemv(True))
# Assert they produce the same output
......@@ -1402,7 +1414,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
assert sum(isinstance(node.op, Gemv) for node in topo) == 1
# call gemv directly for mixed broadcast pattern.
o = theano.tensor.blas.gemv_no_inplace(v2, 0.5, m, v1, 0.25)
o = gemv_no_inplace(v2, 0.5, m, v1, 0.25)
f = theano.function([], o, mode=mode_blas_opt)
assert np.allclose(
f(), 0.5 * np.dot(m.get_value(), v1.get_value()) + 0.25 * v2.get_value()
......@@ -1411,12 +1423,12 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
assert sum(isinstance(node.op, Gemv) for node in topo) == 1
def test_gemv_dimensions(self):
A = T.matrix("A")
x, y = T.vectors("x", "y")
A = tt.matrix("A")
x, y = tt.vectors("x", "y")
alpha = theano.shared(theano._asarray(1.0, dtype=config.floatX), name="alpha")
beta = theano.shared(theano._asarray(1.0, dtype=config.floatX), name="beta")
z = beta * y + alpha * T.dot(A, x)
z = beta * y + alpha * tt.dot(A, x)
f = theano.function([A, x, y], z)
# Matrix value
......@@ -1487,7 +1499,7 @@ class BaseGemv(object):
+ beta.get_value() * y.get_value()
)
oy = alpha * T.dot(a, x) + beta * y
oy = alpha * tt.dot(a, x) + beta * y
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1507,7 +1519,7 @@ class BaseGemv(object):
desired_oy = matrixmultiply(a_v, x_v)
oy = T.dot(a, x)
oy = tt.dot(a, x)
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1523,7 +1535,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(transpose(a_v), x_v) + beta_v * y_v
oy = alpha * T.dot(a.T, x) + beta * y
oy = alpha * tt.dot(a.T, x) + beta * y
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1539,7 +1551,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(a_v, x_v[::2]) + beta_v * y_v
oy = alpha * T.dot(a, x[::2]) + beta * y
oy = alpha * tt.dot(a, x[::2]) + beta * y
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1555,7 +1567,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(transpose(a_v), x_v[::2]) + beta_v * y_v
oy = alpha * T.dot(a.T, x[::2]) + beta * y
oy = alpha * tt.dot(a.T, x[::2]) + beta * y
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1571,7 +1583,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(a_v, x_v) + beta_v * y_v[::2]
oy = alpha * T.dot(a, x) + beta * y[::2]
oy = alpha * tt.dot(a, x) + beta * y[::2]
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1587,7 +1599,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(transpose(a_v), x_v) + beta_v * y_v[::2]
oy = alpha * T.dot(a.T, x) + beta * y[::2]
oy = alpha * tt.dot(a.T, x) + beta * y[::2]
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1607,7 +1619,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(a_v, x_v) + beta_v * y_v
oy = alpha * T.dot(a, x) + beta * y
oy = alpha * tt.dot(a, x) + beta * y
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1627,7 +1639,7 @@ class BaseGemv(object):
desired_oy = alpha_v * matrixmultiply(transpose(a_v), x_v) + beta_v * y_v
oy = alpha * T.dot(a.T, x) + beta * y
oy = alpha * tt.dot(a.T, x) + beta * y
oy_func = theano.function([], oy, mode=self.mode)
......@@ -1647,12 +1659,12 @@ class BaseGemv(object):
x_v = x_v.astype("float32")
y_v = y_v.astype("float32")
alpha = T.dscalar("alpha")
alpha = tt.dscalar("alpha")
a = self.shared(a_v)
x = self.shared(x_v)
y = self.shared(y_v)
rval = T.dot(a, x) * alpha + y
rval = tt.dot(a, x) * alpha + y
f = theano.function([alpha], rval, mode=self.mode)
# this function is currently optimized so that the gemv is
......@@ -1671,14 +1683,14 @@ class BaseGemv(object):
class TestSgemv(BaseGemv, unittest_tools.OptimizationTestMixin):
dtype = float32
gemv = theano.tensor.blas.gemv_no_inplace
gemv_inplace = theano.tensor.blas.gemv_inplace
gemv = gemv_no_inplace
gemv_inplace = gemv_inplace
class TestDgemv(BaseGemv, unittest_tools.OptimizationTestMixin):
dtype = float64
gemv = theano.tensor.blas.gemv_no_inplace
gemv_inplace = theano.tensor.blas.gemv_inplace
gemv = gemv_no_inplace
gemv_inplace = gemv_inplace
# The optimization to put Gemv don't work for complex type for now.
......@@ -1696,30 +1708,30 @@ class TestDgemv(BaseGemv, unittest_tools.OptimizationTestMixin):
class TestGerMakeNode:
def setup_method(self):
self.iv = T.tensor(dtype="int32", broadcastable=(False,))
self.fv = T.tensor(dtype="float32", broadcastable=(False,))
self.fv1 = T.tensor(dtype="float32", broadcastable=(True,))
self.dv = T.tensor(dtype="float64", broadcastable=(False,))
self.dv1 = T.tensor(dtype="float64", broadcastable=(True,))
self.cv = T.tensor(dtype="complex64", broadcastable=(False,))
self.zv = T.tensor(dtype="complex128", broadcastable=(False,))
self.fv_2 = T.tensor(dtype="float32", broadcastable=(False,))
self.fv1_2 = T.tensor(dtype="float32", broadcastable=(True,))
self.dv_2 = T.tensor(dtype="float64", broadcastable=(False,))
self.dv1_2 = T.tensor(dtype="float64", broadcastable=(True,))
self.cv_2 = T.tensor(dtype="complex64", broadcastable=(False,))
self.zv_2 = T.tensor(dtype="complex128", broadcastable=(False,))
self.fm = T.fmatrix()
self.dm = T.dmatrix()
self.cm = T.cmatrix()
self.zm = T.zmatrix()
self.fa = T.fscalar()
self.da = T.dscalar()
self.ca = T.cscalar()
self.za = T.zscalar()
self.iv = tt.tensor(dtype="int32", broadcastable=(False,))
self.fv = tt.tensor(dtype="float32", broadcastable=(False,))
self.fv1 = tt.tensor(dtype="float32", broadcastable=(True,))
self.dv = tt.tensor(dtype="float64", broadcastable=(False,))
self.dv1 = tt.tensor(dtype="float64", broadcastable=(True,))
self.cv = tt.tensor(dtype="complex64", broadcastable=(False,))
self.zv = tt.tensor(dtype="complex128", broadcastable=(False,))
self.fv_2 = tt.tensor(dtype="float32", broadcastable=(False,))
self.fv1_2 = tt.tensor(dtype="float32", broadcastable=(True,))
self.dv_2 = tt.tensor(dtype="float64", broadcastable=(False,))
self.dv1_2 = tt.tensor(dtype="float64", broadcastable=(True,))
self.cv_2 = tt.tensor(dtype="complex64", broadcastable=(False,))
self.zv_2 = tt.tensor(dtype="complex128", broadcastable=(False,))
self.fm = tt.fmatrix()
self.dm = tt.dmatrix()
self.cm = tt.cmatrix()
self.zm = tt.zmatrix()
self.fa = tt.fscalar()
self.da = tt.dscalar()
self.ca = tt.cscalar()
self.za = tt.zscalar()
def test_works_on_all_valid_dtypes(self):
assert self.fm.type == ger(self.fm, self.fa, self.fv, self.fv_2).type
......@@ -1729,7 +1741,7 @@ class TestGerMakeNode:
def test_fails_on_invalid_dtypes(self):
with pytest.raises(TypeError):
ger(T.imatrix(), T.iscalar(), T.ivector(), T.ivector())
ger(tt.imatrix(), tt.iscalar(), tt.ivector(), tt.ivector())
def test_fails_for_nonscalar_alpha(self):
with pytest.raises(TypeError):
......@@ -1783,10 +1795,10 @@ class TestGer(unittest_tools.OptimizationTestMixin):
self.mode = theano.compile.get_default_mode().including("fast_run")
self.mode = self.mode.excluding("c_blas", "scipy_blas")
dtype = self.dtype = "float64" # optimization isn't dtype-dependent
self.A = T.tensor(dtype=dtype, broadcastable=(False, False))
self.a = T.tensor(dtype=dtype, broadcastable=())
self.x = T.tensor(dtype=dtype, broadcastable=(False,))
self.y = T.tensor(dtype=dtype, broadcastable=(False,))
self.A = tt.tensor(dtype=dtype, broadcastable=(False, False))
self.a = tt.tensor(dtype=dtype, broadcastable=())
self.x = tt.tensor(dtype=dtype, broadcastable=(False,))
self.y = tt.tensor(dtype=dtype, broadcastable=(False,))
self.ger = ger
self.ger_destructive = ger_destructive
self.gemm = gemm_no_inplace
......@@ -1797,11 +1809,11 @@ class TestGer(unittest_tools.OptimizationTestMixin):
return theano.function(inputs, outputs, self.mode, updates=updates)
def b(self, bval):
return T.as_tensor_variable(np.asarray(bval, dtype=self.dtype))
return tt.as_tensor_variable(np.asarray(bval, dtype=self.dtype))
def test_b_0_triggers_ger(self):
# test local_gemm_to_ger opt
assert T.blas.local_gemm_to_ger.transform(
assert local_gemm_to_ger.transform(
gemm_no_inplace(
self.A,
self.a,
......@@ -1813,7 +1825,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
def test_b_1_triggers_ger(self):
# test local_gemm_to_ger opt
assert T.blas.local_gemm_to_ger.transform(
assert local_gemm_to_ger.transform(
gemm_no_inplace(
self.A,
self.a,
......@@ -1825,7 +1837,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
def test_b_other_does_not_triggers_ger(self):
# test local_gemm_to_ger opt
assert not T.blas.local_gemm_to_ger.transform(
assert not local_gemm_to_ger.transform(
gemm_no_inplace(
self.A,
self.a,
......@@ -1837,7 +1849,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
def test_b_nonconst_does_not_triggers_ger(self):
# test local_gemm_to_ger opt
assert not T.blas.local_gemm_to_ger.transform(
assert not local_gemm_to_ger.transform(
gemm_no_inplace(
self.A,
self.a,
......@@ -1848,12 +1860,12 @@ class TestGer(unittest_tools.OptimizationTestMixin):
)
def test_outer(self):
f = self.function([self.x, self.y], T.outer(self.x, self.y))
f = self.function([self.x, self.y], tt.outer(self.x, self.y))
self.assertFunctionContains(f, self.ger_destructive)
f(np.random.rand(5).astype(self.dtype), np.random.rand(4).astype(self.dtype))
def test_A_plus_outer(self):
f = self.function([self.A, self.x, self.y], self.A + T.outer(self.x, self.y))
f = self.function([self.A, self.x, self.y], self.A + tt.outer(self.x, self.y))
self.assertFunctionContains(f, self.ger)
f(
np.random.rand(5, 4).astype(self.dtype),
......@@ -1868,7 +1880,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
def test_A_plus_scaled_outer(self):
f = self.function(
[self.A, self.x, self.y], self.A + 0.1 * T.outer(self.x, self.y)
[self.A, self.x, self.y], self.A + 0.1 * tt.outer(self.x, self.y)
)
self.assertFunctionContains(f, self.ger)
f(
......@@ -1886,7 +1898,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
f = self.function(
[self.A, self.x, self.y],
np.asarray(0.2, self.dtype) * self.A
+ np.asarray(0.1, self.dtype) * T.outer(self.x, self.y),
+ np.asarray(0.1, self.dtype) * tt.outer(self.x, self.y),
)
