Skip to content

P
pytensor
提交 e103e7fb 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.nnet

上级 daacd069
隐藏空白字符变更
内嵌 并排
正在显示 包含 430 行增加 和 436 行删除
tests/tensor/nnet/speed_test_conv.py
浏览文件 @ e103e7fb
......@@ -2,7 +2,7 @@ import time
import numpy as N
import theano.tensor as T
import theano.tensor as tt
from theano import function, Mode
from theano.tensor.nnet.conv import ConvOp
......@@ -39,7 +39,7 @@ def flip(kern, kshp):
global_rng = N.random.RandomState(3423489)
dmatrix4 = T.TensorType("float64", (False, False, False, False))
dmatrix4 = tt.TensorType("float64", (False, False, False, False))
def exec_multilayer_conv_nnet_old(
......@@ -51,7 +51,7 @@ def exec_multilayer_conv_nnet_old(
nkerns,
unroll_batch=0,
unroll_kern=0,
img=T.dmatrix(),
img=tt.dmatrix(),
validate=True,
conv_op_py=False,
do_print=True,
......@@ -64,7 +64,7 @@ def exec_multilayer_conv_nnet_old(
# build actual input images
imgval = global_rng.rand(bsize, imshp[0], imshp[1], imshp[2])
a = T.dmatrix()
a = tt.dmatrix()
kerns = [a for i in nkerns]
inputs4 = dmatrix4()
kerns4 = dmatrix4()
......@@ -180,7 +180,7 @@ def exec_multilayer_conv_nnet(
nkerns,
unroll_batch=0,
unroll_kern=0,
img=T.dmatrix(),
img=tt.dmatrix(),
do_print=True,
repeat=1,
unroll_patch=False,
......@@ -191,7 +191,7 @@ def exec_multilayer_conv_nnet(
# build actual input images
imgval = global_rng.rand(bsize, imshp[0], imshp[1], imshp[2])
a = T.dmatrix()
a = tt.dmatrix()
kerns = [a for i in nkerns]
inputs4 = dmatrix4()
kerns4 = dmatrix4()
......@@ -277,7 +277,7 @@ def speed_multilayer_conv():
] # (1,1)]#(2,2) bugged
convmodes = ["valid", "full"]
# do_convolve2 = False
a = T.dmatrix()
a = tt.dmatrix()
kerns = [a for i in nkerns]
assert len(kshps) == len(nkerns) == len(kerns)
......
tests/tensor/nnet/test_bn.py
浏览文件 @ e103e7fb
......@@ -3,7 +3,7 @@ import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
from collections import OrderedDict
......@@ -29,11 +29,11 @@ def test_BNComposite():
M = 1 + np.random.random([20]).astype("float32")
V = 1 + np.random.random([20]).astype("float32")
x = theano.tensor.matrix("x")
b = theano.tensor.vector("b")
g = theano.tensor.vector("g")
m = theano.tensor.vector("m")
v = theano.tensor.vector("v")
x = tt.matrix("x")
b = tt.vector("b")
g = tt.vector("g")
m = tt.vector("m")
v = tt.vector("v")
x.tag.test_value = np.random.rand(2, 2).astype(theano.config.floatX)
b.tag.test_value = np.random.rand(2).astype(theano.config.floatX)
......@@ -65,11 +65,11 @@ def test_batch_normalization():
M = 1 + np.random.random([20]).astype("float32")
V = 1 + np.random.random([20]).astype("float32")
x = theano.tensor.matrix("x")
b = theano.tensor.vector("b")
g = theano.tensor.vector("g")
m = theano.tensor.vector("m")
v = theano.tensor.vector("v")
x = tt.matrix("x")
b = tt.vector("b")
g = tt.vector("g")
m = tt.vector("m")
v = tt.vector("v")
bn_ref_op = bn_ref(x, g, b, m, v)
f_ref = theano.function([x, g, b, m, v], [bn_ref_op])
......@@ -123,11 +123,11 @@ def test_bn_feature_maps():
M = 1 + np.random.random([3]).astype("float32")
V = 1 + np.random.random([3]).astype("float32")
x = theano.tensor.tensor4("x")
b = theano.tensor.vector("b")
g = theano.tensor.vector("g")
m = theano.tensor.vector("m")
v = theano.tensor.vector("v")
x = tt.tensor4("x")
b = tt.vector("b")
g = tt.vector("g")
m = tt.vector("m")
v = tt.vector("v")
bn_ref_op = bn_ref(
x,
......@@ -170,7 +170,7 @@ def test_batch_normalization_train():
utt.seed_rng()
for axes in ("per-activation", "spatial", (1, 2, 3, 4)):
for vartype in (T.tensor5, T.tensor3, T.vector):
for vartype in (tt.tensor5, tt.tensor3, tt.vector):
x, scale, bias, running_mean, running_var = (
vartype(n)
for n in ("x", "scale", "bias", "running_mean", "running_var")
......@@ -211,11 +211,11 @@ def test_batch_normalization_train():
axes2 = axes
x_mean2 = x.mean(axis=axes2, keepdims=True)
x_var2 = x.var(axis=axes2, keepdims=True)
x_invstd2 = T.inv(T.sqrt(x_var2 + eps))
scale2 = T.addbroadcast(scale, *axes2)
bias2 = T.addbroadcast(bias, *axes2)
x_invstd2 = tt.inv(tt.sqrt(x_var2 + eps))
scale2 = tt.addbroadcast(scale, *axes2)
bias2 = tt.addbroadcast(bias, *axes2)
out2 = (x - x_mean2) * (scale2 * x_invstd2) + bias2
m = T.cast(T.prod(x.shape) / T.prod(scale.shape), theano.config.floatX)
m = tt.cast(tt.prod(x.shape) / tt.prod(scale.shape), theano.config.floatX)
out_running_mean2 = (
running_mean * (1 - running_average_factor)
+ x_mean2 * running_average_factor
......@@ -226,14 +226,14 @@ def test_batch_normalization_train():
)
# backward pass
dy = vartype("dy")
grads = T.grad(None, wrt=[x, scale, bias], known_grads={out: dy})
grads = tt.grad(None, wrt=[x, scale, bias], known_grads={out: dy})
# reference backward pass
grads2 = T.grad(None, wrt=[x, scale, bias], known_grads={out2: dy})
grads2 = tt.grad(None, wrt=[x, scale, bias], known_grads={out2: dy})
# second-order backward pass
dx = vartype("dinputs")
dscale = vartype("dscale")
dbias = vartype("dbias")
grad_grads = T.grad(
grad_grads = tt.grad(
None,
wrt=[x, dy, scale],
known_grads=OrderedDict(
......@@ -252,7 +252,7 @@ def test_batch_normalization_train():
return_disconnected="zero",
)
# reference second-order backward pass
grad_grads2 = T.grad(
grad_grads2 = tt.grad(
None,
wrt=[x, dy, scale],
known_grads=OrderedDict(
......@@ -354,7 +354,7 @@ def test_batch_normalization_train_grad_grad():
utt.seed_rng()
for axes in ("per-activation", "spatial", (1, 2, 3, 4)):
for vartype in (T.tensor5, T.tensor4, T.tensor3, T.matrix, T.vector):
for vartype in (tt.tensor5, tt.tensor4, tt.tensor3, tt.matrix, tt.vector):
# run these experiments with float64 for sufficient numerical stability
x, dy, scale, x_mean, x_invstd = (
vartype(n, dtype="float64")
......@@ -425,10 +425,10 @@ def test_batch_normalization_train_without_running_averages():
utt.seed_rng()
x, scale, bias, dy = (
T.tensor4("x"),
T.tensor4("scale"),
T.tensor4("bias"),
T.tensor4("dy"),
tt.tensor4("x"),
tt.tensor4("scale"),
tt.tensor4("bias"),
tt.tensor4("dy"),
)
data_shape = (5, 10, 30, 25)
param_shape = (1, 10, 30, 25)
......@@ -438,7 +438,7 @@ def test_batch_normalization_train_without_running_averages():
x, scale, bias, "per-activation"
)
# backward pass
grads = T.grad(None, wrt=[x, scale, bias], known_grads={out: dy})
grads = tt.grad(None, wrt=[x, scale, bias], known_grads={out: dy})
# compile
f = theano.function([x, scale, bias, dy], [out, x_mean, x_invstd] + grads)
# check if the abstract Ops have been replaced
......@@ -465,7 +465,7 @@ def test_batch_normalization_train_without_running_averages():
def test_batch_normalization_train_broadcast():
for axes in ("per-activation", "spatial", (1, 2, 3, 4)):
for vartype in (T.tensor5, T.tensor4, T.tensor3, T.matrix, T.vector):
for vartype in (tt.tensor5, tt.tensor4, tt.tensor3, tt.matrix, tt.vector):
x = vartype("x")
ndim = x.ndim
eps = 5e-3 # some non-standard value to test if it's used
......@@ -492,7 +492,7 @@ def test_batch_normalization_train_broadcast():
params_dimshuffle[axis] = i
# construct non-broadcasted parameter variables
param_type = T.TensorType(x.dtype, (False,) * len(non_bc_axes))
param_type = tt.TensorType(x.dtype, (False,) * len(non_bc_axes))
scale, bias, running_mean, running_var = (
param_type(n) for n in ("scale", "bias", "running_mean", "running_var")
)
......@@ -545,7 +545,7 @@ def test_batch_normalization_train_broadcast():
# compile to compute all differences
f = theano.function(
[x, scale, bias, running_mean, running_var], T.sum(sum(results))
[x, scale, bias, running_mean, running_var], tt.sum(sum(results))
)
# the paired ops are exactly the same, so the optimizer should have
......@@ -570,7 +570,7 @@ def test_batch_normalization_train_broadcast():
@pytest.mark.slow
def test_batch_normalization_test():
for axes in ("per-activation", "spatial", (1, 2, 3, 4)):
for vartype in (T.tensor5, T.tensor3, T.vector):
for vartype in (tt.tensor5, tt.tensor3, tt.vector):
x, scale, bias, mean, var = (
vartype(n) for n in ("x", "scale", "bias", "mean", "var")
)
......@@ -593,14 +593,16 @@ def test_batch_normalization_test():
else:
axes2 = axes
scale2, bias2, mean2, var2 = (
T.addbroadcast(t, *axes2) for t in (scale, bias, mean, var)
tt.addbroadcast(t, *axes2) for t in (scale, bias, mean, var)
)
out2 = (x - mean2) * (scale2 / T.sqrt(var2 + eps)) + bias2
out2 = (x - mean2) * (scale2 / tt.sqrt(var2 + eps)) + bias2
# backward pass
dy = vartype("dy")
grads = T.grad(None, wrt=[x, scale, bias, mean, var], known_grads={out: dy})
grads = tt.grad(
None, wrt=[x, scale, bias, mean, var], known_grads={out: dy}
)
# reference backward pass
grads2 = T.grad(
grads2 = tt.grad(
None, wrt=[x, scale, bias, mean, var], known_grads={out2: dy}
)
# compile
......@@ -649,7 +651,7 @@ def test_batch_normalization_test():
def test_batch_normalization_broadcastable():
# check if the broadcastable pattern is preserved by the optimizations
x, dy, scale, bias, mean, var = (
T.scalar(n).dimshuffle(["x"] * 5)
tt.scalar(n).dimshuffle(["x"] * 5)
for n in ("x", "dy", "scale", "bias", "mean", "var")
)
......@@ -659,8 +661,8 @@ def test_batch_normalization_broadcastable():
)
out_test = bn.batch_normalization_test(x, scale, bias, mean, var, "spatial")
# backward pass
grads_train = T.grad(None, wrt=[x, scale, bias], known_grads={out_train: dy})
grads_test = T.grad(None, wrt=[x, scale, bias], known_grads={out_test: dy})
grads_train = tt.grad(None, wrt=[x, scale, bias], known_grads={out_train: dy})
grads_test = tt.grad(None, wrt=[x, scale, bias], known_grads={out_test: dy})
# compile
f = theano.function(
[x, scale, bias, mean, var, dy],
......
