Skip to content

P
pytensor
Unverified 提交 9fcb536e authored 作者: George Ho's avatar George Ho 提交者： GitHub
浏览文件
操作

Remove theano.dot (#149)

This replaces `theano.tensor.basic.dot` with the old `theano.dot`, removes the latter, and renames `theano.tensor.basic.dot` to `theano.tensor.basic.dense_dot`.
上级 4249a01c
隐藏空白字符变更
内嵌 并排
正在显示 包含 130 行增加 和 107 行删除
tests/compile/test_pfunc.py
浏览文件 @ 9fcb536e
......@@ -786,7 +786,7 @@ class TestAliasingRules:
theano.In(m1, mutable=True),
theano.In(m2, mutable=True),
],
theano.dot((x * 2), m1) + theano.dot((y * 3), m2),
theano.tensor.dot((x * 2), m1) + theano.tensor.dot((y * 3), m2),
)
# Test 1. If the same variable is given twice
......@@ -851,9 +851,9 @@ class TestAliasingRules:
theano.In(m3, mutable=True),
],
(
theano.dot((x * 2), m1)
+ theano.dot((y * 3), m2)
+ theano.dot((z * 4), m3)
theano.tensor.dot((x * 2), m1)
+ theano.tensor.dot((y * 3), m2)
+ theano.tensor.dot((z * 4), m3)
),
)
......
tests/scan_module/test_scan.py
浏览文件 @ 9fcb536e
......@@ -485,8 +485,10 @@ class TestScan:
def f_rnn_cmpl(u1_t, u2_t, x_tm1, y_tm1, W_in1):
return [
theano.dot(u1_t, W_in1) + u2_t * W_in2 + theano.dot(x_tm1, W),
theano.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1)
+ u2_t * W_in2
+ theano.tensor.dot(x_tm1, W),
theano.tensor.dot(x_tm1, W_out),
]
outputs, updates = theano.scan(
......@@ -539,11 +541,11 @@ class TestScan:
def f_rnn_cmpl(u1_t, u2_tm1, u2_t, u2_tp1, x_tm1, y_tm1, y_tm3, W_in1):
return [
theano.dot(u1_t, W_in1)
theano.tensor.dot(u1_t, W_in1)
+ (u2_t + u2_tm1 * u2_tp1) * W_in2
+ theano.dot(x_tm1, W),
(y_tm1 + y_tm3) * theano.dot(x_tm1, W_out),
theano.dot(u1_t, W_in1),
+ theano.tensor.dot(x_tm1, W),
(y_tm1 + y_tm3) * theano.tensor.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1),
]
outputs, updates = theano.scan(
......@@ -1107,13 +1109,13 @@ class TestScan:
def f(u1_t, u2_t, y0_tm3, y0_tm2, y0_tm1, y1_tm1):
y0_t = (
theano.dot(theano.dot(u1_t, W1), W2)
theano.tensor.dot(theano.tensor.dot(u1_t, W1), W2)
+ 0.1 * y0_tm1
+ 0.33 * y0_tm2
+ 0.17 * y0_tm3
)
y1_t = theano.dot(u2_t, W2) + y1_tm1
y2_t = theano.dot(u1_t, W1)
y1_t = theano.tensor.dot(u2_t, W2) + y1_tm1
y2_t = theano.tensor.dot(u1_t, W1)
nwW1 = W1 + 0.1
nwW2 = W2 + 0.05
# return outputs followed by a list of updates
......@@ -1250,11 +1252,15 @@ class TestScan:
trng = theano.tensor.shared_randomstreams.RandomStreams(utt.fetch_seed())
def f(vsample_tm1):
hmean_t = theano.tensor.nnet.sigmoid(theano.dot(vsample_tm1, W) + bhid)
hmean_t = theano.tensor.nnet.sigmoid(
theano.tensor.dot(vsample_tm1, W) + bhid
)
hsample_t = theano.tensor.cast(
trng.binomial(hmean_t.shape, 1, hmean_t), dtype="float32"
)
vmean_t = theano.tensor.nnet.sigmoid(theano.dot(hsample_t, W.T) + bvis)
vmean_t = theano.tensor.nnet.sigmoid(
theano.tensor.dot(hsample_t, W.T) + bvis
)
return theano.tensor.cast(