# Why gemm? This make the graph simpler did we test that it
# make it faster?
......@@ -1906,7 +1918,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
# test corner case shape and dtype
f = self.function(
[self.A, self.x, self.y], self.A + 0.1 * T.outer(self.x, self.y)
[self.A, self.x, self.y], self.A + 0.1 * tt.outer(self.x, self.y)
)
self.assertFunctionContains(f, self.ger)
f(
......@@ -1956,7 +1968,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
[self.x, self.y],
[],
updates=[
(A, A + T.constant(0.1, dtype=self.dtype) * T.outer(self.x, self.y))
(A, A + tt.constant(0.1, dtype=self.dtype) * tt.outer(self.x, self.y))
],
)
self.assertFunctionContains(f, self.ger_destructive)
......@@ -1970,7 +1982,7 @@ class TestGer(unittest_tools.OptimizationTestMixin):
class TestBlasStrides:
dtype = "float64"
shared = staticmethod(tensor._shared)
shared = staticmethod(tt._shared)
mode = theano.compile.get_default_mode()
mode = mode.including("fast_run").excluding("gpu", "c_blas", "scipy_blas")
rng = np.random.RandomState(seed=unittest_tools.fetch_seed())
......@@ -1995,17 +2007,13 @@ class TestBlasStrides:
bt_dev = b_t.get_value(borrow=False, return_internal_type=True)
ct_dev = c_t.get_value(borrow=False, return_internal_type=True)
f_nn = theano.function([], [], updates=[(a, tensor.dot(b, c))], mode=self.mode)
f_nn = theano.function([], [], updates=[(a, tt.dot(b, c))], mode=self.mode)
# print 'class name:', self.__class__.__name__
# theano.printing.debugprint(f_nn)
f_nt = theano.function(
[], [], updates=[(a, tensor.dot(b, c_t.T))], mode=self.mode
)
f_tn = theano.function(
[], [], updates=[(a, tensor.dot(b_t.T, c))], mode=self.mode
)
f_nt = theano.function([], [], updates=[(a, tt.dot(b, c_t.T))], mode=self.mode)
f_tn = theano.function([], [], updates=[(a, tt.dot(b_t.T, c))], mode=self.mode)
f_tt = theano.function(
[], [], updates=[(a, tensor.dot(b_t.T, c_t.T))], mode=self.mode
[], [], updates=[(a, tt.dot(b_t.T, c_t.T))], mode=self.mode
)
# Try with all stride patterns, and all transposed pattern
......@@ -2066,17 +2074,15 @@ class TestBlasStrides:
bt_dev = b_t.get_value(borrow=False, return_internal_type=True)
ct_dev = c_t.get_value(borrow=False, return_internal_type=True)
f_nn = theano.function(
[], [], updates=[(a, l * tensor.dot(b, c))], mode=self.mode
)
f_nn = theano.function([], [], updates=[(a, l * tt.dot(b, c))], mode=self.mode)
f_nt = theano.function(
[], [], updates=[(a, l * tensor.dot(b, c_t.T))], mode=self.mode
[], [], updates=[(a, l * tt.dot(b, c_t.T))], mode=self.mode
)
f_tn = theano.function(
[], [], updates=[(a, l * tensor.dot(b_t.T, c))], mode=self.mode
[], [], updates=[(a, l * tt.dot(b_t.T, c))], mode=self.mode
)
f_tt = theano.function(
[], [], updates=[(a, l * tensor.dot(b_t.T, c_t.T))], mode=self.mode
[], [], updates=[(a, l * tt.dot(b_t.T, c_t.T))], mode=self.mode
)
# Try with all stride patterns, and all transposed pattern
......@@ -2140,30 +2146,30 @@ class TestBlasStrides:
ct_dev = c_t.get_value(borrow=False, return_internal_type=True)
f_nnn = theano.function(
[], [], updates=[(a, (l * a + tensor.dot(b, c)))], mode=self.mode
[], [], updates=[(a, (l * a + tt.dot(b, c)))], mode=self.mode
)
f_nnt = theano.function(
[], [], updates=[(a, (l * a + tensor.dot(b, c_t.T)))], mode=self.mode
[], [], updates=[(a, (l * a + tt.dot(b, c_t.T)))], mode=self.mode
)
f_ntn = theano.function(
[], [], updates=[(a, (l * a + tensor.dot(b_t.T, c)))], mode=self.mode
[], [], updates=[(a, (l * a + tt.dot(b_t.T, c)))], mode=self.mode
)
f_ntt = theano.function(
[], [], updates=[(a, (l * a + tensor.dot(b_t.T, c_t.T)))], mode=self.mode
[], [], updates=[(a, (l * a + tt.dot(b_t.T, c_t.T)))], mode=self.mode
)
f_tnn = theano.function(
[], [], updates=[(a_t, (l * a_t + tensor.dot(b, c).T))], mode=self.mode
[], [], updates=[(a_t, (l * a_t + tt.dot(b, c).T))], mode=self.mode
)
f_tnt = theano.function(
[], [], updates=[(a_t, (l * a_t + tensor.dot(b, c_t.T).T))], mode=self.mode
[], [], updates=[(a_t, (l * a_t + tt.dot(b, c_t.T).T))], mode=self.mode
)
f_ttn = theano.function(
[], [], updates=[(a_t, (l * a_t + tensor.dot(b_t.T, c).T))], mode=self.mode
[], [], updates=[(a_t, (l * a_t + tt.dot(b_t.T, c).T))], mode=self.mode
)
f_ttt = theano.function(
[],
[],
updates=[(a_t, (l * a_t + tensor.dot(b_t.T, c_t.T).T))],
updates=[(a_t, (l * a_t + tt.dot(b_t.T, c_t.T).T))],
mode=self.mode,
)
......@@ -2260,11 +2266,11 @@ class TestBlasStrides:
c_dev = c.get_value(borrow=False, return_internal_type=True)
f_n = theano.function(
[], [], updates=[(a, (a + l * tensor.dot(b, c)))], mode=self.mode
[], [], updates=[(a, (a + l * tt.dot(b, c)))], mode=self.mode
)
f_t = theano.function(
[], [], updates=[(a, (a + l * tensor.dot(b_t.T, c)))], mode=self.mode
[], [], updates=[(a, (a + l * tt.dot(b_t.T, c)))], mode=self.mode
)
# Try with all stride patterns, and all transposed pattern
......@@ -2313,11 +2319,11 @@ class TestBlasStrides:
c_dev = c.get_value(borrow=False, return_internal_type=True)
f_n = theano.function(
[], [], updates=[(a, (a + l * tensor.outer(b, c)))], mode=self.mode
[], [], updates=[(a, (a + l * tt.outer(b, c)))], mode=self.mode
)
f_t = theano.function(
[], [], updates=[(a_t, (a_t + l * tensor.outer(b, c).T))], mode=self.mode
[], [], updates=[(a_t, (a_t + l * tt.outer(b, c).T))], mode=self.mode
)
# Try with all stride patterns, and all transposed patterns
......@@ -2365,8 +2371,8 @@ class TestBlasStrides:
b = theano.shared(bval[:, :5], borrow=True)
c = theano.shared(cval[:3, :5], borrow=True)
s = theano.tensor.scalar()
upd_c = s * c + theano.tensor.dot(a, b)
s = tt.scalar()
upd_c = s * c + tt.dot(a, b)
f = theano.function([s], [], updates={c: upd_c})
f(0)
......@@ -2376,38 +2382,38 @@ class TestBlasStrides:
class TestInferShape(unittest_tools.InferShapeTester):
def test_dot22(self):
x, y = T.matrices("xy")
x, y = tt.matrices("xy")
self._compile_and_check(
[x, y],
[T.blas._dot22(x, y)],
[_dot22(x, y)],
[
np.random.random((2, 3)).astype(config.floatX),
np.random.random((3, 4)).astype(config.floatX),
],
T.blas.Dot22,
Dot22,
)
def test_dot22scalar(self):
x, y = T.matrices("xy")
a = T.scalar("a")
x, y = tt.matrices("xy")
a = tt.scalar("a")
self._compile_and_check(
[x, y, a],
[T.blas._dot22scalar(x, y, a)],
[_dot22scalar(x, y, a)],
[
np.random.random((2, 3)).astype(config.floatX),
np.random.random((3, 4)).astype(config.floatX),