tests/tensor/nnet/test_conv.py
浏览文件 @ e103e7fb
......@@ -5,9 +5,9 @@ import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
from theano.tensor.nnet import conv
from theano.tensor.nnet import conv, conv2d
from theano.tensor.basic import _allclose, NotScalarConstantError
from tests import unittest_tools as utt
......@@ -28,9 +28,9 @@ class TestConv2D(utt.InferShapeTester):
conv2d = staticmethod(conv.conv2d)
def setup_method(self):
self.input = T.tensor4("input", dtype=self.dtype)
self.input = tt.tensor4("input", dtype=self.dtype)
self.input.name = "default_V"
self.filters = T.tensor4("filters", dtype=self.dtype)
self.filters = tt.tensor4("filters", dtype=self.dtype)
self.filters.name = "default_filters"
super().setup_method()
......@@ -64,12 +64,12 @@ class TestConv2D(utt.InferShapeTester):
"""
if N_image_shape is None:
N_image_shape = [
T.get_scalar_constant_value(T.as_tensor_variable(x))
tt.get_scalar_constant_value(tt.as_tensor_variable(x))
for x in image_shape
]
if N_filter_shape is None:
N_filter_shape = [
T.get_scalar_constant_value(T.as_tensor_variable(x))
tt.get_scalar_constant_value(tt.as_tensor_variable(x))
for x in filter_shape
]
......@@ -391,7 +391,7 @@ class TestConv2D(utt.InferShapeTester):
def test_shape_Constant_tensor(self):
# Tests convolution where the {image,filter}_shape is a Constant tensor.
as_t = T.as_tensor_variable
as_t = tt.as_tensor_variable
self.validate((as_t(3), as_t(2), as_t(7), as_t(5)), (5, 2, 2, 3), "valid")
self.validate(as_t([3, 2, 7, 5]), (5, 2, 2, 3), "valid")
self.validate(as_t((3, 2, 7, 5)), (5, 2, 2, 3), "valid")
......@@ -563,11 +563,11 @@ class TestConv2D(utt.InferShapeTester):
# Make sure errors are raised when image and kernel are not 4D tensors
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=T.dmatrix())
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=tt.dmatrix())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", filters=T.dvector())
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", filters=tt.dvector())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=T.dtensor3())
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=tt.dtensor3())
def test_gcc_crash(self):
# gcc 4.3.0 20080428 (Red Hat 4.3.0-8)
......@@ -629,8 +629,8 @@ class TestConv2D(utt.InferShapeTester):
r = np.asarray(np.random.rand(*shape), dtype="float64")
return r * 2 - 1
adtens = T.dtensor4()
bdtens = T.dtensor4()
adtens = tt.dtensor4()
bdtens = tt.dtensor4()
aivec_val = [4, 5, 6, 3]
bivec_val = [7, 5, 3, 2]
adtens_val = rand(*aivec_val)
......@@ -737,20 +737,18 @@ class TestConv2D(utt.InferShapeTester):
# code from that ticket.
def test_broadcast_grad():
# rng = numpy.random.RandomState(utt.fetch_seed())
x1 = T.tensor4("x")
x1 = tt.tensor4("x")
# x1_data = rng.randn(1, 1, 300, 300)
sigma = T.scalar("sigma")
sigma = tt.scalar("sigma")
# sigma_data = 20
window_radius = 3
filter_1d = T.arange(-window_radius, window_radius + 1)
filter_1d = tt.arange(-window_radius, window_radius + 1)
filter_1d = filter_1d.astype(theano.config.floatX)
filter_1d = T.exp(-0.5 * filter_1d ** 2 / sigma ** 2)
filter_1d = tt.exp(-0.5 * filter_1d ** 2 / sigma ** 2)
filter_1d = filter_1d / filter_1d.sum()
filter_W = filter_1d.dimshuffle(["x", "x", 0, "x"])
y = theano.tensor.nnet.conv2d(
x1, filter_W, border_mode="full", filter_shape=[1, 1, None, None]
)
y = conv2d(x1, filter_W, border_mode="full", filter_shape=[1, 1, None, None])
theano.grad(y.sum(), sigma)
tests/tensor/nnet/test_corr.py
浏览文件 @ e103e7fb
import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
from six import integer_types
from theano.tensor.nnet import corr, conv
......@@ -29,9 +29,9 @@ class TestCorr2D(utt.InferShapeTester):
dtype = theano.config.floatX
def setup_method(self):
self.input = T.tensor4("input", dtype=self.dtype)
self.input = tt.tensor4("input", dtype=self.dtype)
self.input.name = "default_V"
self.filters = T.tensor4("filters", dtype=self.dtype)
self.filters = tt.tensor4("filters", dtype=self.dtype)
self.filters.name = "default_filters"
# This tests can run even when theano.config.blas.ldflags is empty.
super().setup_method()
......@@ -55,10 +55,10 @@ class TestCorr2D(utt.InferShapeTester):
if not theano.config.cxx:
pytest.skip("Need cxx to test conv2d")
N_image_shape = [
T.get_scalar_constant_value(T.as_tensor_variable(x)) for x in image_shape
tt.get_scalar_constant_value(tt.as_tensor_variable(x)) for x in image_shape
]
N_filter_shape = [
T.get_scalar_constant_value(T.as_tensor_variable(x)) for x in filter_shape
tt.get_scalar_constant_value(tt.as_tensor_variable(x)) for x in filter_shape
]
if input is None:
......@@ -255,7 +255,7 @@ class TestCorr2D(utt.InferShapeTester):
def test_shape_Constant_tensor(self):
# Tests correlation where the {image,filter}_shape is a Constant tensor.
as_t = T.as_tensor_variable
as_t = tt.as_tensor_variable
border_modes = ["valid", "full", "half", (1, 1), (2, 1), (1, 2), (3, 3), 1]
for border_mode in border_modes:
......@@ -290,11 +290,11 @@ class TestCorr2D(utt.InferShapeTester):
# Make sure errors are raised when image and kernel are not 4D tensors
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=T.dmatrix())
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=tt.dmatrix())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", filters=T.dvector())
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", filters=tt.dvector())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=T.dtensor3())
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=tt.dtensor3())
@pytest.mark.skipif(not theano.config.cxx, reason="Need cxx for this test")
def test_dtype_upcast(self):
......@@ -313,8 +313,8 @@ class TestCorr2D(utt.InferShapeTester):
for a_dtype in dtypes:
for b_dtype in dtypes:
c_dtype = theano.scalar.upcast(a_dtype, b_dtype)
a_tens = T.tensor4(dtype=a_dtype)
b_tens = T.tensor4(dtype=b_dtype)
a_tens = tt.tensor4(dtype=a_dtype)
b_tens = tt.tensor4(dtype=b_dtype)
a_tens_val = rand(a_shape, dtype=a_dtype)
b_tens_val = rand(b_shape, dtype=b_dtype)
......@@ -334,8 +334,8 @@ class TestCorr2D(utt.InferShapeTester):
corrMM = corr.CorrMM
adtens = T.dtensor4()
bdtens = T.dtensor4()
adtens = tt.dtensor4()
bdtens = tt.dtensor4()
aivec_vals = [
[4, 5, 6, 3],
[6, 2, 8, 3],
......@@ -385,8 +385,8 @@ class TestCorr2D(utt.InferShapeTester):
corrMM = corr.CorrMM
gradW = corr.CorrMM_gradWeights
adtens = T.dtensor4()
bdtens = T.dtensor4()
adtens = tt.dtensor4()
bdtens = tt.dtensor4()
aivec_vals = [
[1, 5, 6, 3],
[8, 2, 7, 3],
......@@ -441,8 +441,8 @@ class TestCorr2D(utt.InferShapeTester):
corrMM = corr.CorrMM
gradI = corr.CorrMM_gradInputs
adtens = T.dtensor4()
bdtens = T.dtensor4()
adtens = tt.dtensor4()
bdtens = tt.dtensor4()
aivec_vals = [
[1, 5, 6, 3],
[8, 2, 7, 3],
......@@ -510,8 +510,8 @@ class TestGroupCorr2d(TestGroupedConvNoOptim):
groups = 3
bottom = np.random.rand(3, 6, 5, 5).astype(theano.config.floatX)
kern = np.random.rand(9, 2, 3, 3).astype(theano.config.floatX)
bottom_sym = T.tensor4("bottom")
kern_sym = T.tensor4("kern")
bottom_sym = tt.tensor4("bottom")
kern_sym = tt.tensor4("kern")
# grouped convolution graph
conv_group = self.conv(num_groups=groups)(bottom_sym, kern_sym)
......@@ -527,7 +527,7 @@ class TestGroupCorr2d(TestGroupedConvNoOptim):
)
for i in range(groups)
]
concatenated_output = T.concatenate(split_conv_output, axis=1)
concatenated_output = tt.concatenate(split_conv_output, axis=1)
conv_func = theano.function(
[bottom_sym, kern_sym], concatenated_output, mode=self.mode
)
......
tests/tensor/nnet/test_corr3d.py
浏览文件 @ e103e7fb
import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
from six import integer_types
......@@ -23,9 +23,9 @@ class TestCorr3D(utt.InferShapeTester):
dtype = theano.config.floatX
def setup_method(self):
self.input = T.tensor5("input", dtype=self.dtype)
self.input = tt.tensor5("input", dtype=self.dtype)
self.input.name = "default_V"
self.filters = T.tensor5("filters", dtype=self.dtype)
self.filters = tt.tensor5("filters", dtype=self.dtype)
self.filters.name = "default_filters"