trng.binomial(vmean_t.shape, 1, vmean_t), dtype="float32"
)
......@@ -1463,8 +1469,10 @@ class TestScan:
def f_rnn_cmpl(u1_t, u2_t, x_tm1, y_tm1, W_in1):
return [
theano.dot(u1_t, W_in1) + u2_t * W_in2 + theano.dot(x_tm1, W),
theano.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1)
+ u2_t * W_in2
+ theano.tensor.dot(x_tm1, W),
theano.tensor.dot(x_tm1, W_out),
]
cost, updates = scan_project_sum(
......@@ -1532,11 +1540,11 @@ class TestScan:
def f_rnn_cmpl(u1_t, u2_tm1, u2_t, u2_tp1, x_tm1, y_tm1, y_tm3, W_in1):
return [
theano.dot(u1_t, W_in1)
theano.tensor.dot(u1_t, W_in1)
+ (u2_t + u2_tm1 * u2_tp1) * W_in2
+ theano.dot(x_tm1, W),
(y_tm1 + y_tm3) * theano.dot(x_tm1, W_out),
theano.dot(u1_t, W_in1),
+ theano.tensor.dot(x_tm1, W),
(y_tm1 + y_tm3) * theano.tensor.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1),
]
# We change the compute_test_value[_opt] flag to run the
......@@ -1795,10 +1803,10 @@ for{cpu,scan_fn}.2 [id H] ''
def f_rnn_cmpl(u1_t, u2_tm1, u2_t, u2_tp1, x_tm1, y_tm1, y_tm3, W_in1):
return [
theano.dot(u1_t, W_in1)
theano.tensor.dot(u1_t, W_in1)
+ (u2_t + u2_tm1 * u2_tp1) * W_in2
+ theano.dot(x_tm1, W),
(y_tm1 + y_tm3) * theano.dot(x_tm1, W_out),
+ theano.tensor.dot(x_tm1, W),
(y_tm1 + y_tm3) * theano.tensor.dot(x_tm1, W_out),
]
cost, updates = scan_project_sum(
......@@ -1853,7 +1861,7 @@ for{cpu,scan_fn}.2 [id H] ''
trng1 = theano.tensor.shared_randomstreams.RandomStreams(123)
x_t = (
theano.tensor.cast(u2_t, theano.config.floatX)
+ theano.dot(u_t, W_in)
+ theano.tensor.dot(u_t, W_in)
+ x_tm1
+ trng1.uniform(low=-1.1, high=1.1, dtype=theano.config.floatX)
)
......@@ -1935,7 +1943,7 @@ for{cpu,scan_fn}.2 [id H] ''
def f_rnn_cmpl(u_t, x_tm1, W_in):
trng1 = theano.tensor.shared_randomstreams.RandomStreams(123)
rnd_nb = trng1.uniform(low=-0.1, high=0.1)
x_t = theano.dot(u_t, W_in) + x_tm1 + rnd_nb
x_t = theano.tensor.dot(u_t, W_in) + x_tm1 + rnd_nb
x_t = theano.tensor.cast(x_t, dtype=theano.config.floatX)
return x_t
......@@ -2026,8 +2034,10 @@ for{cpu,scan_fn}.2 [id H] ''
# prior results: h_tm2, h_tm1
# non-sequences: W_ih, W_hh, W_ho, b_h
def one_step(x_t, h_tm2, h_tm1, W_ih, W_hh, b_h, W_ho, b_o):
h_t = tensor.tanh(theano.dot(x_t, W_ih) + theano.dot(h_tm2, W_hh) + b_h)
y_t = theano.dot(h_t, W_ho) + b_o
h_t = tensor.tanh(
theano.tensor.dot(x_t, W_ih) + theano.tensor.dot(h_tm2, W_hh) + b_h
)
y_t = theano.tensor.dot(h_t, W_ho) + b_o
return [h_t, y_t]
# hidden and outputs of the entire sequence
......@@ -2181,7 +2191,7 @@ for{cpu,scan_fn}.2 [id H] ''
A = theano.tensor.matrix("A")
fc1 = theano.shared(0.5, name="fc1")
fc2 = theano.shared(0.9, name="fc2")
y = fc1 * theano.dot(x * x, theano.dot(A, x))
y = fc1 * theano.tensor.dot(x * x, theano.tensor.dot(A, x))
y.name = "y"
gy = theano.tensor.grad(y, x)
gy.name = "gy"
......@@ -2326,8 +2336,10 @@ for{cpu,scan_fn}.2 [id H] ''
return [
y_tm3 + 1,