np.asarray(0.5, dtype=config.floatX),
],
T.blas.Dot22Scalar,
Dot22Scalar,
)
def test_gemm(self):
x, y, z = T.matrices("xyz")
a = T.scalar("a")
b = T.scalar("b")
x, y, z = tt.matrices("xyz")
a = tt.scalar("a")
b = tt.scalar("b")
self._compile_and_check(
[x, y, a, z, b],
[T.blas.gemm(z, a, x, y, b)],
[gemm(z, a, x, y, b)],
[
np.random.random((2, 3)).astype(config.floatX),
np.random.random((3, 4)).astype(config.floatX),
......@@ -2415,17 +2421,17 @@ class TestInferShape(unittest_tools.InferShapeTester):
np.random.random((2, 4)).astype(config.floatX),
np.asarray(0.5, dtype=config.floatX),
],
T.blas.Gemm,
Gemm,
)
def test_gemv(self):
A = T.matrix("A")
x, y = T.vectors("xy")
a = T.scalar("a")
b = T.scalar("b")
A = tt.matrix("A")
x, y = tt.vectors("xy")
a = tt.scalar("a")
b = tt.scalar("b")
self._compile_and_check(
[y, a, A, x, b],
[T.blas.gemv(y, a, A, x, b)],
[gemv(y, a, A, x, b)],
[
np.random.random((2,)).astype(config.floatX),
np.asarray(0.5, dtype=config.floatX),
......@@ -2433,21 +2439,21 @@ class TestInferShape(unittest_tools.InferShapeTester):
np.random.random((3,)).astype(config.floatX),
np.asarray(0.5, dtype=config.floatX),
],
T.blas.Gemv,
Gemv,
)
def test_ger(self):
A = T.matrix("A")
x, y = T.vectors("xy")
a = T.scalar("a")
A = tt.matrix("A")
x, y = tt.vectors("xy")
a = tt.scalar("a")
self._compile_and_check(
[A, a, x, y],
[T.blas.ger(A, a, x, y)],
[ger(A, a, x, y)],
[
np.random.random((2, 3)).astype(config.floatX),
np.asarray(0.5, dtype=config.floatX),
np.random.random((2,)).astype(config.floatX),
np.random.random((3,)).astype(config.floatX),
],
T.blas.Ger,
Ger,
)
tests/tensor/test_elemwise.py
浏览文件 @ ef279e19
......@@ -7,15 +7,14 @@ import pytest
import numpy as np
import theano
import theano.tensor as tt
import tests.unittest_tools as utt
from copy import copy
from theano import gof, scalar, config
from theano import tensor
from theano.tensor import TensorType, as_tensor_variable
from theano.compile.mode import get_default_mode, Mode
from theano.tensor.elemwise import (
......@@ -24,7 +23,10 @@ from theano.tensor.elemwise import (
DimShuffle,
Prod,
ProdWithoutZeros,
Sum,
)
from theano.tensor.type import values_eq_approx_remove_nan
from theano.tensor.nnet import sigmoid
from tests import unittest_tools
......@@ -126,37 +128,37 @@ class TestReduceAxes:
def test_sum_axes(self):
axes = [None, 0, 1, [0, 1], np.array(1), [np.array(0), np.array(1)]]
for a in axes:
x = tensor.matrix()
x = tt.matrix()
x.sum(a)
def test_mean_axes(self):
axes = [None, 0, 1, [0, 1], np.array(1), [np.array(0), np.array(1)]]
for a in axes:
x = tensor.matrix()
x = tt.matrix()
x.mean(a)
def test_max_axes(self):
axes = [None, 0, 1, [0, 1], np.array(1), [np.array(0), np.array(1)]]
for a in axes:
x = tensor.matrix()
x = tt.matrix()
x.max(a)
def test_min_axes(self):
axes = [None, 0, 1, [0, 1], np.array(1), [np.array(0), np.array(1)]]
for a in axes:
x = tensor.matrix()
x = tt.matrix()
x.min(a)
def test_argmax_axes(self):
axes = [None, 0, 1, [0, 1], np.array(1), [np.array(0), np.array(1)]]
for a in axes:
x = tensor.matrix()
x = tt.matrix()
x.argmax(a)
def test_var_axes(self):
axes = [None, 0, 1, [0, 1], np.array(1), [np.array(0), np.array(1)]]
for a in axes:
x = tensor.matrix()
x = tt.matrix()
x.var(a)
......@@ -302,7 +304,7 @@ class TestBroadcast:
assert (xv == yv).all()
def test_fill_var(self):
x = tensor.matrix()
x = tt.matrix()
x.fill(3)
def test_fill_grad(self):
......@@ -310,7 +312,7 @@ class TestBroadcast:
# https://groups.google.com/d/topic/theano-users/nQshB8gUA6k/discussion
x = TensorType(config.floatX, [0, 1, 0])("x")
y = TensorType(config.floatX, [0, 1, 0])("y")
e = tensor.second(x, y)
e = tt.second(x, y)
theano.grad(e.sum(), y)
@pytest.mark.skipif(
......@@ -414,7 +416,7 @@ class TestCAReduce(unittest_tools.InferShapeTester):
f = theano.function([x], e, mode=mode)
xv = np.asarray(np.random.rand(*xsh))
if dtype not in tensor.discrete_dtypes:
if dtype not in tt.discrete_dtypes:
xv = np.asarray(xv, dtype=dtype)
else:
xv = np.asarray(xv < 0.5, dtype=dtype)
......@@ -441,12 +443,12 @@ class TestCAReduce(unittest_tools.InferShapeTester):
axis2.append(a)
assert len(axis2) == len(tosum)
tosum = tuple(axis2)
if tensor_op == tensor.all:
if tensor_op == tt.all:
for axis in reversed(sorted(tosum)):
zv = np.all(zv, axis)
if len(tosum) == 0:
zv = zv != 0
elif tensor_op == tensor.any:
elif tensor_op == tt.any:
for axis in reversed(sorted(tosum)):
zv = np.any(zv, axis)
if len(tosum) == 0:
......@@ -537,11 +539,9 @@ class TestCAReduce(unittest_tools.InferShapeTester):
self.with_mode(Mode(linker="py"), scalar.maximum, dtype=dtype)
self.with_mode(Mode(linker="py"), scalar.minimum, dtype=dtype)
self.with_mode(
Mode(linker="py"), scalar.and_, dtype=dtype, tensor_op=tensor.all
)
self.with_mode(
Mode(linker="py"), scalar.or_, dtype=dtype, tensor_op=tensor.any
Mode(linker="py"), scalar.and_, dtype=dtype, tensor_op=tt.all
)
self.with_mode(Mode(linker="py"), scalar.or_, dtype=dtype, tensor_op=tt.any)
for dtype in ["int8", "uint8"]:
self.with_mode(Mode(linker="py"), scalar.or_, dtype=dtype)
self.with_mode(Mode(linker="py"), scalar.and_, dtype=dtype)
......@@ -562,14 +562,14 @@ class TestCAReduce(unittest_tools.InferShapeTester):
scalar.or_,
dtype=dtype,
test_nan=True,
tensor_op=tensor.any,
tensor_op=tt.any,
)
self.with_mode(
Mode(linker="py"),
scalar.and_,
dtype=dtype,
test_nan=True,
tensor_op=tensor.all,
tensor_op=tt.all,
)
@pytest.mark.skipif(
......@@ -591,12 +591,8 @@ class TestCAReduce(unittest_tools.InferShapeTester):
for dtype in ["bool", "floatX", "int8", "uint8"]:
self.with_mode(Mode(linker="c"), scalar.minimum, dtype=dtype)
self.with_mode(Mode(linker="c"), scalar.maximum, dtype=dtype)
self.with_mode(
Mode(linker="c"), scalar.and_, dtype=dtype, tensor_op=tensor.all
)
self.with_mode(
Mode(linker="c"), scalar.or_, dtype=dtype, tensor_op=tensor.any
)
self.with_mode(Mode(linker="c"), scalar.and_, dtype=dtype, tensor_op=tt.all)
self.with_mode(Mode(linker="c"), scalar.or_, dtype=dtype, tensor_op=tt.any)
for dtype in ["bool", "int8", "uint8"]:
self.with_mode(Mode(linker="c"), scalar.or_, dtype=dtype)
self.with_mode(Mode(linker="c"), scalar.and_, dtype=dtype)
......@@ -664,7 +660,7 @@ class TestProd:
# second time, with some added complexity
# verify_grad takes the sum of the matrices anyway
def fn(x2):
return theano.tensor.sqr(Prod(axis=1)(x2))