# This tests can run even when theano.config.blas.ldflags is empty.
super().setup_method()
......@@ -50,10 +50,10 @@ class TestCorr3D(utt.InferShapeTester):
pytest.skip("Need cxx for this test")
N_image_shape = [
T.get_scalar_constant_value(T.as_tensor_variable(x)) for x in image_shape
tt.get_scalar_constant_value(tt.as_tensor_variable(x)) for x in image_shape
]
N_filter_shape = [
T.get_scalar_constant_value(T.as_tensor_variable(x)) for x in filter_shape
tt.get_scalar_constant_value(tt.as_tensor_variable(x)) for x in filter_shape
]
if input is None:
......@@ -296,7 +296,7 @@ class TestCorr3D(utt.InferShapeTester):
)
def test_shape_Constant_tensor(self, border_mode):
# Tests correlation where the {image,filter}_shape is a Constant tensor
as_t = T.as_tensor_variable
as_t = tt.as_tensor_variable
self.validate(
(as_t(3), as_t(2), as_t(7), as_t(5), as_t(5)), (5, 2, 2, 3, 3), border_mode
)
......@@ -327,13 +327,17 @@ class TestCorr3D(utt.InferShapeTester):
def test_wrong_input(self):
# Make sure errors are raised when image and kernel are not 5D tensors
with pytest.raises(Exception):
self.validate((3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=T.dmatrix())
self.validate((3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=tt.dmatrix())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=T.vector())
self.validate((3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=tt.vector())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=T.dtensor3())
self.validate(
(3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=tt.dtensor3()
)
with pytest.raises(Exception):
self.validate((3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=T.dtensor4())
self.validate(
(3, 2, 8, 8, 8), (4, 2, 5, 5, 5), "valid", input=tt.dtensor4()
)
@pytest.mark.skipif(not theano.config.cxx, reason="Need cxx for this test")
def test_dtype_upcast(self):
......@@ -352,8 +356,8 @@ class TestCorr3D(utt.InferShapeTester):
for a_dtype in dtypes:
for b_dtype in dtypes:
c_dtype = theano.scalar.upcast(a_dtype, b_dtype)
a_tens = T.tensor5(dtype=a_dtype)
b_tens = T.tensor5(dtype=b_dtype)
a_tens = tt.tensor5(dtype=a_dtype)
b_tens = tt.tensor5(dtype=b_dtype)
a_tens_val = rand(a_shape, dtype=a_dtype)
b_tens_val = rand(b_shape, dtype=b_dtype)
......@@ -373,8 +377,8 @@ class TestCorr3D(utt.InferShapeTester):
corr3dMM = corr3d.Corr3dMM
adtens = T.dtensor5()
bdtens = T.dtensor5()
adtens = tt.dtensor5()
bdtens = tt.dtensor5()
aivec_vals = [
[4, 5, 6, 3, 3],
[6, 2, 8, 3, 3],
......@@ -422,8 +426,8 @@ class TestCorr3D(utt.InferShapeTester):
corr3dMM = corr3d.Corr3dMM
gradW = corr3d.Corr3dMMGradWeights
adtens = T.dtensor5()
bdtens = T.dtensor5()
adtens = tt.dtensor5()
bdtens = tt.dtensor5()
aivec_vals = [
[1, 5, 6, 3, 3],
[8, 2, 7, 3, 3],
......@@ -482,8 +486,8 @@ class TestCorr3D(utt.InferShapeTester):
corr3dMM = corr3d.Corr3dMM
gradI = corr3d.Corr3dMMGradInputs
adtens = T.dtensor5()
bdtens = T.dtensor5()
adtens = tt.dtensor5()
bdtens = tt.dtensor5()
aivec_vals = [
[1, 5, 6, 3, 3],
[8, 2, 7, 3, 3],
......
tests/tensor/nnet/test_ctc.py
浏览文件 @ e103e7fb
import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
from theano.tensor.nnet.ctc import (
ctc_available,
......@@ -128,7 +128,7 @@ class TestCTC:
t_cost = ctc(t_activations, t_labels, t_activation_times)
# Symbolic gradient of CTC cost
t_grad = T.grad(T.mean(t_cost), t_activations)
t_grad = tt.grad(tt.mean(t_cost), t_activations)
# Compile symbolic functions
train = theano.function([], [t_cost, t_grad])
......
tests/tensor/nnet/test_neighbours.py
浏览文件 @ e103e7fb
import pytest
import numpy as np
import theano
import theano.tensor as T
import theano.tensor as tt
from theano import change_flags
from theano import shared, function
......@@ -30,7 +30,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
images = shared(
np.arange(np.prod(shape), dtype=dtype).reshape(shape)
)
neib_shape = T.as_tensor_variable(pshape)
neib_shape = tt.as_tensor_variable(pshape)
f = function(
[],
......@@ -60,7 +60,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
shape = (2, 3, 4, 4)
for dtype in self.dtypes:
images = shared(np.arange(np.prod(shape), dtype=dtype).reshape(shape))
neib_shape = T.as_tensor_variable((2, 2))
neib_shape = tt.as_tensor_variable((2, 2))
for border in ["valid", "ignore_borders"]:
f = function(
......@@ -114,8 +114,8 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
images = shared(
np.asarray(np.arange(np.prod(shape)).reshape(shape), dtype=dtype)
)
neib_shape = T.as_tensor_variable((3, 3))
neib_step = T.as_tensor_variable((2, 2))
neib_shape = tt.as_tensor_variable((3, 3))
neib_step = tt.as_tensor_variable((2, 2))
for border in ["valid", "ignore_borders"]:
f = function(
[],
......@@ -170,7 +170,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
images = shared(np.arange(np.prod(shape), dtype=dtype).reshape(shape))
for neib_shape in [(3, 2), (2, 3)]:
neib_shape = T.as_tensor_variable(neib_shape)
neib_shape = tt.as_tensor_variable(neib_shape)
f = function([], images2neibs(images, neib_shape), mode=self.mode)
with pytest.raises(TypeError):
f()
......@@ -252,8 +252,8 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
images = shared(
np.asarray(np.arange(np.prod(shape)).reshape(shape), dtype=dtype)
)
neib_shape = T.as_tensor_variable(neib_shape)
neib_step = T.as_tensor_variable(neib_step)
neib_shape = tt.as_tensor_variable(neib_shape)
neib_step = tt.as_tensor_variable(neib_step)
expected = np.asarray(expected)
f = function(
......@@ -304,8 +304,8 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
x.shape[2] + 2 * extra[0],
x.shape[3] + 2 * extra[1],
)
padded_x = T.zeros(padded_shape)
padded_x = T.set_subtensor(
padded_x = tt.zeros(padded_shape)
padded_x = tt.set_subtensor(
padded_x[:, :, extra[0] : -extra[0], extra[1] : -extra[1]], x
)
x_using_valid = images2neibs(
......@@ -341,8 +341,8 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
x.shape[2] + 2 * extra[0],
x.shape[3] + 2 * extra[1],
)
padded_x = T.zeros(padded_shape)
padded_x = T.set_subtensor(
padded_x = tt.zeros(padded_shape)
padded_x = tt.set_subtensor(
padded_x[:, :, extra[0] : -extra[0], extra[1] : -extra[1]], x
)
x_using_valid = images2neibs(
......@@ -361,7 +361,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
images = shared(np.arange(np.prod(shape), dtype=dtype).reshape(shape))
for neib_shape in [(3, 2), (2, 3)]:
neib_shape = T.as_tensor_variable(neib_shape)
neib_shape = tt.as_tensor_variable(neib_shape)
f = function(
[],
......@@ -373,7 +373,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
for shape in [(2, 3, 2, 3), (2, 3, 3, 2)]:
images = shared(np.arange(np.prod(shape)).reshape(shape))
neib_shape = T.as_tensor_variable((3, 3))
neib_shape = tt.as_tensor_variable((3, 3))
f = function(
[],
images2neibs(images, neib_shape, mode="wrap_centered"),
......@@ -385,7 +385,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
# Test a valid shapes
shape = (2, 3, 3, 3)
images = shared(np.arange(np.prod(shape)).reshape(shape))
neib_shape = T.as_tensor_variable((3, 3))
neib_shape = tt.as_tensor_variable((3, 3))
f = function(
[],
......@@ -467,33 +467,33 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
def test_neibs_valid_with_inconsistent_borders(self):
shape = (2, 3, 5, 5)
images = T.dtensor4()
images = tt.dtensor4()
images_val = np.arange(np.prod(shape), dtype="float32").reshape(shape)
f = theano.function(
[images], T.sqr(images2neibs(images, (2, 2), mode="valid")), mode=self.mode
[images], tt.sqr(images2neibs(images, (2, 2), mode="valid")), mode=self.mode
)
with pytest.raises(TypeError):
f(images_val)
def test_neibs_half_with_inconsistent_borders(self):
shape = (2, 3, 5, 5)
images = T.dtensor4()
images = tt.dtensor4()
images_val = np.arange(np.prod(shape), dtype="float32").reshape(shape)
f = theano.function(
[images], T.sqr(images2neibs(images, (2, 2), mode="half")), mode=self.mode
[images], tt.sqr(images2neibs(images, (2, 2), mode="half")), mode=self.mode
)
with pytest.raises(TypeError):
f(images_val)
def test_neibs_full_with_inconsistent_borders(self):
shape = (2, 3, 5, 5)
images = T.dtensor4()
images = tt.dtensor4()
images_val = np.arange(np.prod(shape), dtype="float32").reshape(shape)
f = theano.function(
[images], T.sqr(images2neibs(images, (2, 2), mode="full")), mode=self.mode
[images], tt.sqr(images2neibs(images, (2, 2), mode="full")), mode=self.mode
)
with pytest.raises(TypeError):
f(images_val)
......@@ -503,12 +503,12 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
# or that we crash in a few other case found while
# investigating that case
img = T.tensor4("img")
patches = T.nnet.neighbours.images2neibs(img, [16, 16])
img = tt.tensor4("img")
patches = tt.nnet.neighbours.images2neibs(img, [16, 16])
extractPatches = theano.function([img], patches, mode=self.mode)
patsRecovery = T.matrix("patsRecovery")
original_size = T.ivector("original_size")
patsRecovery = tt.matrix("patsRecovery")
original_size = tt.ivector("original_size")
for mode in ["valid", "ignore_borders"]:
out = neibs2images(patsRecovery, (16, 16), original_size, mode=mode)
......@@ -528,7 +528,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
def speed_neibs(self):
shape = (100, 40, 18, 18)
images = shared(np.arange(np.prod(shape), dtype="float32").reshape(shape))
neib_shape = T.as_tensor_variable((3, 3))
neib_shape = tt.as_tensor_variable((3, 3))
f = function([], images2neibs(images, neib_shape), mode=self.mode)
......@@ -538,7 +538,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
def speed_neibs_wrap_centered(self):
shape = (100, 40, 18, 18)
images = shared(np.arange(np.prod(shape), dtype="float32").reshape(shape))
neib_shape = T.as_tensor_variable((3, 3))
neib_shape = tt.as_tensor_variable((3, 3))
f = function(
[], images2neibs(images, neib_shape, mode="wrap_centered"), mode=self.mode
......@@ -550,7 +550,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
def speed_neibs_half(self):
shape = (100, 40, 18, 18)
images = shared(np.arange(np.prod(shape), dtype="float32").reshape(shape))
neib_shape = T.as_tensor_variable((3, 3))
neib_shape = tt.as_tensor_variable((3, 3))
f = function([], images2neibs(images, neib_shape, mode="half"), mode=self.mode)
......@@ -560,7 +560,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
def speed_neibs_full(self):
shape = (100, 40, 18, 18)
images = shared(np.arange(np.prod(shape), dtype="float32").reshape(shape))
neib_shape = T.as_tensor_variable((3, 3))
neib_shape = tt.as_tensor_variable((3, 3))
f = function([], images2neibs(images, neib_shape, mode="full"), mode=self.mode)
......@@ -570,7 +570,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
def test_infer_shape(self):
shape = (100, 40, 6, 3)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(2, 1), mode="valid")],
......@@ -585,7 +585,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
)
shape = (100, 40, 5, 4)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(2, 1), mode="ignore_borders")],
......@@ -594,7 +594,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
)
shape = (100, 40, 5, 3)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(2, 3), mode="ignore_borders")],
......@@ -604,7 +604,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
shape = (100, 40, 6, 7)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(2, 2), mode="ignore_borders")],
......@@ -613,7 +613,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
)
shape = (100, 40, 5, 10)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(3, 3), mode="wrap_centered")],
......@@ -622,7 +622,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
)
shape = (100, 40, 6, 4)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(2, 1), mode="half")],
......@@ -637,7 +637,7 @@ class TestImages2Neibs(unittest_tools.InferShapeTester):
)
shape = (100, 40, 6, 5)
images = np.ones(shape).astype("float32")
x = T.ftensor4()
x = tt.ftensor4()
self._compile_and_check(
[x],
[images2neibs(x, neib_shape=(2, 1), mode="full")],
......