y_tm3 + 2,
theano.dot(u1_t, W_in1) + u2_t * W_in2 + theano.dot(x_tm1, W),
y_tm1 + theano.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1)
+ u2_t * W_in2
+ theano.tensor.dot(x_tm1, W),
y_tm1 + theano.tensor.dot(x_tm1, W_out),
]
outputs, updates = theano.scan(
......@@ -2407,8 +2419,10 @@ for{cpu,scan_fn}.2 [id H] ''
def f_rnn_cmpl(u1_t, u2_t, x_tm1, y_tm1, y_tm3, W_in1):
return [
y_tm3 + 1,
theano.dot(u1_t, W_in1) + u2_t * W_in2 + theano.dot(x_tm1, W),
y_tm1 + theano.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1)
+ u2_t * W_in2
+ theano.tensor.dot(x_tm1, W),
y_tm1 + theano.tensor.dot(x_tm1, W_out),
]
_outputs, updates = theano.scan(
......@@ -4022,8 +4036,10 @@ for{cpu,scan_fn}.2 [id H] ''
def f_rnn_cmpl(u1_t, u2_t, x_tm1, y_tm1, y_tm3, W_in1):
return [
y_tm3 + 1,
theano.dot(u1_t, W_in1) + u2_t * W_in2 + theano.dot(x_tm1, W),
y_tm1 + theano.dot(x_tm1, W_out),
theano.tensor.dot(u1_t, W_in1)
+ u2_t * W_in2
+ theano.tensor.dot(x_tm1, W),
y_tm1 + theano.tensor.dot(x_tm1, W_out),
]
rval, updates = theano.scan(
......@@ -4069,7 +4085,7 @@ for{cpu,scan_fn}.2 [id H] ''
A = theano.tensor.matrix("A")
z, updates = theano.scan(
theano.dot, sequences=[], non_sequences=[x, A], n_steps=2
theano.tensor.dot, sequences=[], non_sequences=[x, A], n_steps=2
)
f = theano.function([x, A], z)
topo = f.maker.fgraph.toposort()
......@@ -5603,7 +5619,7 @@ def test_compute_test_value_grad_cast():
w = theano.shared(np.random.randn(4, 3).astype(floatX), name="w")
outputs, _ = theano.scan(
lambda i, h, w: (theano.dot(h[i], w), i),
lambda i, h, w: (theano.tensor.dot(h[i], w), i),
outputs_info=[None, 0],
non_sequences=[h, w],
n_steps=3,
......
tests/sparse/test_basic.py
浏览文件 @ 9fcb536e
......@@ -535,7 +535,7 @@ class TestSparseInferShape(utt.InferShapeTester):
(tensor.matrix(), SparseType("csr", "float32")()),
]:
sparse_out = theano.dot(x, y)
sparse_out = tensor.dot(x, y)
if isinstance(x, sparse.SparseVariable):
x = tensor.matrix()
if isinstance(y, sparse.SparseVariable):
......@@ -1342,7 +1342,7 @@ class TestStructuredDot:
for sparse_format_b in ["csc", "csr", "bsr"]:
a = SparseType(sparse_format_a, dtype=sparse_dtype)()
b = SparseType(sparse_format_b, dtype=sparse_dtype)()
d = theano.dot(a, b)
d = theano.tensor.dot(a, b)
f = theano.function([a, b], theano.Out(d, borrow=True))
for M, N, K, nnz in [
(4, 3, 2, 3),
......@@ -1364,7 +1364,7 @@ class TestStructuredDot:
a = SparseType("csc", dtype=sparse_dtype)()
b = tensor.matrix(dtype=dense_dtype)
d = theano.dot(a, b)
d = theano.tensor.dot(a, b)
f = theano.function([a, b], theano.Out(d, borrow=True))
for M, N, K, nnz in [
......@@ -1412,7 +1412,7 @@ class TestStructuredDot:
a = SparseType("csr", dtype=sparse_dtype)()
b = tensor.matrix(dtype=dense_dtype)
d = theano.dot(a, b)
d = theano.tensor.dot(a, b)
f = theano.function([a, b], d)
for M, N, K, nnz in [
......