return tt.sqr(Prod(axis=1)(x2))
unittest_tools.verify_grad(fn, [x_val], mode=self.mode)
......@@ -674,20 +670,20 @@ class TestProd:
x_val = np.asarray(
[[1.0, 2.0, 3.0], [0.0, 5.0, 6.0], [0.0, 0.0, 9.0]], dtype="float32"
)
x = theano.tensor.dmatrix()
x = tt.dmatrix()
# sanity check
p = Prod(axis=1)(x)
# Uncomment this for debugging if needed
# x2 = theano.tensor.dmatrix()
# x2 = tt.dmatrix()
# p2 = Prod(axis=1)(x2)
# fn = theano.function([x, x2], [p - p2], mode=self.mode)
# print("hand computed diff for each row")
# x2_val = np.asarray([[1., 2., 3.003], [0.003, 5., 6], [
# 0., 0., 9.01]])
# print(fn(x_val, x2_val))
# fn2 = theano.function([x], [theano.tensor.grad(p.sum(), x)],
# fn2 = theano.function([x], [tt.grad(p.sum(), x)],
# mode=self.mode)
# print("real grad")
# print(fn2(x_val))
......@@ -700,10 +696,10 @@ class TestProd:
# second time, with some added complexity
# verify_grad takes the sum of the matrices anyway
# def fn5(x5):
# return theano.tensor.sqr(Prod(axis=1)(x5))
# return tt.sqr(Prod(axis=1)(x5))
# x4 = theano.tensor.dmatrix()
# p4 = theano.tensor.sqr(Prod(axis=1)(x4))
# x4 = tt.dmatrix()
# p4 = tt.sqr(Prod(axis=1)(x4))
# fn4 = theano.function([x4], p4)
# print("with sqr")
# print(fn4(x_val))
......@@ -713,7 +709,7 @@ class TestProd:
@pytest.mark.slow
def test_prod_no_zeros_in_input(self):
x = theano.tensor.dmatrix()
x = tt.dmatrix()
x_val = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype="float32")
pwz = Prod(axis=1, no_zeros_in_input=True)(x)
fn = theano.function([x], pwz, mode=self.mode)
......@@ -754,7 +750,7 @@ class TestProd:
unittest_tools.verify_grad(second_deriv, [x_val], mode=self.mode)
def test_prod_without_zeros(self):
x = theano.tensor.dmatrix()
x = tt.dmatrix()
x_val = np.array([[1, 2, 3], [0, 5, 6], [0, 0, 9]], dtype="float32")
pwz = ProdWithoutZeros(axis=1)(x)
fn = theano.function([x], pwz, mode=self.mode)
......@@ -766,14 +762,14 @@ class TestProd:
@pytest.mark.xfail(raises=theano.gradient.NullTypeGradError)
def test_prod_without_zeros_grad(self):
x = theano.tensor.dmatrix()
x = tt.dmatrix()
pwz_a1 = ProdWithoutZeros(axis=0)(x)
pwz_grad = theano.grad(theano.tensor.sum(pwz_a1), x)
pwz_grad = theano.grad(tt.sum(pwz_a1), x)
theano.function([x], pwz_grad, mode=self.mode)
@pytest.mark.slow
def test_other_grad_tests(self):
x = theano.tensor.dmatrix()
x = tt.dmatrix()
x_val1 = np.array([[1, 2, 3], [0, 5, 6], [0, 0, 9]], dtype="float32")
x_val2 = np.array(
[[1, 2, 0], [0, 5, 6], [7, 8, 9], [9, 10, 0]], dtype="float32"
......@@ -781,7 +777,7 @@ class TestProd:
rng = rng = np.random.RandomState(43)
p = Prod(axis=1)
grad_p = theano.tensor.grad(p(x).sum(), x)
grad_p = tt.grad(p(x).sum(), x)
grad_fn = theano.function([x], grad_p, mode=self.mode)
assert np.allclose(
grad_fn(x_val1), [[6.0, 3.0, 2.0], [30.0, 0.0, 0.0], [0.0, 0.0, 0.0]]
......@@ -792,7 +788,7 @@ class TestProd:
)
p_axis0 = Prod(axis=0)
grad_p_axis0 = theano.tensor.grad(p_axis0(x).sum(), x)
grad_p_axis0 = tt.grad(p_axis0(x).sum(), x)
grad_fn_axis0 = theano.function([x], grad_p_axis0, mode=self.mode)
assert np.allclose(
grad_fn_axis0(x_val2),
......@@ -804,12 +800,12 @@ class TestProd:
],
)
tensor.verify_grad(p, [x_val1], rng=rng, mode=self.mode)
tt.verify_grad(p, [x_val1], rng=rng, mode=self.mode)
def test_mul_without_zeros_zeros(self):
a = np.zeros((3, 3))
x = theano.tensor.dmatrix()
x = tt.dmatrix()
mul1 = ProdWithoutZeros(axis=0)(x)
......@@ -838,35 +834,32 @@ class TestIsInfIsNan:
[np.nan, np.inf, -np.inf, 0, 1, -1],
]
]
self.scalar = tensor.scalar()
self.vector = tensor.vector()
self.scalar = tt.scalar()
self.vector = tt.vector()
self.mode = get_default_mode()
if isinstance(self.mode, theano.compile.debugmode.DebugMode):
# Disable the check preventing usage of NaN / Inf values.
self.mode = copy(self.mode)
self.mode.check_isfinite = False
def run_isfunc(self, isfunc):
for input in (self.scalar, self.vector):
theano_isfunc = theano.function(
[input], getattr(tensor, isfunc)(input), mode=self.mode
)
numpy_isfunc = getattr(np, isfunc)
def run_isfunc(self, tt_func, np_func):
for args in (self.scalar, self.vector):
theano_isfunc = theano.function([args], tt_func(args), mode=self.mode)
for x in self.test_vals:
if (x.ndim == 0 and input is not self.scalar) or (
x.ndim == 1 and input is not self.vector
if (x.ndim == 0 and args is not self.scalar) or (
x.ndim == 1 and args is not self.vector
):
# We only test with the appropriate input type.
continue
t_out = theano_isfunc(x)
n_out = numpy_isfunc(x)
n_out = np_func(x)
assert (t_out == n_out).all(), (t_out, n_out)
def test_isinf(self):
return self.run_isfunc("isinf")
self.run_isfunc(tt.isinf, np.isinf)
def test_isnan(self):
return self.run_isfunc("isnan")
self.run_isfunc(tt.isnan, np.isnan)
class TestReduceDtype:
......@@ -882,7 +875,7 @@ class TestReduceDtype:
for method in self.methods:
for idx, dtype in enumerate(self.dtypes):
axis = self.axes[idx % len(self.axes)]
x = tensor.matrix(dtype=dtype)
x = tt.matrix(dtype=dtype)
s = getattr(x, method)(axis=axis)
assert (
s.dtype
......@@ -910,7 +903,7 @@ class TestReduceDtype:
for method in self.methods:
for idx, dtype in enumerate(self.dtypes):
axis = self.axes[idx % len(self.axes)]
x = tensor.matrix(dtype=dtype)
x = tt.matrix(dtype=dtype)
s = getattr(x, method)(axis=axis)
assert (
s.owner.op.acc_dtype
......@@ -942,7 +935,7 @@ class TestReduceDtype:
idx = 0
for method in self.methods:
for input_dtype in self.dtypes:
x = tensor.matrix(dtype=input_dtype)
x = tt.matrix(dtype=input_dtype)
for output_dtype in self.dtypes:
# Only tests case where both input and output are complex.
icomplex = input_dtype.startswith("complex")
......@@ -977,7 +970,7 @@ class TestReduceDtype:
if "complex" in input_dtype:
continue
# Check that we can take the gradient
tensor.grad(var.sum(), x, disconnected_inputs="ignore")
tt.grad(var.sum(), x, disconnected_inputs="ignore")
idx += 1
def test_reduce_custom_acc_dtype(self):
......@@ -987,7 +980,7 @@ class TestReduceDtype:
idx = 0
for method in self.methods:
for input_dtype in self.dtypes:
x = tensor.matrix(dtype=input_dtype)
x = tt.matrix(dtype=input_dtype)
for acc_dtype in self.dtypes:
# If the accumulator is a complex, the gradient of the reduce will
# cast the complex to the input dtype. We can't call the normal
......@@ -1002,8 +995,8 @@ class TestReduceDtype:
# We always allow int/uint inputs with float/complex outputs.
upcasted_dtype = scalar.upcast(input_dtype, acc_dtype)
if acc_dtype == upcasted_dtype or (
input_dtype in tensor.discrete_dtypes
and acc_dtype in tensor.continuous_dtypes
input_dtype in tt.discrete_dtypes
and acc_dtype in tt.continuous_dtypes
):
var = getattr(x, method)(acc_dtype=acc_dtype, axis=axis)
assert var.owner.op.acc_dtype == acc_dtype
......@@ -1011,7 +1004,7 @@ class TestReduceDtype:
if "complex" in input_dtype:
continue
# Check that we can take the gradient
tensor.grad(var.sum(), x, disconnected_inputs="ignore")
tt.grad(var.sum(), x, disconnected_inputs="ignore")
else:
with pytest.raises(TypeError):
getattr(x(method), acc_dtype=acc_dtype, axis=axis)
......@@ -1040,9 +1033,9 @@ class TestMeanDtype:
axes = [None, 0, 1, [], [0], [1], [0, 1]]
for idx, dtype in enumerate(map(str, theano.scalar.all_types)):
axis = axes[idx % len(axes)]
x = tensor.matrix(dtype=dtype)
x = tt.matrix(dtype=dtype)
m = x.mean(axis=axis)
if dtype in tensor.discrete_dtypes:
if dtype in tt.discrete_dtypes:
assert m.dtype == "float64"
else:
assert m.dtype == dtype, (m, m.dtype, dtype)
......@@ -1059,7 +1052,7 @@ class TestMeanDtype:
axes = [None, 0, 1, [], [0], [1], [0, 1]]
idx = 0
for input_dtype in map(str, theano.scalar.all_types):
x = tensor.matrix(dtype=input_dtype)
x = tt.matrix(dtype=input_dtype)
for sum_dtype in map(str, theano.scalar.all_types):
axis = axes[idx % len(axes)]
# If the inner sum cannot be created, it will raise a
......@@ -1070,7 +1063,7 @@ class TestMeanDtype:
pass
else:
# Executed if no TypeError was raised
if sum_dtype in tensor.discrete_dtypes:
if sum_dtype in tt.discrete_dtypes:
assert mean_var.dtype == "float64", (mean_var.dtype, sum_dtype)
else:
assert mean_var.dtype == sum_dtype, (mean_var.dtype, sum_dtype)
......@@ -1086,11 +1079,11 @@ class TestMeanDtype:
if "complex" in mean_var.dtype:
continue
try:
tensor.grad(mean_var.sum(), x, disconnected_inputs="ignore")
tt.grad(mean_var.sum(), x, disconnected_inputs="ignore")
except NotImplementedError:
# TrueDiv does not seem to have a gradient when
# the numerator is complex.
if mean_var.dtype in tensor.complex_dtypes:
if mean_var.dtype in tt.complex_dtypes:
pass
else:
raise
......@@ -1114,7 +1107,7 @@ class TestProdWithoutZerosDtype:
axes = [None, 0, 1, [], [0], [1], [0, 1]]
for idx, dtype in enumerate(map(str, theano.scalar.all_types)):
axis = axes[idx % len(axes)]
x = ProdWithoutZeros(axis=axis)(tensor.matrix(dtype=dtype))
x = ProdWithoutZeros(axis=axis)(tt.matrix(dtype=dtype))
assert (
x.dtype
== dict(
......@@ -1135,7 +1128,7 @@ class TestProdWithoutZerosDtype:
axes = [None, 0, 1, [], [0], [1], [0, 1]]
for idx, dtype in enumerate(map(str, theano.scalar.all_types)):
axis = axes[idx % len(axes)]
x = tensor.matrix(dtype=dtype)
x = tt.matrix(dtype=dtype)
p = ProdWithoutZeros(axis=axis)(x)
assert (
p.owner.op.acc_dtype
......@@ -1168,7 +1161,7 @@ class TestProdWithoutZerosDtype:
axes = [None, 0, 1, [], [0], [1], [0, 1]]
idx = 0
for input_dtype in map(str, theano.scalar.all_types):
x = tensor.matrix(dtype=input_dtype)
x = tt.matrix(dtype=input_dtype)
for output_dtype in map(str, theano.scalar.all_types):
axis = axes[idx % len(axes)]
prod_woz_var = ProdWithoutZeros(axis=axis, dtype=output_dtype)(x)
......@@ -1189,15 +1182,15 @@ class TestProdWithoutZerosDtype:
axes = [None, 0, 1, [], [0], [1], [0, 1]]
idx = 0
for input_dtype in map(str, theano.scalar.all_types):
x = tensor.matrix(dtype=input_dtype)
x = tt.matrix(dtype=input_dtype)
for acc_dtype in map(str, theano.scalar.all_types):
axis = axes[idx % len(axes)]
# If acc_dtype would force a downcast, we expect a TypeError
# We always allow int/uint inputs with float/complex outputs.
upcasted_dtype = scalar.upcast(input_dtype, acc_dtype)
if acc_dtype == upcasted_dtype or (
input_dtype in tensor.discrete_dtypes
and acc_dtype in tensor.continuous_dtypes
input_dtype in tt.discrete_dtypes
and acc_dtype in tt.continuous_dtypes
):
prod_woz_var = ProdWithoutZeros(axis=axis, acc_dtype=acc_dtype)(x)
assert prod_woz_var.owner.op.acc_dtype == acc_dtype
......@@ -1220,7 +1213,7 @@ class TestBitOpReduceGrad:
self.rng = np.random.RandomState(unittest_tools.fetch_seed())
def test_all_grad(self):
x = tensor.bmatrix("x")
x = tt.bmatrix("x")
x_all = x.all()
gx = theano.grad(x_all, x)
f = theano.function([x], gx)
......@@ -1231,7 +1224,7 @@ class TestBitOpReduceGrad:
assert np.all(gx_val == 0)
def test_any_grad(self):
x = tensor.bmatrix("x")
x = tt.bmatrix("x")
x_all = x.any()
gx = theano.grad(x_all, x)
f = theano.function([x], gx)
......@@ -1244,8 +1237,8 @@ class TestBitOpReduceGrad:
class TestElemwise(unittest_tools.InferShapeTester):
def test_elemwise_grad_bool(self):
x = theano.tensor.scalar(dtype="bool")
y = theano.tensor.bscalar()
x = tt.scalar(dtype="bool")
y = tt.bscalar()
z = x * y
dx, dy = theano.grad(z, [x, y])
......@@ -1279,7 +1272,7 @@ class TestElemwise(unittest_tools.InferShapeTester):
# Elemwise.perform used to compute the product
# of input shapes to check if there was a zero in them,
# it overflowed in this case.
a, b, c, d, e, f = tensor.vectors("abcdef")
a, b, c, d, e, f = tt.vectors("abcdef")
s = a + b + c + d + e + f
g = theano.function(
[a, b, c, d, e, f], s, mode=theano.compile.Mode(linker="py")
......@@ -1315,7 +1308,7 @@ def test_clip_grad():
# test the gradient of clip
def func(x, y, z):
return theano.tensor.clip(x, y, z)
return tt.clip(x, y, z)
# use an x value less than y, an x value between y and z, and an x value
# greater than z
......@@ -1323,35 +1316,38 @@ def test_clip_grad():
def test_grad_useless_sum():
# Test absence of useless sum.
#
# When an operation (such as T.mul) is done on a broadcastable vector and
# a matrix, the gradient in backward path is computed for the broadcasted
# vector. So a sum reverts the broadcasted vector to a vector. In the case
# of operations on two broadcastable vectors, the sum should not be generated.
#
# This test checks whether there is a useless sum in the gradient
# computations.
"""
Test absence of useless sum.
When an operation (such as `theano.tensor.mul`) is done on a broadcastable
vector and a matrix, the gradient in backward path is computed for the
broadcasted vector. So a sum reverts the broadcasted vector to a vector. In
the case of operations on two broadcastable vectors, the sum should not be
generated.
This test checks whether there is a useless sum in the gradient
computations.
"""
mode = theano.compile.get_default_mode().including("canonicalize")
mode.check_isfinite = False
x = TensorType(theano.config.floatX, (True,))("x")
l = tensor.log(1.0 - tensor.nnet.sigmoid(x))[0]
g = tensor.grad(l, x)
l = tt.log(1.0 - sigmoid(x))[0]
g = tt.grad(l, x)
nodes = theano.gof.graph.ops([x], [g])
f = theano.function([x], g, mode=mode)
test_values = [-100, -1, 0, 1, 100]
outputs = []
old_values_eq_approx = staticmethod(TensorType.values_eq_approx)
TensorType.values_eq_approx = staticmethod(tensor.type.values_eq_approx_remove_nan)
TensorType.values_eq_approx = staticmethod(values_eq_approx_remove_nan)
try:
for test_value in test_values:
outputs.append(f(np.array([test_value]).astype("float32")))
finally:
TensorType.values_eq_approx = old_values_eq_approx
assert not any([isinstance(node.op, theano.tensor.elemwise.Sum) for node in nodes])
assert not any([isinstance(node.op, Sum) for node in nodes])
assert np.allclose(
outputs, [[-3.72007598e-44], [-0.26894142], [-0.5], [-0.73105858], [-1.0]]
)
......@@ -1360,22 +1356,21 @@ def test_grad_useless_sum():
def test_elemwise_grad_broadcast():
# This crashed in the past.
x = tensor.tensor(dtype="float32", broadcastable=(True, False, False, False))
y = tensor.tensor(dtype="float32", broadcastable=(True, True, False, False))
x = tt.tensor(dtype="float32", broadcastable=(True, False, False, False))
y = tt.tensor(dtype="float32", broadcastable=(True, True, False, False))
theano.grad(theano.tensor.tanh(x).sum(), x)
theano.grad(theano.tensor.tanh(x + y).sum(), y)
theano.grad(theano.tensor.tanh(x + y).sum(), [x, y])
theano.grad(tt.tanh(x).sum(), x)
theano.grad(tt.tanh(x + y).sum(), y)
theano.grad(tt.tanh(x + y).sum(), [x, y])
def test_clip_grad_int():
# test that integers don't crash clip gradient
x = tensor.iscalar()
y = tensor.iscalar()
z = tensor.iscalar()
c = tensor.clip(x, y, z)
tensor.grad(c, [x, y, z])
x = tt.iscalar()
y = tt.iscalar()
z = tt.iscalar()
c = tt.clip(x, y, z)
tt.grad(c, [x, y, z])
def test_not_implemented_elemwise_grad():
......@@ -1394,20 +1389,10 @@ def test_not_implemented_elemwise_grad():
dy_dx = n
return [theano.gradient.grad_not_implemented(self, 0, n), gz * dy_dx]
test_op = tensor.Elemwise(TestOp())
x = tensor.scalar()