tests/tensor/nnet/test_nnet.py
浏览文件 @ e103e7fb
......@@ -3,11 +3,9 @@ import pytest
import numpy as np
import theano
import theano.tensor as tt
from theano import config
from theano import tensor as T
from theano import tensor
from theano import gof
from theano.gof.opt import check_stack_trace
from theano import printing
......@@ -42,9 +40,10 @@ from theano.tensor.nnet import (
binary_crossentropy,
sigmoid_binary_crossentropy,
confusion_matrix,
logsoftmax,
)
from theano.tensor import matrix, vector, lvector, scalar
from theano.tensor.nnet.nnet import softsign
from theano.tensor.nnet.nnet import softsign, LogSoftmax
from tests import unittest_tools as utt
from tests.tensor.test_basic import (
......@@ -99,7 +98,7 @@ class TestSoftmax(utt.InferShapeTester):
self._compile_and_check([admat], [Softmax()(admat)], [admat_val], Softmax)
def test_vector(self):
x = T.vector()
x = tt.vector()
f = theano.function([x], softmax_op(x))
xv = np.random.randn(6).astype(config.floatX)
......@@ -144,8 +143,8 @@ class TestSoftmaxWithBias(utt.InferShapeTester):
)
W = theano.shared(value=initial_W, name="W")
vbias = theano.shared(value=0.1, name="vbias") # 0.01
hid = T.vector("hid")
f = theano.function([hid], T.nnet.softmax_op(T.dot(hid, W.T) + vbias))
hid = tt.vector("hid")
f = theano.function([hid], softmax_op(tt.dot(hid, W.T) + vbias))
ops = [node.op for node in f.maker.fgraph.toposort()]
assert softmax_with_bias not in ops
assert softmax_op in ops
......@@ -156,7 +155,7 @@ class TestSoftmaxWithBias(utt.InferShapeTester):
def test_softmax_with_bias_trace(self):
a = theano.shared(np.random.randn(3).astype(config.floatX))
b = theano.shared(np.float32(np.random.randn()))
sm = T.nnet.softmax(a + b)
sm = softmax(a + b)
f = theano.function([], sm)
assert check_stack_trace(f, ops_to_check="last")
......@@ -202,7 +201,7 @@ class TestLogSoftmax(utt.InferShapeTester):
utt.verify_grad(f, [np.random.rand(3, 4)])
def test_vector(self):
x = T.vector()
x = tt.vector()
f = theano.function([x], logsoftmax_op(x))
xv = np.random.randn(6).astype(config.floatX)
......@@ -218,16 +217,16 @@ class TestLogSoftmax(utt.InferShapeTester):
m = theano.config.mode
m = theano.compile.get_mode(m)
m.check_isfinite = False
x, y = tensor.matrices("xy")
x, y = tt.matrices("xy")
# regular softmax and crossentropy
sm = tensor.nnet.softmax(x)
cm = tensor.nnet.categorical_crossentropy(sm, y)
sm = softmax(x)
cm = categorical_crossentropy(sm, y)
# numerically stable log-softmax with crossentropy
logsm = tensor.nnet.logsoftmax(x)
sm2 = tensor.exp(logsm) # just used to show equivalence with sm
cm2 = -tensor.sum(y * logsm, axis=1)
grad = tensor.grad(cm2.mean(), x)
logsm = logsoftmax(x)
sm2 = tt.exp(logsm) # just used to show equivalence with sm
cm2 = -tt.sum(y * logsm, axis=1)
grad = tt.grad(cm2.mean(), x)
# create some inputs into a softmax that are large and labels
a = np.exp(10 * np.random.rand(5, 10).astype(theano.config.floatX))
......@@ -263,14 +262,12 @@ class TestLogSoftmax(utt.InferShapeTester):
# Check that Log(Softmax(x)) is substituted with Logsoftmax(x). Note that
# only the forward pass is checked (i.e., doesn't check the gradient)
x, y = tensor.matrices("xy")
sm = tensor.nnet.softmax(x)
logsm = tensor.log(sm)
x, y = tt.matrices("xy")
sm = softmax(x)
logsm = tt.log(sm)
f = theano.function([x], logsm)
assert isinstance(
f.maker.fgraph.outputs[0].owner.op, theano.tensor.nnet.nnet.LogSoftmax
)
assert check_stack_trace(f, ops_to_check=theano.tensor.nnet.nnet.LogSoftmax)
assert isinstance(f.maker.fgraph.outputs[0].owner.op, LogSoftmax)
assert check_stack_trace(f, ops_to_check=LogSoftmax)
def test_local_softmax_grad_optimization_and_big_input(self):
# Test the Logsoftmax's grad substitution.
......@@ -287,8 +284,8 @@ class TestLogSoftmax(utt.InferShapeTester):
a = np.exp(10 * np.random.rand(5, 10).astype(theano.config.floatX))
def myfunc(x):
sm = tensor.nnet.softmax(x)
logsm = tensor.log(sm)
sm = softmax(x)
logsm = tt.log(sm)
return logsm
# We set step to 0.1 because for big values we need a big epsilon
......@@ -302,19 +299,17 @@ class TestLogSoftmax(utt.InferShapeTester):
# but with a different elemwise operation than true_div is not
# optimized.
x = T.matrix("x")
y = T.log(T.nnet.softmax(x))
g = T.grad(y.sum(), x)
x = tt.matrix("x")
y = tt.log(softmax(x))
g = tt.grad(y.sum(), x)
softmax_grad_node = g.owner
assert softmax_grad_node.op == softmax_grad
true_div_node = softmax_grad_node.inputs[0].owner
assert true_div_node.op == tensor.true_div
assert true_div_node.op == tt.true_div
# We replace the elemwise true_div op by an elemwise add.
new_g = softmax_grad(
tensor.add(*true_div_node.inputs), softmax_grad_node.inputs[1]
)
new_g = softmax_grad(tt.add(*true_div_node.inputs), softmax_grad_node.inputs[1])
fgraph = gof.FunctionGraph([x], [new_g])
theano.compile.mode.optdb.query(theano.compile.mode.OPT_FAST_RUN).optimize(
......@@ -361,7 +356,7 @@ class TestCrossentropySoftmax1Hot:
y_idx = [3]
def f(a):
return crossentropy_softmax_1hot(T.shape_padleft(a), y_idx)[0]
return crossentropy_softmax_1hot(tt.shape_padleft(a), y_idx)[0]
utt.verify_grad(f, [np.random.rand(4)])
......@@ -369,7 +364,7 @@ class TestCrossentropySoftmax1Hot:
y_idx = [3]
def f(a, b):
return crossentropy_softmax_1hot(T.shape_padleft(a) + b, y_idx)[0]
return crossentropy_softmax_1hot(tt.shape_padleft(a) + b, y_idx)[0]
utt.verify_grad(f, [np.random.rand(4), np.random.rand(4)])
......@@ -380,7 +375,7 @@ class TestCrossentropySoftmax1HotWithBiasDx(utt.InferShapeTester):
def f(sm):
# Class indices
y = np.random.randint(low=0, high=5, size=10).astype(class_dtype)
return theano.tensor.nnet.crossentropy_softmax_1hot_with_bias_dx(
return crossentropy_softmax_1hot_with_bias_dx(
np.random.rand(10), sm, y # Gradient w.r.t. NLL. # Softmax output.
)
......@@ -398,7 +393,7 @@ class TestCrossentropySoftmax1HotWithBiasDx(utt.InferShapeTester):
softmax_output /= softmax_output.sum(axis=1).reshape(10, 1)
def f(dy):
return theano.tensor.nnet.crossentropy_softmax_1hot_with_bias_dx(
return crossentropy_softmax_1hot_with_bias_dx(
dy, softmax_output, rng.randint(low=0, high=5, size=10)
)
......@@ -438,7 +433,7 @@ class TestCrossentropySoftmax1HotWithBiasDx(utt.InferShapeTester):
class TestCrossentropySoftmaxArgmax1HotWithBias(utt.InferShapeTester):
def setup_method(self):
self.op = theano.tensor.nnet.crossentropy_softmax_argmax_1hot_with_bias
self.op = crossentropy_softmax_argmax_1hot_with_bias
super().setup_method()
def test_grads(self):
......@@ -504,7 +499,7 @@ class TestCrossentropySoftmaxArgmax1HotWithBias(utt.InferShapeTester):
class TestPrepend(utt.InferShapeTester):
def test_prepend_constant(self):
x = tensor.matrix("x")
x = tt.matrix("x")
y = Prepend_scalar_constant_to_each_row(4.0)(x)
f = theano.function([x], y)
m = np.random.rand(3, 5).astype(config.floatX)
......@@ -514,7 +509,7 @@ class TestPrepend(utt.InferShapeTester):
def test_prepend_basic(self):
"""Test basic functionality."""