tests/tensor/test_basic.py
浏览文件 @ 9fcb536e
......@@ -114,9 +114,9 @@ from theano.tensor import (
constant,
cscalar,
default,
dense_dot,
diag,
dmatrix,
dot,
dscalar,
dscalars,
dtensor3,
......@@ -708,7 +708,7 @@ TestConjBroadcast = makeBroadcastTester(
TestDot = makeTester(
name="DotTester",
op=dot,
op=dense_dot,
expected=lambda x, y: np.dot(x, y),
checks={},
good=dict(
......@@ -1140,7 +1140,7 @@ class TestAlloc:
(some_matrix[idx, idx], 1),
],
):
derp = sum(dot(subtensor, variables))
derp = sum(dense_dot(subtensor, variables))
fobj = theano.function([some_vector], derp, mode=self.mode)
grad_derp = theano.grad(derp, some_vector)
......@@ -3659,7 +3659,7 @@ class TestDot:
return type(x), x.dtype, x.shape
nz = np.dot(x, y)
tz = eval_outputs([dot(as_tensor_variable(x), as_tensor_variable(y))])
tz = eval_outputs([dense_dot(as_tensor_variable(x), as_tensor_variable(y))])
assert tz.dtype == nz.dtype, (tz.dtype, tz.dtype.num, nz.dtype, nz.dtype.num)
assert tz.shape == nz.shape, (tz.shape, nz.shape)
utt.assert_allclose(nz, tz, rtol=1e-4, atol=1e-4)
......@@ -3797,7 +3797,7 @@ class TestDot:
def not_aligned(self, x, y):
with change_flags(compute_test_value="off"):
z = dot(x, y)
z = dense_dot(x, y)
with pytest.raises(ValueError):
eval_outputs([z])
......@@ -3813,19 +3813,19 @@ class TestDot:
self.not_aligned(rand(5, 4, 3), rand(6, 7, 8))
def test_grad(self):
utt.verify_grad(dot, [rand(2, 3), rand(3, 2)])
utt.verify_grad(dot, [rand(2), rand(2, 3)])
utt.verify_grad(dot, [rand(3, 2), rand(2)])
utt.verify_grad(dot, [rand(2), rand(2)])
utt.verify_grad(dot, [rand(), rand(2)])
utt.verify_grad(dot, [rand(), rand(2, 5)])
utt.verify_grad(dot, [rand(2), rand()])
utt.verify_grad(dot, [rand(2, 5), rand()])
utt.verify_grad(dot, [rand(2, 3, 4), rand(4)])
utt.verify_grad(dot, [rand(3), rand(2, 3, 4)])
utt.verify_grad(dot, [rand(4, 3), rand(2, 3, 4)])
utt.verify_grad(dot, [rand(2, 3, 4), rand(4, 5)])
utt.verify_grad(dot, [rand(2, 3, 4), rand(3, 4, 5)])
utt.verify_grad(dense_dot, [rand(2, 3), rand(3, 2)])
utt.verify_grad(dense_dot, [rand(2), rand(2, 3)])
utt.verify_grad(dense_dot, [rand(3, 2), rand(2)])
utt.verify_grad(dense_dot, [rand(2), rand(2)])
utt.verify_grad(dense_dot, [rand(), rand(2)])
utt.verify_grad(dense_dot, [rand(), rand(2, 5)])
utt.verify_grad(dense_dot, [rand(2), rand()])
utt.verify_grad(dense_dot, [rand(2, 5), rand()])
utt.verify_grad(dense_dot, [rand(2, 3, 4), rand(4)])
utt.verify_grad(dense_dot, [rand(3), rand(2, 3, 4)])
utt.verify_grad(dense_dot, [rand(4, 3), rand(2, 3, 4)])
utt.verify_grad(dense_dot, [rand(2, 3, 4), rand(4, 5)])
utt.verify_grad(dense_dot, [rand(2, 3, 4), rand(3, 4, 5)])
@pytest.mark.slow
def test_broadcastable_patterns(self):
......@@ -3882,7 +3882,7 @@ class TestDot:
):
y = TensorType(dtype=dtype1, broadcastable=bc1)()
z = dot(x, y)
z = dense_dot(x, y)
t = TensorType(dtype=dtype0, broadcastable=z.broadcastable)()
rval = z * 3 + 2 * t
......