# The call to `grad` used to crash.
tensor.grad(test_op(2, x), x)
# Verify that trying to use the not implemented gradient fails.
try:
tensor.grad(test_op(x, 2), x)
assert False
except theano.gradient.NullTypeGradError:
pass
test_op = tt.Elemwise(TestOp())
x = tt.scalar()
assert isinstance(tt.grad(test_op(2, x), x), gof.graph.Variable)
if __name__ == "__main__":
t = TestElemwise("setUp")
t.setup_method()
t.test_infer_shape()
# Verify that trying to use the not implemented gradient fails.
with pytest.raises(theano.gradient.NullTypeGradError):
tt.grad(test_op(x, 2), x)
tests/tensor/test_fft.py
浏览文件 @ ef279e19
import numpy as np
import pytest
import theano
import theano.tensor as tt
from theano import tensor as T
from theano.tensor import fft
from tests import unittest_tools as utt
......@@ -31,7 +32,7 @@ class TestFFT:
def test_1Drfft(self):
inputs_val = np.random.random((1, N)).astype(theano.config.floatX)
x = T.matrix("x")
x = tt.matrix("x")
rfft = fft.rfft(x)
f_rfft = theano.function([x], rfft)
res_rfft = f_rfft(inputs_val)
......
tests/tensor/test_gc.py
浏览文件 @ ef279e19
import numpy as np
import time
import six.moves.cPickle as pickle
import numpy as np
import theano
from theano import tensor as T
import time
import theano.tensor as tt
def test_no_reuse():
x = T.lvector()
y = T.lvector()
x = tt.lvector()
y = tt.lvector()
f = theano.function([x, y], x + y)
# provide both inputs in the first call
......@@ -22,7 +24,7 @@ def test_no_reuse():
def test_gc_never_pickles_temporaries():
x = T.dvector()
x = tt.dvector()
r = x
for i in range(2): # TODO: 30 causes like LONG compilation due to MERGE
......@@ -105,7 +107,7 @@ def test_merge_opt_runtime():
#
# Ironically, there is actually no merging to do in this graph.
x = T.dvector()
x = tt.dvector()
r = x
for i in range(50):
r = r + r / 10
......
tests/tensor/test_merge.py
浏览文件 @ ef279e19
import numpy as np
import theano.tensor.basic as tt
from theano.gof.type import Type
from theano.gof.graph import Variable, Apply
from theano.gof.op import Op
from theano.gof.opt import MergeOptimizer
from theano.gof.fg import FunctionGraph as Env
import theano.tensor.basic as T
from theano.gof.fg import FunctionGraph
def as_variable(x):
def is_variable(x):
if not isinstance(x, Variable):
raise TypeError("not a Variable", x)
return x
......@@ -30,7 +31,7 @@ class MyOp(Op):
self.x = x
def make_node(self, *inputs):
inputs = list(map(as_variable, inputs))
inputs = list(map(is_variable, inputs))
for input in inputs:
if not isinstance(input.type, MyType):
raise Exception("Error 1")
......@@ -65,9 +66,9 @@ def test_merge_with_weird_eq():
# numpy arrays don't compare equal like other python objects
# SCALAR CASE
x = T.constant(np.asarray(1), name="x")
y = T.constant(np.asarray(1), name="y")
g = Env([x, y], [x + y])
x = tt.constant(np.asarray(1), name="x")
y = tt.constant(np.asarray(1), name="y")
g = FunctionGraph([x, y], [x + y])
MergeOptimizer().optimize(g)
assert len(g.apply_nodes) == 1
......@@ -77,9 +78,9 @@ def test_merge_with_weird_eq():
# NONSCALAR CASE
# This was created to test TensorConstantSignature
x = T.constant(np.ones(5), name="x")
y = T.constant(np.ones(5), name="y")
g = Env([x, y], [x + y])
x = tt.constant(np.ones(5), name="x")
y = tt.constant(np.ones(5), name="y")
g = FunctionGraph([x, y], [x + y])
MergeOptimizer().optimize(g)
assert len(g.apply_nodes) == 1
......
tests/tensor/test_mlp.py
浏览文件 @ ef279e19
"""
This is a minimized version of the mlp.py in the tutorial. We removed stuff that make this mlp don't work.
But this test a bug that we saw. This bug made the Shape_i object not being lifted, that caused the CrossentropySoftmax... op not being inserted.
This is a minimized version of the mlp.py in the tutorial. We removed stuff
that make this mlp don't work. But this test a bug that we saw. This bug made
the Shape_i object not being lifted, that caused the CrossentropySoftmax... op
not being inserted.
"""
__docformat__ = "restructedtext en"
from collections import OrderedDict
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
def gen_data():
......@@ -49,7 +49,7 @@ def gen_data():
# floats it doesn't make sense) therefore instead of returning
# ``shared_y`` we will have to cast it to int. This little hack
# lets ous get around this issue
return shared_x, T.cast(shared_y, "int32")
return shared_x, tt.cast(shared_y, "int32")
test_set_x, test_set_y = shared_dataset(test_set)
valid_set_x, valid_set_y = shared_dataset(valid_set)
......@@ -96,11 +96,11 @@ class LogisticRegression(object):
)
# compute vector of class-membership probabilities in symbolic form
self.p_y_given_x = T.nnet.softmax(T.dot(input, self.W))
self.p_y_given_x = tt.nnet.softmax(tt.dot(input, self.W))
# compute prediction as class whose probability is maximal in
# symbolic form
self.y_pred = T.argmax(self.p_y_given_x, axis=1)
self.y_pred = tt.argmax(self.p_y_given_x, axis=1)
# parameters of the model
self.params = [self.W]
......@@ -128,11 +128,11 @@ class LogisticRegression(object):
# LP[T.arange(y.shape[0]),y] is a vector v containing [LP[0,y[0]], LP[1,y[1]], LP[2,y[2]], ..., LP[n-1,y[n-1]]]
# and T.mean(LP[T.arange(y.shape[0]),y]) is the mean (across minibatch examples) of the elements in v,
# i.e., the mean log-likelihood across the minibatch.
return T.log(self.p_y_given_x[T.arange(y.shape[0]), y])
return tt.log(self.p_y_given_x[tt.arange(y.shape[0]), y])
class HiddenLayer(object):
def __init__(self, rng, input, n_in, n_out, activation=T.tanh, name_prefix=""):
def __init__(self, rng, input, n_in, n_out, activation=tt.tanh, name_prefix=""):
"""
Typical hidden layer of a MLP: units are fully-connected and have
sigmoidal activation function. Weight matrix W is of shape (n_in,n_out)
......@@ -174,7 +174,7 @@ class HiddenLayer(object):
)
self.W = theano.shared(value=W_values, name=name_prefix + "W")
self.output = T.dot(input, self.W)
self.output = tt.dot(input, self.W)
# parameters of the model
self.params = [self.W]
......@@ -222,7 +222,7 @@ class MLP(object):
input=input,
n_in=n_in,
n_out=n_hidden,
activation=T.tanh,
activation=tt.tanh,
name_prefix="hid_",
)
......@@ -284,9 +284,9 @@ def test_mlp():
# print '... building the model'
# allocate symbolic variables for the data
index = T.lscalar() # index to a [mini]batch
x = T.matrix("x") # the data is presented as rasterized images
y = T.ivector("y") # the labels are presented as 1D vector of
index = tt.lscalar() # index to a [mini]batch
x = tt.matrix("x") # the data is presented as rasterized images
y = tt.ivector("y") # the labels are presented as 1D vector of
# [int] labels
rng = np.random.RandomState(1234)
......@@ -303,7 +303,7 @@ def test_mlp():
# the resulting gradients will be stored in a list gparams
gparams = []
for param in classifier.params:
gparam = T.grad(cost, param)
gparam = tt.grad(cost, param)
gparams.append(gparam)
# Some optimizations needed are tagged with 'fast_run'
......@@ -312,7 +312,7 @@ def test_mlp():
updates2 = OrderedDict()
updates2[classifier.hiddenLayer.params[0]] = T.grad(
updates2[classifier.hiddenLayer.params[0]] = tt.grad(
cost, classifier.hiddenLayer.params[0]
)
train_model = theano.function(
......@@ -328,7 +328,7 @@ def test_mlp():
# theano.printing.debugprint(train_model, print_type=True)
assert any(
[
isinstance(i.op, T.nnet.CrossentropySoftmax1HotWithBiasDx)
isinstance(i.op, tt.nnet.CrossentropySoftmax1HotWithBiasDx)
for i in train_model.maker.fgraph.toposort()
]
)
......@@ -348,11 +348,7 @@ def test_mlp():
# theano.printing.debugprint(train_model, print_type=True)
assert any(
[
isinstance(i.op, T.nnet.CrossentropySoftmax1HotWithBiasDx)
isinstance(i.op, tt.nnet.CrossentropySoftmax1HotWithBiasDx)
for i in train_model.maker.fgraph.toposort()
]
)
if __name__ == "__main__":
test_mlp()
tests/tensor/test_opt.py
浏览文件 @ ef279e19
This source diff could not be displayed because it is too large. You can view the blob instead.