x = tensor.matrix("x")
x = tt.matrix("x")
y = Prepend_scalar_to_each_row()(5.0, x)
f = theano.function([x], y)
m = np.ones((3, 5), dtype="float32")
......@@ -562,8 +557,8 @@ class TestCrossentropyCategorical1HotGrad(utt.InferShapeTester):
class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
def test_grad(self):
x = tensor.matrix("x")
one_of_n = tensor.lvector("one_of_n")
x = tt.matrix("x")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
xe = op(x, one_of_n)
f = theano.function([x, one_of_n], xe)
......@@ -574,7 +569,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
def oplike(x):
return op(x, [0, 1])
tensor.verify_grad(oplike, [x_val], rng=np.random)
tt.verify_grad(oplike, [x_val], rng=np.random)
def test_infer_shape(self):
admat = matrix()
......@@ -590,8 +585,8 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
)
def test_softmax_optimizations(self):
x = tensor.matrix("x")
one_of_n = tensor.lvector("one_of_n")
x = tt.matrix("x")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
# xe = op(x, one_of_n)
......@@ -604,8 +599,8 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
assert fgraph.outputs[0].owner.op == crossentropy_softmax_argmax_1hot_with_bias
def test_softmax_optimizations_vector(self):
x = tensor.vector("x")
one_of_n = tensor.lvector("one_of_n")
x = tt.vector("x")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
fgraph = gof.FunctionGraph([x, one_of_n], [op(softmax_op(x), one_of_n)])
assert fgraph.outputs[0].owner.op == op
......@@ -616,9 +611,9 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
assert fgraph.outputs[0].owner.op == crossentropy_softmax_argmax_1hot_with_bias
def test_softmax_optimizations_w_bias(self):
x = tensor.matrix("x")
b = tensor.vector("b")
one_of_n = tensor.lvector("one_of_n")
x = tt.matrix("x")
b = tt.vector("b")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
# xe = op(x, one_of_n)
......@@ -644,14 +639,14 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
assert fgraph.outputs[0].owner.op == crossentropy_softmax_argmax_1hot_with_bias
def test_softmax_optimizations_w_bias2(self):
x = tensor.matrix("x")
b = tensor.vector("b")
c = tensor.vector("c")
one_of_n = tensor.lvector("one_of_n")
x = tt.matrix("x")
b = tt.vector("b")
c = tt.vector("c")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
fgraph = gof.FunctionGraph(
[x, b, c, one_of_n], [op(softmax_op(T.add(x, b, c)), one_of_n)]
[x, b, c, one_of_n], [op(softmax_op(tt.add(x, b, c)), one_of_n)]
)
assert fgraph.outputs[0].owner.op == op
......@@ -672,9 +667,9 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
assert fgraph.outputs[0].owner.op == crossentropy_softmax_argmax_1hot_with_bias
def test_softmax_optimizations_w_bias_vector(self):
x = tensor.vector("x")
b = tensor.vector("b")
one_of_n = tensor.lvector("one_of_n")
x = tt.vector("x")
b = tt.vector("b")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
fgraph = gof.FunctionGraph([x, b, one_of_n], [op(softmax_op(x + b), one_of_n)])
assert fgraph.outputs[0].owner.op == op
......@@ -695,12 +690,12 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
assert fgraph.outputs[0].owner.op == crossentropy_softmax_argmax_1hot_with_bias
def test_softmax_grad_optimizations(self):
x = tensor.matrix("x")
one_of_n = tensor.lvector("one_of_n")
x = tt.matrix("x")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
xe = op(softmax_op(x), one_of_n)
sum_xe = tensor.sum(xe)
g_x = tensor.grad(sum_xe, x)
sum_xe = tt.sum(xe)
g_x = tt.grad(sum_xe, x)
fgraph = gof.FunctionGraph([x, one_of_n], [g_x])
assert check_stack_trace(
fgraph, ops_to_check=[crossentropy_softmax_1hot_with_bias_dx, softmax_op]
......@@ -737,12 +732,12 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
assert not has_softmaxdx
def test_softmax_grad_optimizations_vector(self):
x = tensor.vector("x")
one_of_n = tensor.lvector("one_of_n")
x = tt.vector("x")
one_of_n = tt.lvector("one_of_n")
op = crossentropy_categorical_1hot
xe = op(softmax_op(x), one_of_n)
sum_xe = tensor.sum(xe)
g_x = tensor.grad(sum_xe, x)
sum_xe = tt.sum(xe)
g_x = tt.grad(sum_xe, x)
fgraph = gof.FunctionGraph([x, one_of_n], [g_x])
# print 'BEFORE'
......@@ -786,16 +781,16 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
x_val = rng.randn(3, 5).astype(config.floatX)
b_val = rng.randn(5).astype(config.floatX)
y_val = np.asarray([2, 4, 1])
x = T.matrix("x")
b = T.vector("b")
y = T.lvector("y")
x = tt.matrix("x")
b = tt.vector("b")
y = tt.lvector("y")
# Basic case
expressions = [
T.sum(-T.log(softmax(x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x))[T.arange(y.shape[0]), y]),
T.sum(-T.log(softmax(x))[T.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
]
for expr in expressions:
# Verify the optimizer worked on the expressions
......@@ -809,14 +804,14 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
ops = [node.op for node in f.maker.fgraph.toposort()]
assert len(ops) == 4
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, y_val)
except Exception:
theano.printing.debugprint(f)
raise
# Also verify the gradient wrt x
g = theano.function([x, y], T.grad(expr, x), mode=mode)
g = theano.function([x, y], tt.grad(expr, x), mode=mode)
assert check_stack_trace(
g, ops_to_check=[crossentropy_softmax_1hot_with_bias_dx, softmax_op]
)
......@@ -835,10 +830,10 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# Test that a biased softmax is optimized correctly
bias_expressions = [
T.sum(-T.log(softmax(x + b)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(b + x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x + b))[T.arange(y.shape[0]), y]),
T.sum(-T.log(softmax(b + x))[T.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(x + b)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(b + x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x + b))[tt.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(b + x))[tt.arange(y.shape[0]), y]),
]
for expr in bias_expressions:
......@@ -856,7 +851,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
except Exception:
theano.printing.debugprint(f)
raise
g = theano.function([x, b, y], T.grad(expr, x), mode=mode)
g = theano.function([x, b, y], tt.grad(expr, x), mode=mode)
assert check_stack_trace(
g,
ops_to_check=[
......@@ -879,10 +874,10 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# Test that using "mean" instead of sum works, too
mean_expressions = [
T.mean(-T.log(softmax(x)[T.arange(y.shape[0]), y])),
-T.mean(T.log(softmax(x)[T.arange(y.shape[0]), y])),
-T.mean(T.log(softmax(x))[T.arange(y.shape[0]), y]),
T.mean(-T.log(softmax(x))[T.arange(y.shape[0]), y]),
tt.mean(-tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-tt.mean(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-tt.mean(tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
tt.mean(-tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
]
for expr in mean_expressions:
......@@ -896,13 +891,13 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
ops = [node.op for node in f.maker.fgraph.toposort()]
assert len(ops) == 6
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, y_val)
except Exception:
theano.printing.debugprint(f)
raise
g = theano.function([x, y], T.grad(expr, x), mode=mode)
g = theano.function([x, y], tt.grad(expr, x), mode=mode)
assert check_stack_trace(
g, ops_to_check=[crossentropy_softmax_1hot_with_bias_dx, softmax_op]
)
......@@ -922,10 +917,10 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
raise
mean_bias_expressions = [
T.mean(-T.log(softmax(x + b)[T.arange(y.shape[0]), y])),
-T.mean(T.log(softmax(b + x)[T.arange(y.shape[0]), y])),
-T.mean(T.log(softmax(x + b))[T.arange(y.shape[0]), y]),
T.mean(-T.log(softmax(b + x))[T.arange(y.shape[0]), y]),
tt.mean(-tt.log(softmax(x + b)[tt.arange(y.shape[0]), y])),
-tt.mean(tt.log(softmax(b + x)[tt.arange(y.shape[0]), y])),
-tt.mean(tt.log(softmax(x + b))[tt.arange(y.shape[0]), y]),
tt.mean(-tt.log(softmax(b + x))[tt.arange(y.shape[0]), y]),
]
for expr in mean_bias_expressions:
......@@ -939,11 +934,11 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
ops = [node.op for node in f.maker.fgraph.toposort()]
assert len(ops) == 4
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
except Exception:
theano.printing.debugprint(f)
raise
g = theano.function([x, b, y], T.grad(expr, x), mode=mode)
g = theano.function([x, b, y], tt.grad(expr, x), mode=mode)
assert check_stack_trace(
g,
ops_to_check=[
......@@ -972,14 +967,14 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
rng = np.random.RandomState(utt.fetch_seed())
x_val = rng.randn(3, 5).astype(config.floatX)
y_val = np.asarray([2, 4, 1], dtype="int64")
x = T.matrix("x")
y = T.lvector("y")
yi = T.cast(y, "int32")
x = tt.matrix("x")
y = tt.lvector("y")
yi = tt.cast(y, "int32")
expressions = [
T.sum(-T.log(softmax(x)[T.arange(yi.shape[0]), yi])),
-T.sum(T.log(softmax(x)[T.arange(yi.shape[0]), yi])),
-T.sum(T.log(softmax(x))[T.arange(yi.shape[0]), yi]),
T.sum(-T.log(softmax(x))[T.arange(yi.shape[0]), yi]),
tt.sum(-tt.log(softmax(x)[tt.arange(yi.shape[0]), yi])),
-tt.sum(tt.log(softmax(x)[tt.arange(yi.shape[0]), yi])),
-tt.sum(tt.log(softmax(x))[tt.arange(yi.shape[0]), yi]),
tt.sum(-tt.log(softmax(x))[tt.arange(yi.shape[0]), yi]),
]
for expr in expressions:
......@@ -991,14 +986,14 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
ops = [node.op for node in f.maker.fgraph.toposort()]
assert len(ops) == 5
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, y_val)
except Exception:
theano.printing.debugprint(f)
raise
# Also verify the gradient wrt x
g = theano.function([x, y], T.grad(expr, x), mode=mode)
g = theano.function([x, y], tt.grad(expr, x), mode=mode)
if verbose:
theano.printing.debugprint(g)
try:
......@@ -1021,13 +1016,13 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
x_val = rng.randn(5).astype(config.floatX)
y_val = np.asarray([2])
x = T.vector("x")
y = T.lvector("y")
x = tt.vector("x")
y = tt.lvector("y")
# Test that a biased softmax is optimized correctly
bias_expressions = [
T.sum(-T.log(softmax(x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x)[T.arange(y.shape[0]), y])),
tt.sum(-tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
]
for expr in bias_expressions:
......@@ -1038,12 +1033,12 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
ops = [node.op for node in f.maker.fgraph.toposort()]
assert len(ops) == 5
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, y_val)
except Exception:
theano.printing.debugprint(f)
raise
g = theano.function([x, y], T.grad(expr, x), mode=mode)
g = theano.function([x, y], tt.grad(expr, x), mode=mode)
if verbose:
printing.debugprint(g)
try:
......@@ -1067,16 +1062,16 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
b_val = rng.randn(5).astype(config.floatX)
y_val = np.asarray([2])
x = T.vector("x")
b = T.vector("b")
y = T.lvector("y")
x = tt.vector("x")
b = tt.vector("b")
y = tt.lvector("y")
# Test that a biased softmax is optimized correctly
bias_expressions = [
T.sum(-T.log(softmax(x + b)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(b + x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x + b))[T.arange(y.shape[0]), y]),
T.sum(-T.log(softmax(b + x))[T.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(x + b)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(b + x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x + b))[tt.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(b + x))[tt.arange(y.shape[0]), y]),