tests/tensor/test_blas.py
浏览文件 @ 9fcb536e
......@@ -1254,7 +1254,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
rng = np.random.RandomState(unittest_tools.fetch_seed())
v = theano.shared(np.array(rng.uniform(size=(2,)), dtype="float32"))
w = theano.shared(np.array(rng.uniform(size=(2,)), dtype="float32"))
f = theano.function([], theano.dot(v, w), mode=mode_blas_opt)
f = theano.function([], theano.tensor.dot(v, w), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
self.assertFunctionContains0(f, tt.dot)
......@@ -1268,7 +1268,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
rng = np.random.RandomState(unittest_tools.fetch_seed())
v = theano.shared(np.array(rng.uniform(size=(2,)), dtype="float32"))
m = theano.shared(np.array(rng.uniform(size=(2, 3)), dtype="float32"))
f = theano.function([], theano.dot(v, m), mode=mode_blas_opt)
f = theano.function([], theano.tensor.dot(v, m), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
self.assertFunctionContains0(f, tt.dot)
......@@ -1285,7 +1285,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
rng = np.random.RandomState(unittest_tools.fetch_seed())
v = theano.shared(np.array(rng.uniform(size=(2,)), dtype="float32"))
m = theano.shared(np.array(rng.uniform(size=(3, 2)), dtype="float32"))
f = theano.function([], theano.dot(m, v), mode=mode_blas_opt)
f = theano.function([], theano.tensor.dot(m, v), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
self.assertFunctionContains0(f, tt.dot)
......@@ -1306,7 +1306,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
v2 = theano.shared(v2_orig)
m = theano.shared(np.array(rng.uniform(size=m_shp), dtype="float32"))
f = theano.function([], v2 + theano.dot(m, v1), mode=mode_blas_opt)
f = theano.function([], v2 + theano.tensor.dot(m, v1), mode=mode_blas_opt)
# Assert they produce the same output
assert np.allclose(f(), np.dot(m.get_value(), v1.get_value()) + v2_orig)
......@@ -1317,7 +1317,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
# test the inplace version
g = theano.function(
[], [], updates=[(v2, v2 + theano.dot(m, v1))], mode=mode_blas_opt
[], [], updates=[(v2, v2 + theano.tensor.dot(m, v1))], mode=mode_blas_opt
)
# Assert they produce the same output
......@@ -1355,7 +1355,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
v2 = theano.shared(v2_orig)
m = theano.shared(np.array(rng.uniform(size=(2, 3)), dtype="float32"))
f = theano.function([], v2 + theano.dot(v1, m), mode=mode_blas_opt)
f = theano.function([], v2 + theano.tensor.dot(v1, m), mode=mode_blas_opt)
# Assert they produce the same output
assert np.allclose(f(), np.dot(v1.get_value(), m.get_value()) + v2.get_value())
......@@ -1365,7 +1365,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
# test the inplace version
g = theano.function(
[], [], updates=[(v2, v2 + theano.dot(v1, m))], mode=mode_blas_opt
[], [], updates=[(v2, v2 + theano.tensor.dot(v1, m))], mode=mode_blas_opt
)
# Assert they produce the same output
......@@ -1397,7 +1397,7 @@ class TestGemv(unittest_tools.OptimizationTestMixin):
np.array(rng.uniform(size=(1, 2)), dtype="float32"),
broadcastable=(True, False),
)
o = theano.dot(m, v1)
o = theano.tensor.dot(m, v1)
f = theano.function([], o + v2, mode=mode_blas_opt)
# Assert they produce the same output
......