tests/tensor/test_subtensor.py
浏览文件 @ ef279e19
......@@ -7,7 +7,7 @@ import numpy as np
import theano
import theano.scalar as scal
import theano.tensor as tensor
import theano.tensor as tt
from numpy.testing import assert_array_equal
......@@ -74,7 +74,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
"""
def setup_method(self):
self.shared = tensor._shared
self.shared = _shared
self.dtype = theano.config.floatX
mode = theano.compile.mode.get_default_mode()
self.mode = mode.including("local_useless_subtensor")
......@@ -312,7 +312,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
(lambda: n[: (2 ** 63)])()
def test_list_slice(self):
x = tensor.arange(100).reshape((5, 5, 4))
x = tt.arange(100).reshape((5, 5, 4))
res = x[[slice(1, -1)] * x.ndim].eval()
x = np.arange(100).reshape((5, 5, 4))
np.allclose(res, x[[slice(1, -1)] * x.ndim])
......@@ -567,7 +567,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
n = self.shared(data)
z = scal.constant(subi).astype("int32")
t = n[z:, z]
gn = tensor.grad(tensor.sum(tensor.exp(t)), n)
gn = tt.grad(tt.sum(tt.exp(t)), n)
f = inplace_func([], gn, mode=self.mode)
topo = f.maker.fgraph.toposort()
......@@ -598,7 +598,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
mv = np.asarray(rand(*m_shape), dtype=self.dtype)
t = op(n[:z, :z], m)
gn, gm = tensor.grad(tensor.sum(t), [n, m])
gn, gm = tt.grad(tt.sum(t), [n, m])
utt.verify_grad(lambda m: op(n[:z, :z], m), [mv], mode=self.mode)
utt.verify_grad(lambda nn: op(nn[:z, :z], mv), [data], mode=self.mode)
......@@ -606,7 +606,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
data = np.asarray(rand(2, 3), dtype=self.dtype)
n = self.shared(data)
t = n[1, 0]
gn = tensor.grad(tensor.sum(tensor.exp(t)), n)
gn = tt.grad(tt.sum(tt.exp(t)), n)
f = self.function([], gn)
topo = f.maker.fgraph.toposort()
topo_ = [node for node in topo if not isinstance(node.op, DeepCopyOp)]
......@@ -632,7 +632,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
# optimized for that case.
(rand(4, 4, 2, 3), [3, 3, 1, 1, 2, 2, 0, 0, -1, -2, -3, -4]),
# Test with TensorConstant index.
(rand(4, 2, 3), tensor.constant([3, 3, 1, 1, 2, 2, 0, 0])),
(rand(4, 2, 3), tt.constant([3, 3, 1, 1, 2, 2, 0, 0])),
]:
data = np.asarray(data, dtype=self.dtype)
n = self.shared(data)
......@@ -717,7 +717,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
def test_adv_sub1_broadcast(self):
v = np.arange(3, dtype=self.dtype).reshape((1, 3))
n = self.shared(v * 5, broadcastable=(True, False))
idx = tensor.lvector()
idx = lvector()
t = n[idx]
assert isinstance(t.owner.op, AdvancedSubtensor1)
......@@ -779,13 +779,13 @@ class TestSubtensor(utt.OptimizationTestMixin):
# test set_subtensor broadcast
self.dtype = "float32"
x = tensor.tensor4("x", dtype=self.dtype)
x = tt.tensor4("x", dtype=self.dtype)
indexes = theano.shared(np.int32([1, 2, 3, 4]))
W = self.shared(np.random.random((10, 10, 3, 3)).astype(self.dtype))
h = x + W
h = tensor.set_subtensor(h[indexes], h[indexes])
g = tensor.grad(h.sum(), W)
h = tt.set_subtensor(h[indexes], h[indexes])
g = tt.grad(h.sum(), W)
N = 2
if (
theano.config.mode == "FAST_COMPILE"
......@@ -800,7 +800,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
# The idx can be a broadcastable vector.
ones = np.ones((4, 3), dtype=self.dtype)
n = self.shared(ones * 5)
idx = tensor.TensorType(dtype="int64", broadcastable=(True,))()
idx = tt.TensorType(dtype="int64", broadcastable=(True,))()
assert idx.type.broadcastable == (True,)
t = n[idx]
assert isinstance(t.owner.op, AdvancedSubtensor1)
......@@ -849,9 +849,9 @@ class TestSubtensor(utt.OptimizationTestMixin):
v_data = np.array(np.arange(5), dtype=self.dtype)
t_data = self.shared(v_data)
start = tensor.iscalar("b")
stop = tensor.iscalar("e")
step = tensor.iscalar("s")
start = iscalar("b")
stop = iscalar("e")
step = iscalar("s")
f = self.function(
[start, stop, step],
t_data[start:stop:step].shape,
......@@ -866,18 +866,18 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(f(start, stop, step) == v_data[start:stop:step].shape)
def test_slice_canonical_form_0(self):
start = tensor.iscalar("b")
stop = tensor.iscalar("e")
step = tensor.iscalar("s")
length = tensor.iscalar("l")
start = iscalar("b")
stop = iscalar("e")
step = iscalar("s")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(start, stop, step), length)
f = self.function(
[start, stop, step, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -895,17 +895,17 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(t_out.shape == v_out.shape)
def test_slice_canonical_form_1(self):
stop = tensor.iscalar("e")
step = tensor.iscalar("s")
length = tensor.iscalar("l")
stop = iscalar("e")
step = iscalar("s")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(None, stop, step), length)
f = self.function(
[stop, step, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -922,17 +922,17 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(t_out.shape == v_out.shape)
def test_slice_canonical_form_2(self):
start = tensor.iscalar("b")
step = tensor.iscalar("s")
length = tensor.iscalar("l")
start = iscalar("b")
step = iscalar("s")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(start, None, step), length)
f = self.function(
[start, step, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -949,17 +949,17 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(t_out.shape == v_out.shape)
def test_slice_canonical_form_3(self):
start = tensor.iscalar("b")
stop = tensor.iscalar("e")
length = tensor.iscalar("l")
start = iscalar("b")
stop = iscalar("e")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(start, stop, None), length)
f = self.function(
[start, stop, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -976,16 +976,16 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(t_out.shape == v_out.shape)
def test_slice_canonical_form_4(self):
step = tensor.iscalar("s")
length = tensor.iscalar("l")
step = iscalar("s")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(None, None, step), length)
f = self.function(
[step, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -1001,16 +1001,16 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(t_out.shape == v_out.shape)
def test_slice_canonical_form_5(self):
start = tensor.iscalar("b")
length = tensor.iscalar("l")
start = iscalar("b")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(start, None, None), length)
f = self.function(
[start, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -1026,16 +1026,16 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.all(t_out.shape == v_out.shape)
def test_slice_canonical_form_6(self):
stop = tensor.iscalar("e")
length = tensor.iscalar("l")
stop = iscalar("e")
length = iscalar("l")
cnf = get_canonical_form_slice(slice(None, stop, None), length)
f = self.function(
[stop, length],
[
tensor.as_tensor_variable(cnf[0].start),
tensor.as_tensor_variable(cnf[0].stop),
tensor.as_tensor_variable(cnf[0].step),
tensor.as_tensor_variable(cnf[1]),
tt.as_tensor_variable(cnf[0].start),
tt.as_tensor_variable(cnf[0].stop),
tt.as_tensor_variable(cnf[0].step),
tt.as_tensor_variable(cnf[1]),
],
N=0,
op=subtensor_ops,
......@@ -1057,7 +1057,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
# Should stay on the cpu.
idx_ = _shared(np.asarray(idx))
t = n[idx_]
gn = tensor.grad(tensor.sum(tensor.exp(t)), n)
gn = tt.grad(tt.sum(tt.exp(t)), n)
f = self.function([], [gn, gn.shape], op=AdvancedIncSubtensor1)
topo = f.maker.fgraph.toposort()
if not self.fast_compile:
......@@ -1083,13 +1083,13 @@ class TestSubtensor(utt.OptimizationTestMixin):
assert np.allclose(gshape, data.shape)
def fct(t):
return tensor.sum(t[idx_])
return tt.sum(t[idx_])
utt.verify_grad(fct, [data], mode=self.mode)
# Test the grad of the grad (e.i. AdvancedIncSubtensor1.grad)
def fct2(t):
return tensor.grad(tensor.sum(t[idx_]), t)
return tt.grad(tt.sum(t[idx_]), t)
utt.verify_grad(fct2, [data], mode=self.mode)
......@@ -1195,11 +1195,11 @@ class TestSubtensor(utt.OptimizationTestMixin):