]
for expr in bias_expressions:
......@@ -1088,7 +1083,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# [big_op, sum, dim_shuffle]
assert len(ops) == 3
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, b_val, y_val)
except Exception:
theano.printing.debugprint(f)
......@@ -1097,7 +1092,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
backup = config.warn.sum_div_dimshuffle_bug
config.warn.sum_div_dimshuffle_bug = False
try:
g = theano.function([x, b, y], T.grad(expr, x), mode=mode)
g = theano.function([x, b, y], tt.grad(expr, x), mode=mode)
finally:
config.warn.sum_div_dimshuffle_bug = backup
......@@ -1127,17 +1122,17 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
b_val = rng.randn(5).astype(config.floatX)
y_val = np.asarray([2])
x = T.vector("x")
b = T.vector("b")
y_ = T.lvector("y_")
x = tt.vector("x")
b = tt.vector("b")
y_ = tt.lvector("y_")
y = y_.flatten()
# Test that a biased softmax is optimized correctly
bias_expressions = [
T.sum(-T.log(softmax(x + b)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(b + x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x + b))[T.arange(y.shape[0]), y]),
T.sum(-T.log(softmax(b + x))[T.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(x + b)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(b + x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x + b))[tt.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(b + x))[tt.arange(y.shape[0]), y]),
]
for expr in bias_expressions:
......@@ -1149,7 +1144,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# [big_op, sum, dim_shuffle, flatten]
assert len(ops) <= 4
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, b_val, y_val)
except Exception:
theano.printing.debugprint(f)
......@@ -1158,7 +1153,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
backup = config.warn.sum_div_dimshuffle_bug
config.warn.sum_div_dimshuffle_bug = False
try:
g = theano.function([x, b, y], T.grad(expr, x), mode=mode)
g = theano.function([x, b, y], tt.grad(expr, x), mode=mode)
finally:
config.warn.sum_div_dimshuffle_bug = backup
......@@ -1189,17 +1184,17 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
b_val = rng.randn(5).astype(config.floatX)
y_val = np.asarray([2])
x = T.vector("x")
b = T.vector("b")
y_ = T.lvector("y_")
y = T.specify_shape(y_, (1,))
x = tt.vector("x")
b = tt.vector("b")
y_ = tt.lvector("y_")
y = tt.specify_shape(y_, (1,))
# Test that a biased softmax is optimized correctly
bias_expressions = [
T.sum(-T.log(softmax(x + b)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(b + x)[T.arange(y.shape[0]), y])),
-T.sum(T.log(softmax(x + b))[T.arange(y.shape[0]), y]),
T.sum(-T.log(softmax(b + x))[T.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(x + b)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(b + x)[tt.arange(y.shape[0]), y])),
-tt.sum(tt.log(softmax(x + b))[tt.arange(y.shape[0]), y]),
tt.sum(-tt.log(softmax(b + x))[tt.arange(y.shape[0]), y]),
]
for expr in bias_expressions:
......@@ -1211,7 +1206,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# [big_op, sum, dim_shuffle, specify_shape]
assert len(ops) <= 4
assert crossentropy_softmax_argmax_1hot_with_bias in ops
assert not [1 for o in ops if isinstance(o, T.AdvancedSubtensor)]
assert not [1 for o in ops if isinstance(o, tt.AdvancedSubtensor)]
f(x_val, b_val, y_val)
except Exception:
theano.printing.debugprint(f)
......@@ -1220,7 +1215,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
backup = config.warn.sum_div_dimshuffle_bug
config.warn.sum_div_dimshuffle_bug = False
try:
g = theano.function([x, b, y], T.grad(expr, x), mode=mode)
g = theano.function([x, b, y], tt.grad(expr, x), mode=mode)
finally:
config.warn.sum_div_dimshuffle_bug = backup
......@@ -1246,9 +1241,9 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
rng = np.random.RandomState(utt.fetch_seed())
x_val = rng.randn(3, 5).astype(config.floatX)
y_val = np.asarray([2, 4, 1])
x = T.matrix("x")
y = T.lvector("y")
a = T.scalar("a")
x = tt.matrix("x")
y = tt.lvector("y")
a = tt.scalar("a")
def validate_fn_graph(func):
# The graph of the function should not have softmax anymore
......@@ -1282,22 +1277,22 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# Cases to test
expressions = [
a * T.sum(-T.log(softmax(x)[T.arange(y.shape[0]), y])),
-a * T.sum(T.log(softmax(x)[T.arange(y.shape[0]), y])),
a * (-T.sum(T.log(softmax(x)[T.arange(y.shape[0]), y]))),
a * T.sum(T.log(softmax(x)[T.arange(y.shape[0]), y])),
a * T.sum(-T.log(softmax(x))[T.arange(y.shape[0]), y]),
-a * T.sum(T.log(softmax(x))[T.arange(y.shape[0]), y]),
a * (-T.sum(T.log(softmax(x))[T.arange(y.shape[0]), y])),
a * T.sum(T.log(softmax(x))[T.arange(y.shape[0]), y]),
a * T.mean(-T.log(softmax(x)[T.arange(y.shape[0]), y])),
-a * T.mean(T.log(softmax(x)[T.arange(y.shape[0]), y])),
a * (-T.mean(T.log(softmax(x)[T.arange(y.shape[0]), y]))),
a * T.mean(T.log(softmax(x)[T.arange(y.shape[0]), y])),
a * T.mean(-T.log(softmax(x))[T.arange(y.shape[0]), y]),
-a * T.mean(T.log(softmax(x))[T.arange(y.shape[0]), y]),
a * (-T.mean(T.log(softmax(x))[T.arange(y.shape[0]), y])),
a * T.mean(T.log(softmax(x))[T.arange(y.shape[0]), y]),
a * tt.sum(-tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-a * tt.sum(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
a * (-tt.sum(tt.log(softmax(x)[tt.arange(y.shape[0]), y]))),
a * tt.sum(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
a * tt.sum(-tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
-a * tt.sum(tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
a * (-tt.sum(tt.log(softmax(x))[tt.arange(y.shape[0]), y])),
a * tt.sum(tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
a * tt.mean(-tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
-a * tt.mean(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
a * (-tt.mean(tt.log(softmax(x)[tt.arange(y.shape[0]), y]))),
a * tt.mean(tt.log(softmax(x)[tt.arange(y.shape[0]), y])),
a * tt.mean(-tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
-a * tt.mean(tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
a * (-tt.mean(tt.log(softmax(x))[tt.arange(y.shape[0]), y])),
a * tt.mean(tt.log(softmax(x))[tt.arange(y.shape[0]), y]),
]
for expr in expressions:
......@@ -1312,7 +1307,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
raise
# Verify the gradient wrt x
g = theano.function([x, y, a], T.grad(expr, x), mode=mode)
g = theano.function([x, y, a], tt.grad(expr, x), mode=mode)
try:
assert 3 <= len(g.maker.fgraph.toposort()) <= 6
validate_grad_graph(g)
......@@ -1323,7 +1318,7 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
# Verify the gradient when providing output gradient
h = theano.function(
[x, y, a], T.grad(expr, x, known_grads={expr: a * x.sum()}), mode=mode
[x, y, a], tt.grad(expr, x, known_grads={expr: a * x.sum()}), mode=mode
)
try:
assert 6 <= len(h.maker.fgraph.toposort()) <= 8
......@@ -1335,10 +1330,10 @@ class TestCrossentropyCategorical1Hot(utt.InferShapeTester):
def test_argmax_pushdown():
x = tensor.matrix()
x = tt.matrix()
for sm in [softmax_graph, softmax_op]:
# test that the max_and_argmax is pushed down if the max is not used
out = tensor.max_and_argmax(sm(tensor.exp(tensor.tanh(sigmoid(x)))), axis=-1)[1]
out = tt.max_and_argmax(sm(tt.exp(tt.tanh(sigmoid(x)))), axis=-1)[1]
fgraph = gof.FunctionGraph([x], [out])
theano.compile.mode.optdb.query(theano.compile.mode.OPT_FAST_RUN).optimize(
fgraph
......@@ -1348,11 +1343,11 @@ def test_argmax_pushdown():
# for node in fgraph.toposort():
# print node.op
assert len(fgraph.toposort()) == 1
assert isinstance(fgraph.toposort()[0].op, tensor.basic.Argmax)
assert check_stack_trace(fgraph, ops_to_check=tensor.basic.Argmax)
x = tensor.matrix()
assert isinstance(fgraph.toposort()[0].op, tt.Argmax)
assert check_stack_trace(fgraph, ops_to_check=tt.Argmax)
x = tt.matrix()
# test that the max_and_argmax is not pushed down if the max is used
out = tensor.max_and_argmax(sm(tensor.exp(tensor.tanh(sigmoid(x)))), axis=-1)[0]
out = tt.max_and_argmax(sm(tt.exp(tt.tanh(sigmoid(x)))), axis=-1)[0]
fgraph = gof.FunctionGraph([x], [out])
assert hasattr(fgraph.outputs[0].tag, "trace")
......@@ -1369,17 +1364,17 @@ def test_argmax_pushdown():
# for node in fgraph.toposort():
# print node.op
assert len(fgraph.toposort()) == 3
assert isinstance(fgraph.toposort()[0].op, tensor.Elemwise)
assert isinstance(fgraph.toposort()[0].op, tt.Elemwise)
assert isinstance(fgraph.toposort()[1].op, Softmax)
assert isinstance(fgraph.toposort()[2].op, tensor.CAReduce)
assert isinstance(fgraph.toposort()[2].op, tt.CAReduce)
assert isinstance(fgraph.toposort()[2].op.scalar_op, theano.scalar.Maximum)
def test_argmax_pushdown_bias():
x = tensor.matrix()
b = tensor.vector()
x = tt.matrix()
b = tt.vector()
out = tensor.argmax(softmax_with_bias(x, b), axis=-1)
out = tt.argmax(softmax_with_bias(x, b), axis=-1)
fgraph = gof.FunctionGraph([x, b], [out])
theano.compile.mode.optdb.query(theano.compile.mode.OPT_FAST_RUN).optimize(fgraph)
......@@ -1387,16 +1382,16 @@ def test_argmax_pushdown_bias():
# print 'AFTER'
# for node in fgraph.toposort():
# print node.op
types_to_check = (tensor.DimShuffle, tensor.Elemwise, tensor.Argmax)
types_to_check = (tt.DimShuffle, tt.Elemwise, tt.Argmax)
assert len(fgraph.toposort()) == 3
for i, type in enumerate(types_to_check):
assert isinstance(fgraph.toposort()[i].op, type)
assert check_stack_trace(fgraph, ops_to_check=types_to_check)
x = tensor.matrix()
b = tensor.vector()
out = tensor.max_and_argmax(softmax_with_bias(x, b), axis=-1)[0]
x = tt.matrix()
b = tt.vector()
out = tt.max_and_argmax(softmax_with_bias(x, b), axis=-1)[0]
fgraph = gof.FunctionGraph([x, b], [out])
backup = config.warn.argmax_pushdown_bug
......@@ -1413,9 +1408,9 @@ def test_argmax_pushdown_bias():
# print node.op
assert len(fgraph.toposort()) == 2
assert isinstance(fgraph.toposort()[0].op, SoftmaxWithBias)
assert isinstance(fgraph.toposort()[1].op, tensor.CAReduce)
assert isinstance(fgraph.toposort()[1].op, tt.CAReduce)
assert isinstance(fgraph.toposort()[1].op.scalar_op, theano.scalar.Maximum)
assert check_stack_trace(fgraph, ops_to_check=(SoftmaxWithBias, tensor.CAReduce))
assert check_stack_trace(fgraph, ops_to_check=(SoftmaxWithBias, tt.CAReduce))
def test_asymptotic_32():
......@@ -1427,17 +1422,15 @@ def test_asymptotic_32():
for dtype in "float32", "float64":
if dtype == "float32":
x = tensor.fmatrix()
x2 = tensor.fvector()
x = tt.fmatrix()
x2 = tt.fvector()
else:
x = tensor.dmatrix()
x2 = tensor.dvector()
y = tensor.lvector()
x = tt.dmatrix()
x2 = tt.dvector()
y = tt.lvector()
c = categorical_crossentropy(softmax(x + x2), y)
f = theano.function(
[x, y, x2], [c.sum(), tensor.grad(c.sum(), x)], mode="FAST_RUN"
)
f = theano.function([x, y, x2], [c.sum(), tt.grad(c.sum(), x)], mode="FAST_RUN")
xval = np.zeros((5, 5), dtype=dtype).astype(dtype)
x2val = np.zeros(5, dtype=xval.dtype).astype(dtype)
......@@ -1479,8 +1472,8 @@ class TestSoftmaxOpt:
self.mode = self.mode.including("canonicalize")
def test_basic(self):
c = T.matrix()
p_y = T.exp(c) / T.exp(c).sum(axis=1).dimshuffle(0, "x")
c = tt.matrix()
p_y = tt.exp(c) / tt.exp(c).sum(axis=1).dimshuffle(0, "x")