tests/tensor/test_blas_c.py
浏览文件 @ 9fcb536e
......@@ -147,7 +147,7 @@ class TestCGemv(OptimizationTestMixin):
mode.check_isfinite = False
f = theano.function(
[self.A, self.x, self.y, self.a],
self.a * self.y + theano.dot(self.A, self.x),
self.a * self.y + theano.tensor.dot(self.A, self.x),
mode=mode,
)
Aval = np.ones((3, 1), dtype=self.dtype)
......@@ -160,7 +160,7 @@ class TestCGemv(OptimizationTestMixin):
skip_if_blas_ldflags_empty()
""" Test vector dot matrix """
f = theano.function(
[self.x, self.A], theano.dot(self.x, self.A), mode=self.mode
[self.x, self.A], theano.tensor.dot(self.x, self.A), mode=self.mode
)
# Assert that the dot was optimized somehow
......@@ -180,7 +180,7 @@ class TestCGemv(OptimizationTestMixin):
skip_if_blas_ldflags_empty()
""" Test matrix dot vector """
f = theano.function(
[self.A, self.y], theano.dot(self.A, self.y), mode=self.mode
[self.A, self.y], theano.tensor.dot(self.A, self.y), mode=self.mode
)
# Assert that the dot was optimized somehow
......@@ -220,7 +220,7 @@ class TestCGemv(OptimizationTestMixin):
# test the inplace version
g = theano.function(
[], [], updates=[(v2, v2 + theano.dot(m, v1))], mode=self.mode
[], [], updates=[(v2, v2 + theano.tensor.dot(m, v1))], mode=self.mode
)
# Assert they produce the same output
......
tests/tensor/test_opt.py
浏览文件 @ 9fcb536e
......@@ -3954,14 +3954,14 @@ def test_local_subtensor_of_dot():
return a.shape == b.shape and np.allclose(a, b)
# [cst]
f = theano.function([m1, m2], theano.dot(m1, m2)[1], mode=mode)
f = theano.function([m1, m2], theano.tensor.dot(m1, m2)[1], mode=mode)
topo = f.maker.fgraph.toposort()
assert test_equality(f(d1, d2), np.dot(d1, d2)[1])
# DimShuffle happen in FAST_COMPILE
assert isinstance(topo[-1].op, (CGemv, Gemv, DimShuffle))
# slice
f = theano.function([m1, m2], theano.dot(m1, m2)[1:2], mode=mode)
f = theano.function([m1, m2], theano.tensor.dot(m1, m2)[1:2], mode=mode)
topo = f.maker.fgraph.toposort()
assert test_equality(f(d1, d2), np.dot(d1, d2)[1:2])
assert isinstance(topo[-1].op, Dot22)
......@@ -3972,12 +3972,16 @@ def test_local_subtensor_of_dot():
d1 = np.arange(30).reshape(2, 5, 3).astype(config.floatX)
d2 = np.arange(72).reshape(4, 3, 6).astype(config.floatX) + 100
f = theano.function([m1, m2, idx], theano.dot(m1, m2)[idx, 1:4, :, idx:], mode=mode)
f = theano.function(
[m1, m2, idx], theano.tensor.dot(m1, m2)[idx, 1:4, :, idx:], mode=mode
)
assert test_equality(f(d1, d2, 1), np.dot(d1, d2)[1, 1:4, :, 1:])
# if we return the gradients. We need to use same mode as before.
assert check_stack_trace(f, ops_to_check="last")
f = theano.function([m1, m2, idx], theano.dot(m1, m2)[1:4, :, idx:, idx], mode=mode)
f = theano.function(
[m1, m2, idx], theano.tensor.dot(m1, m2)[1:4, :, idx:, idx], mode=mode
)
assert test_equality(f(d1, d2, 1), np.dot(d1, d2)[1:4, :, 1:, 1])
# Now test that the stack trace is copied over properly,
......