# Symbolic variable to be incremented.
# We create a new one every time in order not to
# have duplicated variables in the function's inputs
data_var = tensor.TensorType(
data_var = tt.TensorType(
broadcastable=[False] * data_n_dims, dtype=self.dtype
)()
# Symbolic variable with rows to be incremented.
idx_var = tensor.vector(dtype="int64")
idx_var = vector(dtype="int64")
n_to_inc = rng.randint(data_shape[0])
if (
n_to_inc == 1
......@@ -1218,7 +1218,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
)
idx_num = idx_num.astype("int64")
# Symbolic variable with increment value.
inc_var = tensor.TensorType(
inc_var = tt.TensorType(
broadcastable=[False] * inc_n_dims, dtype=self.dtype
)()
# Trick for the case where `inc_shape` is the same as
......@@ -1302,8 +1302,8 @@ class TestSubtensor(utt.OptimizationTestMixin):
# Test case provided (and bug detected, gh-607) by John Salvatier
m = matrix("m")
gv = np.array([0, 1, 3])
g = tensor.constant(gv)
i = tensor.lvector("i")
g = tt.constant(gv)
i = lvector("i")
# s1 used to fail
s1 = m[gv, i]
......@@ -1437,7 +1437,7 @@ class TestSubtensor(utt.OptimizationTestMixin):
config.warn.inc_set_subtensor1 = orig_warn
def test_take(self):
a = tensor.matrix()
a = matrix()
f = theano.function(
[a], a.take(0, axis=-1), allow_input_downcast=True, mode=self.mode
)
......@@ -1451,12 +1451,12 @@ class TestIncSubtensor1:
def setup_method(self):
self.rng = np.random.RandomState(seed=utt.fetch_seed())
self.s = tensor.iscalar()
self.v = tensor.fvector()
self.m = tensor.dmatrix()
self.t = tensor.ctensor3()
self.s = iscalar()
self.v = tt.fvector()
self.m = dmatrix()
self.t = ctensor3()
self.adv1q = tensor.lvector() # advanced 1d query
self.adv1q = lvector() # advanced 1d query
def test_cant_adv_idx_into_scalar(self):
with pytest.raises(IndexError):
......@@ -1499,7 +1499,7 @@ class TestIncSubtensor1:
(lambda: inc_subtensor(self.v[self.adv1q](fmatrix())))()
def test_matrix_idx(self):
idx = tensor.lmatrix()
idx = lmatrix()
a = self.m[idx]
a2 = inc_subtensor(a, a)
f = theano.function([self.m, idx], a2)
......@@ -1514,9 +1514,9 @@ class TestIncSubtensor1:
utt.assert_allclose(a2val[3], mval[3] * 2)
def test_inc_bcastableidx(self):
idx = tensor.constant([0])
c_inc = tensor.col()
m_inc = tensor.matrix()
idx = tt.constant([0])
c_inc = tt.col()
m_inc = matrix()
out1 = inc_subtensor(self.m[:, idx], c_inc)
out2 = inc_subtensor(self.m[:, idx], m_inc)
......@@ -1532,7 +1532,7 @@ class TestAdvancedSubtensor:
"""Test inc_subtensor and set_subtensor."""
def setup_method(self):
self.shared = tensor._shared
self.shared = _shared
self.dtype = theano.config.floatX
self.mode = theano.compile.mode.get_default_mode()
......@@ -1552,10 +1552,10 @@ class TestAdvancedSubtensor:
def check(idx, y_val, x_val, true):
x = self.shared(x_val, name="x")
y = tensor.tensor(
y = tt.tensor(
dtype="float32", broadcastable=(False,) * len(y_val.shape), name="y"
)
sym_idx = [tensor.as_tensor_variable(ix) for ix in idx]
sym_idx = [tt.as_tensor_variable(ix) for ix in idx]
expr = advanced_inc_subtensor(x, y, *sym_idx)
f = theano.function([y], expr, mode=self.mode)
rval = f(y_val)
......@@ -1628,7 +1628,7 @@ class TestAdvancedSubtensor:
# optimized for that case.
(rand(4, 4, 2, 3), [3, 3, 1, 1, 2, 2, 0, 0, -1, -2, -3, -4]),
# Test with TensorConstant index.
(rand(2, 4, 3), tensor.constant([3, 3, 1, 1, 2, 2, 0, 0])),
(rand(2, 4, 3), tt.constant([3, 3, 1, 1, 2, 2, 0, 0])),
]:
data = np.asarray(data, dtype=self.dtype)
n = self.shared(data)
......@@ -1704,7 +1704,7 @@ class TestAdvancedSubtensor:
subt = self.m[self.ix1, self.ix12]
a = inc_subtensor(subt, subt)
typ = tensor.TensorType(self.m.type.dtype, self.ix2.type.broadcastable)
typ = tt.TensorType(self.m.type.dtype, self.ix2.type.broadcastable)
assert a.type == typ, (a.type, typ)
f = theano.function(
[self.m, self.ix1, self.ix12], a, allow_input_downcast=True, mode=self.mode
......@@ -1717,7 +1717,7 @@ class TestAdvancedSubtensor:
def test_inc_adv_subtensor_with_broadcasting(self):
inc = dscalar()
a = inc_subtensor(self.m[self.ix1, self.ix12], inc)
g_inc = tensor.grad(a.sum(), inc)
g_inc = tt.grad(a.sum(), inc)
assert a.type == self.m.type, (a.type, self.m.type)
f = theano.function(
......@@ -1737,7 +1737,7 @@ class TestAdvancedSubtensor:
def test_inc_adv_subtensor1_with_broadcasting(self):
inc = dscalar()
a = inc_subtensor(self.m[self.ix1], inc)
g_inc = tensor.grad(a.sum(), inc)
g_inc = tt.grad(a.sum(), inc)
assert a.type == self.m.type, (a.type, self.m.type)
f = theano.function(
......@@ -1782,8 +1782,8 @@ class TestAdvancedSubtensor:
rng = np.random.RandomState(utt.fetch_seed())
a = rng.uniform(size=(3, 3))
b = theano.shared(a)
i = tensor.iscalar()
j = tensor.iscalar()
i = iscalar()
j = iscalar()
z = b[[i, j], :]
f1 = theano.function([i, j], z, mode=self.mode)
cmd = f1(0, 1) == a[[0, 1], :]
......@@ -1791,7 +1791,7 @@ class TestAdvancedSubtensor:
aa = rng.uniform(size=(4, 2, 3))
bb = theano.shared(aa)
k = tensor.iscalar()
k = iscalar()
z = bb[[i, j, k], :, i:k]
f2 = theano.function([i, j, k], z, mode=self.mode)
cmd = f2(0, 1, 2) == aa[[0, 1, 2], :, 0:2]
......@@ -1799,7 +1799,7 @@ class TestAdvancedSubtensor:
def test_adv_sub_3d(self):
# Reported in https://github.com/Theano/Theano/issues/5674
X = tensor.tensor3("X")
X = tt.tensor3("X")
xx = np.zeros((3, 2, 2), config.floatX)
for i in range(3):
......@@ -1821,7 +1821,7 @@ class TestAdvancedSubtensor:
def test_adv_sub_slice(self):
# Reported in https://github.com/Theano/Theano/issues/5898
var = self.shared(np.zeros([3, 3], dtype=config.floatX))
slc = tensor.slicetype()
slc = tt.slicetype()
f = theano.function([slc], var[slc], mode=self.mode)
s = slice(1, 3)
f(s)
......@@ -1833,7 +1833,7 @@ class TestAdvancedSubtensor:
var = self.shared(var_v)
idx1_v = rng.randint(0, 61, size=(5, 4)).astype("int32")
idx1 = self.shared(idx1_v)
idx2 = tensor.arange(4)
idx2 = tt.arange(4)
out = var[:, idx1, idx2]
f = theano.function([], out, mode=self.mode)
out_v = f()
......@@ -1845,8 +1845,8 @@ class TestAdvancedSubtensor:
def test_grad(self):
ones = np.ones((1, 3), dtype=self.dtype)
n = self.shared(ones * 5, broadcastable=(True, False))
idx = tensor.lvector()
idx2 = tensor.lvector()
idx = lvector()
idx2 = lvector()
t = n[idx, idx2]
assert isinstance(t.owner.op, AdvancedSubtensor)
......@@ -1883,7 +1883,7 @@ class TestAdvancedSubtensor:
# Test boolean gradients
def fun(x, y):
return advanced_inc_subtensor(
x, y, tensor.as_tensor(np.array([[True, False], [False, True]]))
x, y, tt.as_tensor(np.array([[True, False], [False, True]]))
)
utt.verify_grad(
......@@ -1897,7 +1897,7 @@ class TestAdvancedSubtensor:
def fun(x, y):
return advanced_set_subtensor(
x, y, tensor.as_tensor(np.array([[True, False], [False, True]]))
x, y, tt.as_tensor(np.array([[True, False], [False, True]]))
)
utt.verify_grad(
......@@ -2216,7 +2216,7 @@ class TestInferShape(utt.InferShapeTester):
check_topo=False,
)
abs_res = n[~tensor.isinf(n)]
abs_res = n[~tt.isinf(n)]
assert abs_res.broadcastable == (False,)
......@@ -2239,7 +2239,7 @@ def test_indexed_result_shape():
if isinstance(x, (slice, type(None))):
return x
else:
return tensor.as_tensor(x)
return tt.as_tensor(x)
def bcast_shape_tuple(x):
if not hasattr(x, "shape"):
......@@ -2250,14 +2250,14 @@ def test_indexed_result_shape():
def compare_index_shapes(test_array, test_idx):
res = indexed_result_shape(
tensor.as_tensor(test_array).shape, [idx_as_tensor(i) for i in test_idx]
tt.as_tensor(test_array).shape, [idx_as_tensor(i) for i in test_idx]
)
exp_res = test_array[test_idx].shape
assert np.array_equal(tuple(get_test_value(r) for r in res), exp_res)
# Test shape-only version
res = indexed_result_shape(
tensor.as_tensor(test_array).shape,
tt.as_tensor(test_array).shape,
[bcast_shape_tuple(idx_as_tensor(i)) for i in test_idx],
indices_are_shapes=True,
)
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论