# test that function contains softmax and no div.
f = theano.function([c], p_y, mode=self.mode)
......@@ -1496,8 +1489,8 @@ class TestSoftmaxOpt:
f(self.rng.rand(3, 4).astype(config.floatX))
def test_basic_keepdims(self):
c = T.matrix()
p_y = T.exp(c) / T.exp(c).sum(axis=1, keepdims=True)
c = tt.matrix()
p_y = tt.exp(c) / tt.exp(c).sum(axis=1, keepdims=True)
# test that function contains softmax and no div.
f = theano.function([c], p_y, mode=self.mode)
......@@ -1514,15 +1507,15 @@ class TestSoftmaxOpt:
@pytest.mark.skip(reason="Optimization not enabled for the moment")
def test_grad(self):
c = T.matrix()
p_y = T.exp(c) / T.exp(c).sum(axis=1).dimshuffle(0, "x")
c = tt.matrix()
p_y = tt.exp(c) / tt.exp(c).sum(axis=1).dimshuffle(0, "x")
# test that function contains softmax and softmaxgrad
w = T.matrix()
w = tt.matrix()
backup = config.warn.sum_div_dimshuffle_bug
config.warn.sum_div_dimshuffle_bug = False
try:
g = theano.function([c, w], T.grad((p_y * w).sum(), c))
g = theano.function([c, w], tt.grad((p_y * w).sum(), c))
finally:
config.warn.sum_div_dimshuffle_bug = backup
g_ops = [n.op for n in g.maker.fgraph.toposort()]
......@@ -1538,8 +1531,8 @@ class TestSoftmaxOpt:
@pytest.mark.skip(reason="Optimization not enabled for the moment")
def test_transpose_basic(self):
# this should be a transposed softmax
c = T.matrix()
p_y = T.exp(c) / T.exp(c).sum(axis=0)
c = tt.matrix()
p_y = tt.exp(c) / tt.exp(c).sum(axis=0)
# test that function contains softmax and no div.
theano.function([c], p_y)
......@@ -1549,7 +1542,7 @@ class TestSoftmaxOpt:
backup = config.warn.sum_div_dimshuffle_bug
config.warn.sum_div_dimshuffle_bug = False
try:
theano.function([c], T.grad(p_y.sum(), c))
theano.function([c], tt.grad(p_y.sum(), c))
finally:
config.warn.sum_div_dimshuffle_bug = backup
# printing.debugprint(g)
......@@ -1557,8 +1550,8 @@ class TestSoftmaxOpt:
@pytest.mark.skip(reason="Optimization not enabled for the moment")
def test_1d_basic(self):
# this should be a softmax, but of a one-row matrix
c = T.vector()
p_y = T.exp(c) / T.exp(c).sum()
c = tt.vector()
p_y = tt.exp(c) / tt.exp(c).sum()
# test that function contains softmax and no div.
theano.function([c], p_y)
......@@ -1568,7 +1561,7 @@ class TestSoftmaxOpt:
backup = config.warn.sum_div_dimshuffle_bug
config.warn.sum_div_dimshuffle_bug = False
try:
theano.function([c], T.grad(p_y.sum(), c))
theano.function([c], tt.grad(p_y.sum(), c))
finally:
config.warn.sum_div_dimshuffle_bug = backup
# printing.debugprint(g)
......@@ -1605,7 +1598,7 @@ def test_stabilize_log_softmax():
x = matrix()
y = softmax(x)
z = theano.tensor.log(y)
z = tt.log(y)
f = theano.function([x], z, mode=mode)
assert check_stack_trace(f, ops_to_check="all")
......@@ -1696,8 +1689,8 @@ def test_h_softmax():
#############
# Build graph
#############
x = tensor.matrix("x")
y = tensor.ivector("y")
x = tt.matrix("x")
y = tt.ivector("y")
# This only computes the output corresponding to the target
y_hat_tg = h_softmax(
......@@ -1779,7 +1772,7 @@ def test_selu():
def test_binary_crossentropy_reshape():
# Reported as https://github.com/Theano/Theano/issues/4086
a = tensor.tensor4("a")
a = tt.tensor4("a")
for c in (
binary_crossentropy(sigmoid(a.reshape((-1, 1))), 1).sum(),
binary_crossentropy(sigmoid(a).reshape((-1, 1)), 1).sum(),
......@@ -1818,7 +1811,7 @@ class TestSigmoidBinaryCrossentropy:
# Test sigmoid_binary_crossentropy(p, t) ==
# binary_crossentropy(sigmoid(p), t).
pred, target = inputs = tensor.vectors("pt")
pred, target = inputs = tt.vectors("pt")
reference_val = binary_crossentropy(sigmoid(pred), target)
f_reference = theano.function(inputs, reference_val)
......@@ -1845,8 +1838,8 @@ def test_confusion_matrix():
conf_mat = np.asarray(conf_mat)
return [conf_mat, order]
x = tensor.vector()
y = tensor.vector()
x = tt.vector()
y = tt.vector()
f = theano.function([x, y], confusion_matrix(x, y))
list_inputs = [
[[0, 1, 2, 1, 0], [0, 0, 2, 1, 2]],
......
tests/tensor/nnet/test_sigm.py
浏览文件 @ e103e7fb
import numpy as np
import theano.tensor.inplace
from theano import tensor as T, config
from theano.tensor import basic as tensor
import theano
import theano.tensor as tt
from theano import config
from theano.tensor.inplace import neg_inplace
from theano.gof.opt import check_stack_trace
from theano.gof.toolbox import is_same_graph
from theano.tensor.nnet import (
......@@ -19,6 +21,7 @@ from theano.tensor.nnet.sigm import (
perform_sigm_times_exp,
register_local_1msigmoid,
simplify_mul,
ScalarSoftplus,
)
from tests import unittest_tools as utt
......@@ -124,7 +127,7 @@ class TestSigmoidOpts:
"""
if excluding is None:
excluding = []
m = theano.config.mode
m = config.mode
if m == "FAST_COMPILE":
mode = theano.compile.mode.get_mode("FAST_RUN")
else:
......@@ -137,68 +140,68 @@ class TestSigmoidOpts:
def test_exp_over_1_plus_exp(self):
m = self.get_mode(excluding=["local_elemwise_fusion"])
x = T.vector()
x = tt.vector()
data = np.random.rand(54).astype(config.floatX)
backup = config.warn.identify_1pexp_bug
config.warn.identify_1pexp_bug = False
try:
# tests exp_over_1_plus_exp
f = theano.function([x], T.exp(x) / (1 + T.exp(x)), mode=m)
f = theano.function([x], tt.exp(x) / (1 + tt.exp(x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] == [sigmoid]
f(data)
f = theano.function([x], T.exp(x) / (2 + T.exp(x)), mode=m)
f = theano.function([x], tt.exp(x) / (2 + tt.exp(x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [sigmoid]
f(data)
f = theano.function([x], T.exp(x) / (1 - T.exp(x)), mode=m)
f = theano.function([x], tt.exp(x) / (1 - tt.exp(x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [sigmoid]
f(data)
f = theano.function([x], T.exp(x + 1) / (1 + T.exp(x)), mode=m)
f = theano.function([x], tt.exp(x + 1) / (1 + tt.exp(x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [sigmoid]
f(data)
# tests inv_1_plus_exp
f = theano.function([x], T.fill(x, 1.0) / (1 + T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, 1.0) / (1 + tt.exp(-x)), mode=m)
# todo: solve issue #4589 first
# assert check_stack_trace(f, ops_to_check=sigmoid)
assert [node.op for node in f.maker.fgraph.toposort()] == [sigmoid]
f(data)
f = theano.function([x], T.fill(x, 1.0) / (2 + T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, 1.0) / (2 + tt.exp(-x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [sigmoid]
f(data)
f = theano.function([x], T.fill(x, 1.0) / (1 - T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, 1.0) / (1 - tt.exp(-x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [sigmoid]
f(data)
f = theano.function([x], T.fill(x, 1.1) / (1 + T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, 1.1) / (1 + tt.exp(-x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [sigmoid]
f(data)
# tests inv_1_plus_exp with neg
f = theano.function([x], T.fill(x, -1.0) / (1 + T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, -1.0) / (1 + tt.exp(-x)), mode=m)
# todo: solve issue #4589 first
# assert check_stack_trace(
# f, ops_to_check=[sigmoid, theano.tensor.inplace.neg_inplace])
# f, ops_to_check=[sigmoid, neg_inplace])
assert [node.op for node in f.maker.fgraph.toposort()] == [
sigmoid,
theano.tensor.inplace.neg_inplace,
neg_inplace,
]
f(data)
f = theano.function([x], T.fill(x, -1.0) / (1 - T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, -1.0) / (1 - tt.exp(-x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
theano.tensor.inplace.neg_inplace,
neg_inplace,
]
f(data)
f = theano.function([x], T.fill(x, -1.0) / (2 + T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, -1.0) / (2 + tt.exp(-x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
theano.tensor.inplace.neg_inplace,
neg_inplace,
]
f(data)
f = theano.function([x], T.fill(x, -1.1) / (1 + T.exp(-x)), mode=m)
f = theano.function([x], tt.fill(x, -1.1) / (1 + tt.exp(-x)), mode=m)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
theano.tensor.inplace.neg_inplace,
neg_inplace,
]
f(data)
......@@ -208,66 +211,66 @@ class TestSigmoidOpts:
# = - (sigm(x) * sigm(x))
f = theano.function(
[x],
(T.fill(x, -1.0) * T.exp(x)) / ((1 + T.exp(x)) * (1 + T.exp(-x))),
(tt.fill(x, -1.0) * tt.exp(x)) / ((1 + tt.exp(x)) * (1 + tt.exp(-x))),
mode=m,
)
# todo: solve issue #4589 first
# assert check_stack_trace(f, ops_to_check=[sigmoid, T.mul])
assert [node.op for node in f.maker.fgraph.toposort()] == [sigmoid, T.mul]
# assert check_stack_trace(f, ops_to_check=[sigmoid, tt.mul])
assert [node.op for node in f.maker.fgraph.toposort()] == [sigmoid, tt.mul]
f(data)
f = theano.function(
[x],
(T.fill(x, -1.1) * T.exp(x)) / ((1 + T.exp(x)) * (1 + T.exp(-x))),
(tt.fill(x, -1.1) * tt.exp(x)) / ((1 + tt.exp(x)) * (1 + tt.exp(-x))),
mode=m,
)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
T.mul,
theano.tensor.inplace.neg_inplace,
tt.mul,
neg_inplace,
]
f(data)
f = theano.function(
[x],
(T.fill(x, -1.0) * T.exp(x)) / ((2 + T.exp(x)) * (1 + T.exp(-x))),
(tt.fill(x, -1.0) * tt.exp(x)) / ((2 + tt.exp(x)) * (1 + tt.exp(-x))),