tests/tensor/test_sharedvar.py
浏览文件 @ 9fcb536e
......@@ -533,7 +533,8 @@ def makeSharedTester(
s = self.cast_value(s)
s_shared = self.shared_constructor(s)
f = theano.function(
[], updates=[(s_shared, theano.dot(a_shared, b_shared) + s_shared)]
[],
updates=[(s_shared, theano.tensor.dot(a_shared, b_shared) + s_shared)],
)
topo = f.maker.fgraph.toposort()
f()
......@@ -569,7 +570,9 @@ def makeSharedTester(
f = theano.function(
[],
s_shared.shape,
updates=[(s_shared, theano.dot(a_shared, b_shared) + s_shared_specify)],
updates=[
(s_shared, theano.tensor.dot(a_shared, b_shared) + s_shared_specify)
],
)
topo = f.maker.fgraph.toposort()
shp = f()
......@@ -606,7 +609,9 @@ def makeSharedTester(
f = theano.function(
[],
s_shared.shape,
updates=[(s_shared, theano.dot(a_shared, b_shared) + s_shared_specify)],
updates=[
(s_shared, theano.tensor.dot(a_shared, b_shared) + s_shared_specify)
],
)
topo = f.maker.fgraph.toposort()
shp = f()
......
theano/__init__.py
浏览文件 @ 9fcb536e
......@@ -160,27 +160,6 @@ np.seterr(all=_all, divide=_divide, over=_over, under=_under, invalid=_invalid)
del _all, _divide, _over, _under, _invalid
def dot(l, r):
"""Return a symbolic dot product.
This is designed to work with both sparse and dense tensors types.
"""
try:
res = l.__dot__(r)
if res is NotImplemented:
raise NotImplementedError()
return res
except (NotImplementedError, AttributeError, TypeError):
res = r.__rdot__(l)
if res is NotImplemented:
raise NotImplementedError()
return res
def get_scalar_constant_value(v):
"""Return the constant scalar (i.e. 0-D) value underlying variable `v`.
......
theano/sparse/basic.py
浏览文件 @ 9fcb536e
......@@ -273,6 +273,8 @@ class _sparse_py_operators:
def __rdot__(right, left):
return structured_dot(left, right)
dot = __dot__
# N.B. THIS IS COMMENTED OUT ON PURPOSE!!!
# Discussion with Fred & James (at least, and maybe others before)
# we decided that casting from a sparse to dense should be explicit
......
theano/tensor/basic.py
浏览文件 @ 9fcb536e
......@@ -6306,7 +6306,24 @@ pprint.assign(
)
def dot(a, b):
def dot(l, r):
"""Return a symbolic dot product.
This is designed to work with both sparse and dense tensors types.
"""
try:
res = l.__dot__(r)
if res is NotImplemented:
raise NotImplementedError
except (NotImplementedError, AttributeError, TypeError):
res = r.__rdot__(l)
if res is NotImplemented:
raise NotImplementedError()
return res
def dense_dot(a, b):
"""
Computes the dot product of two variables.
......
theano/tensor/nlinalg.py
浏览文件 @ 9fcb536e
......@@ -687,18 +687,18 @@ def matrix_power(M, n):
return M
elif n == 2:
return theano.dot(M, M)
return theano.tensor.dot(M, M)
elif n == 3:
return theano.dot(theano.dot(M, M), M)
return theano.tensor.dot(theano.tensor.dot(M, M), M)
result = z = None
while n > 0:
z = M if z is None else theano.dot(z, z)
z = M if z is None else theano.tensor.dot(z, z)
n, bit = divmod(n, 2)
if bit:
result = z if result is None else theano.dot(result, z)
result = z if result is None else theano.tensor.dot(result, z)
return result
......
theano/tensor/var.py
浏览文件 @ 9fcb536e
......@@ -661,10 +661,10 @@ class _tensor_py_operators:
"""The dtype of this tensor."""
def __dot__(left, right):
return theano.tensor.basic.dot(left, right)
return theano.tensor.basic.dense_dot(left, right)
def __rdot__(right, left):
return theano.tensor.basic.dot(left, right)
return theano.tensor.basic.dense_dot(left, right)
dot = __dot__
......
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