mode=m,
)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
T.mul,
theano.tensor.inplace.neg_inplace,
tt.mul,
neg_inplace,
]
f(data)
f = theano.function(
[x],
(T.fill(x, -1.0) * T.exp(x)) / ((1 + T.exp(x)) * (2 + T.exp(-x))),
(tt.fill(x, -1.0) * tt.exp(x)) / ((1 + tt.exp(x)) * (2 + tt.exp(-x))),
mode=m,
)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
T.mul,
theano.tensor.inplace.neg_inplace,
tt.mul,
neg_inplace,
]
f(data)
f = theano.function(
[x],
(T.fill(x, -1.0) * T.exp(x)) / ((1 + T.exp(x)) * (1 + T.exp(x))),
(tt.fill(x, -1.0) * tt.exp(x)) / ((1 + tt.exp(x)) * (1 + tt.exp(x))),
mode=m,
)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
T.mul,
theano.tensor.inplace.neg_inplace,
tt.mul,
neg_inplace,
]
f(data)
f = theano.function(
[x],
(T.fill(x, -1.0) * T.exp(x)) / ((1 + T.exp(x)) * (2 + T.exp(-x))),
(tt.fill(x, -1.0) * tt.exp(x)) / ((1 + tt.exp(x)) * (2 + tt.exp(-x))),
mode=m,
)
assert [node.op for node in f.maker.fgraph.toposort()] != [
sigmoid,
T.mul,
theano.tensor.inplace.neg_inplace,
tt.mul,
neg_inplace,
]
f(data)
......@@ -280,21 +283,21 @@ class TestSigmoidOpts:
return
m = self.get_mode()
x = T.fmatrix()
x = tt.fmatrix()
# tests exp_over_1_plus_exp
f = theano.function([x], 1 - T.exp(x) / (1 + T.exp(x)), mode=m)
assert check_stack_trace(f, ops_to_check=[tensor.neg, sigmoid_inplace])
f = theano.function([x], 1 - tt.exp(x) / (1 + tt.exp(x)), mode=m)
assert check_stack_trace(f, ops_to_check=[tt.neg, sigmoid_inplace])
assert [node.op for node in f.maker.fgraph.toposort()] == [
tensor.neg,
tt.neg,
sigmoid_inplace,
]
# tests inv_1_plus_exp
f = theano.function([x], 1 - T.fill(x, 1.0) / (1 + T.exp(-x)), mode=m)
assert check_stack_trace(f, ops_to_check=[tensor.neg, sigmoid_inplace])
f = theano.function([x], 1 - tt.fill(x, 1.0) / (1 + tt.exp(-x)), mode=m)
assert check_stack_trace(f, ops_to_check=[tt.neg, sigmoid_inplace])
assert [node.op for node in f.maker.fgraph.toposort()] == [
tensor.neg,
tt.neg,
sigmoid_inplace,
]
......@@ -308,36 +311,30 @@ class TestSigmoidOpts:
assert [node.op for node in func.maker.fgraph.toposort()] == ops
m = self.get_mode(excluding=["local_elemwise_fusion", "inplace"])
x, y = tensor.vectors("x", "y")
x, y = tt.vectors("x", "y")
f = theano.function([x], sigmoid(-x) * tensor.exp(x), mode=m)
f = theano.function([x], sigmoid(-x) * tt.exp(x), mode=m)
match(f, [sigmoid])
assert check_stack_trace(f, ops_to_check=sigmoid)
f = theano.function([x], sigmoid(x) * tensor.exp(-x), mode=m)
match(f, [tensor.neg, sigmoid])
f = theano.function([x], sigmoid(x) * tt.exp(-x), mode=m)
match(f, [tt.neg, sigmoid])
assert check_stack_trace(f, ops_to_check=sigmoid)
f = theano.function([x], -(-(-(sigmoid(x)))) * tensor.exp(-x), mode=m)
match(f, [tensor.neg, sigmoid, tensor.neg])
f = theano.function([x], -(-(-(sigmoid(x)))) * tt.exp(-x), mode=m)
match(f, [tt.neg, sigmoid, tt.neg])
# assert check_stack_trace(f, ops_to_check=sigmoid)
f = theano.function(
[x, y],
(
sigmoid(x)
* sigmoid(-y)
* -tensor.exp(-x)
* tensor.exp(x * y)
* tensor.exp(y)
),
(sigmoid(x) * sigmoid(-y) * -tt.exp(-x) * tt.exp(x * y) * tt.exp(y)),
mode=m,
)
topo = f.maker.fgraph.toposort()
for op, nb in [(sigmoid, 2), (tensor.mul, 2), (tensor.neg, 1), (tensor.exp, 1)]:
for op, nb in [(sigmoid, 2), (tt.mul, 2), (tt.neg, 1), (tt.exp, 1)]:
assert sum([n.op == op for n in topo]) == nb
# assert check_stack_trace(f, ops_to_check=[sigmoid, tensor.mul,
# tensor.exp])
# assert check_stack_trace(f, ops_to_check=[sigmoid, tt.mul,
# tt.exp])
def test_perform_sigm_times_exp(self):
# Test the core function doing the `sigm_times_exp` optimization.
......@@ -345,8 +342,8 @@ class TestSigmoidOpts:
# It is easier to test different graph scenarios this way than by
# compiling a theano function.
x, y, z, t = tensor.vectors("x", "y", "z", "t")
exp = tensor.exp
x, y, z, t = tt.vectors("x", "y", "z", "t")
exp = tt.exp
def ok(expr1, expr2):
trees = [parse_mul_tree(e) for e in (expr1, expr2)]
......@@ -386,11 +383,11 @@ class TestSigmoidOpts:
# At some point, this returned nan, because (1 - sigm(x)) was
# on both the numerator and the denominator of a fraction,
# but the two nodes in question had not been merged.
x = tensor.matrix("x")
lr = tensor.scalar("lr")
x = tt.matrix("x")
lr = tt.scalar("lr")
s = sigmoid(x)
l = T.log(1 - s)
l = tt.log(1 - s)
c = l.mean()
ux = x - lr * theano.grad(c, x)
......@@ -403,7 +400,7 @@ class TestSigmoidOpts:
assert not np.isnan(ux_v)
def test_local_ultra_fast_sigmoid(self):
x = tensor.matrix("x")
x = tt.matrix("x")
s = sigmoid(x)
mode = self.get_mode("local_ultra_fast_sigmoid")
......@@ -422,7 +419,7 @@ class TestSigmoidOpts:
f([[-50, -10, -4, -1, 0, 1, 4, 10, 50]])
def test_local_hard_sigmoid(self):
x = tensor.matrix("x")
x = tt.matrix("x")
s = sigmoid(x)
mode = self.get_mode("local_hard_sigmoid")
......@@ -440,7 +437,7 @@ class TestSigmoidOpts:
mode2 = mode.excluding("fusion").excluding("inplace")
f2 = theano.function([x], s, mode=mode2)
assert check_stack_trace(f2, ops_to_check=theano.tensor.clip)
assert check_stack_trace(f2, ops_to_check=tt.clip)
class TestSoftplusOpts:
......@@ -457,56 +454,56 @@ class TestSoftplusOpts:
utt.seed_rng()
def test_logsigm_to_softplus(self):
x = T.vector()
x = tt.vector()
out = T.log(sigmoid(x))
out = tt.log(sigmoid(x))
f = theano.function([x], out, mode=self.m)
# Fix ticket #4581 first
# assert check_stack_trace(
# f, ops_to_check=(theano.scalar.Neg,
# theano.tensor.nnet.sigm.ScalarSoftplus))
# ScalarSoftplus))
topo = f.maker.fgraph.toposort()
assert len(topo) == 3
assert isinstance(topo[0].op.scalar_op, theano.scalar.Neg)
assert isinstance(topo[1].op.scalar_op, theano.tensor.nnet.sigm.ScalarSoftplus)
assert isinstance(topo[1].op.scalar_op, ScalarSoftplus)
assert isinstance(topo[2].op.scalar_op, theano.scalar.Neg)
f(np.random.rand(54).astype(config.floatX))
def test_log1msigm_to_softplus(self):
x = T.matrix()
x = tt.matrix()
out = T.log(1 - sigmoid(x))
out = tt.log(1 - sigmoid(x))
f = theano.function([x], out, mode=self.m)
topo = f.maker.fgraph.toposort()
assert len(topo) == 2
assert isinstance(topo[0].op.scalar_op, theano.tensor.nnet.sigm.ScalarSoftplus)
assert isinstance(topo[0].op.scalar_op, ScalarSoftplus)
assert isinstance(topo[1].op.scalar_op, theano.scalar.Neg)
# assert check_stack_trace(f, ops_to_check='all')
f(np.random.rand(54, 11).astype(config.floatX))
# Same test with a flatten
out = T.log(1 - T.flatten(sigmoid(x)))
out = tt.log(1 - tt.flatten(sigmoid(x)))
f = theano.function([x], out, mode=self.m)
# assert check_stack_trace(f, ops_to_check='all')
topo = f.maker.fgraph.toposort()
assert len(topo) == 3
assert tensor.is_flat(topo[0].outputs[0])
assert isinstance(topo[1].op.scalar_op, theano.tensor.nnet.sigm.ScalarSoftplus)
assert tt.is_flat(topo[0].outputs[0])
assert isinstance(topo[1].op.scalar_op, ScalarSoftplus)
assert isinstance(topo[2].op.scalar_op, theano.scalar.Neg)
f(np.random.rand(54, 11).astype(config.floatX))
# Same test with a reshape
out = T.log(1 - sigmoid(x).reshape([x.size]))
out = tt.log(1 - sigmoid(x).reshape([x.size]))
f = theano.function([x], out, mode=self.m)
topo = f.maker.fgraph.toposort()
# assert len(topo) == 3
assert any(isinstance(node.op, T.Reshape) for node in topo)
assert any(isinstance(node.op, tt.Reshape) for node in topo)
assert any(
isinstance(
getattr(node.op, "scalar_op", None),
theano.tensor.nnet.sigm.ScalarSoftplus,
ScalarSoftplus,
)
for node in topo
)
......@@ -517,16 +514,16 @@ class TestSoftplusOpts:
if m == "FAST_COMPILE":
m = "FAST_RUN"
x = T.vector()
x = tt.vector()
out = T.log(1 + T.exp(x))
out = tt.log(1 + tt.exp(x))
f = theano.function([x], out, mode=self.m)
# Fix ticket #4581 first
# assert check_stack_trace(f, ops_to_check='all')
topo = f.maker.fgraph.toposort()
assert len(topo) == 1
assert isinstance(topo[0].op.scalar_op, theano.tensor.nnet.sigm.ScalarSoftplus)
assert isinstance(topo[0].op.scalar_op, ScalarSoftplus)
f(np.random.rand(54).astype(config.floatX))
......@@ -536,14 +533,14 @@ class TestSigmoidUtils:
"""
def test_compute_mul(self):
x, y, z = tensor.vectors("x", "y", "z")
x, y, z = tt.vectors("x", "y", "z")
tree = (x * y) * -z
mul_tree = parse_mul_tree(tree)
assert parse_mul_tree(compute_mul(mul_tree)) == mul_tree
assert is_same_graph(compute_mul(parse_mul_tree(tree)), tree)
def test_parse_mul_tree(self):
x, y, z = tensor.vectors("x", "y", "z")
x, y, z = tt.vectors("x", "y", "z")
assert parse_mul_tree(x * y) == [False, [[False, x], [False, y]]]
assert parse_mul_tree(-(x * y)) == [True, [[False, x], [False, y]]]
assert parse_mul_tree(-x * y) == [False, [[True, x], [False, y]]]
......@@ -557,8 +554,8 @@ class TestSigmoidUtils:
backup = config.warn.identify_1pexp_bug
config.warn.identify_1pexp_bug = False
try:
x = tensor.vector("x")
exp = tensor.exp
x = tt.vector("x")
exp = tt.exp
assert is_1pexp(1 + exp(x), False) == (False, x)
assert is_1pexp(exp(x) + 1, False) == (False, x)
for neg, exp_arg in map(
......
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