Skip to content

P
pytensor
提交 a2f6c1c9 authored 作者: erakra's avatar erakra
浏览文件
操作

fixing flake8 for test_blas.py

上级 c5cd87fa
显示空白字符变更
内嵌 并排
正在显示 包含 108 行增加 和 116 行删除
theano/tensor/tests/test_blas.py
浏览文件 @ a2f6c1c9
......@@ -81,7 +81,7 @@ class t_gemm(TestCase):
f = inplace_func([tz, ta, tx, ty, tb],
gemm_inplace(tz, ta, tx, ty, tb),
mode=compile.Mode(optimizer=None, linker=l))
new_z = f(z, a, x, y, b)
f(z, a, x, y, b)
z_after = self._gemm(z_orig, a, x, y, b)
# print z_orig, z_after, z, type(z_orig), type(z_after), type(z)
......@@ -96,8 +96,8 @@ class t_gemm(TestCase):
cmp_linker(copy(z), a, x, y, b, 'c|py')
cmp_linker(copy(z), a, x, y, b, 'py')
if (not dtype.startswith("complex")
and theano.config.cxx):
if (not dtype.startswith("complex") and
theano.config.cxx):
# If theano.config.blas.ldflags is empty, Theano will use
# a NumPy C implementation of [sd]gemm_.
cmp_linker(copy(z), a, x, y, b, 'c')
......@@ -105,7 +105,7 @@ class t_gemm(TestCase):
def test0a(self):
Gemm.debug = True
try:
g = gemm_no_inplace([1.], 1., [1.], [1.], 1.)
gemm_no_inplace([1.], 1., [1.], [1.], 1.)
except TypeError as e:
if exc_message(e) is Gemm.E_rank:
return
......@@ -171,7 +171,6 @@ class t_gemm(TestCase):
def test_factorised_scalar(self):
a = T.matrix()
b = T.matrix()
c = T.matrix()
s = theano.shared(np.zeros((5, 5)).astype(config.floatX))
lr1 = T.constant(0.01).astype(config.floatX)
......@@ -182,7 +181,7 @@ class t_gemm(TestCase):
f = theano.function([a, b], updates=[(s, lr1 * T.dot(a, b) +
l2_reg * lr2 * s)],
mode=mode_not_fast_compile).maker.fgraph.toposort()
#[Gemm{inplace}(<TensorType(float64, matrix)>, 0.01,
# [Gemm{inplace}(<TensorType(float64, matrix)>, 0.01,
# <TensorType(float64, matrix)>, <TensorType(float64, matrix)>,
# 2e-06)]
assert len(f) == 1
......@@ -192,7 +191,7 @@ class t_gemm(TestCase):
f = theano.function([a, b], updates=[(s, lr1 * (T.dot(a, b) -
l2_reg * s))],
mode=mode_not_fast_compile).maker.fgraph.toposort()
#[Gemm{inplace}(<TensorType(float64, matrix)>, 0.01,
# [Gemm{inplace}(<TensorType(float64, matrix)>, 0.01,
# <TensorType(float64, matrix)>, <TensorType(float64, matrix)>,
# -2e-06)]
assert len(f) == 1
......@@ -202,7 +201,7 @@ class t_gemm(TestCase):
f = theano.function([a, b],
updates=[(s, s - lr1 * (s * .0002 + T.dot(a, b)))],
mode=mode_not_fast_compile).maker.fgraph.toposort()
#[Gemm{inplace}(<TensorType(float64, matrix)>, -0.01,
# [Gemm{inplace}(<TensorType(float64, matrix)>, -0.01,
# <TensorType(float64, matrix)>, <TensorType(float64, matrix)>,
# 0.999998)]
assert len(f) == 1
......@@ -270,14 +269,14 @@ class t_gemm(TestCase):
def t(z, x, y, a=1.0, b=0.0, l='c|py', dt='float64'):
z, a, x, y, b = [theano._asarray(p, dtype=dt)
for p in (z, a, x, y, b)]
z_orig = z.copy()
# z_orig = z.copy()
z_after = self._gemm(z, a, x, y, b)
tz, ta, tx, ty, tb = [shared(p) for p in (z, a, x, y, b)]
# f = inplace_func([tz,ta,tx,ty,tb], gemm_inplace(tz,ta,tx,ty,tb),
# mode = compile.Mode(optimizer = None, linker=l))
#f(z, a, x, y, b)
# f(z, a, x, y, b)
f = inplace_func([], gemm_inplace(tz, ta, tx, ty, tb),
mode=compile.Mode(optimizer=None, linker=l))
f()
......@@ -355,8 +354,9 @@ class t_gemm(TestCase):
tz_i = gemm_no_inplace(tz[:, :, i], ta, tx[
:, :, i], ty[:, :, i], tb)
g_i = theano.function([], tz_i,
updates=[(tz, T.set_subtensor(tz[:, :, i], tz_i))],
g_i = theano.function(
[], tz_i, updates=[(tz, T.set_subtensor(tz[:, :, i],
tz_i))],
mode=compile.Mode(optimizer=None, linker=l))
for j in xrange(3):
g_i()
......@@ -404,8 +404,8 @@ class t_as_scalar(TestCase):
def test1(self):
"""Test that it fails on nonscalar constants"""
a = T.constant(np.ones(5))
self.assertTrue(None == _as_scalar(a))
self.assertTrue(None == _as_scalar(T.DimShuffle([False], [0, 'x'])(a)))
self.assertTrue(_as_scalar(a) is None)
self.assertTrue(_as_scalar(T.DimShuffle([False], [0, 'x'])(a)) is None)
def test2(self):
"""Test that it works on scalar variables"""
......@@ -420,9 +420,9 @@ class t_as_scalar(TestCase):
def test3(self):
"""Test that it fails on nonscalar variables"""
a = T.matrix()
self.assertTrue(None == _as_scalar(a))
self.assertTrue(None == _as_scalar(T.DimShuffle([False, False],
[0, 'x', 1])(a)))
self.assertTrue(_as_scalar(a) is None)
self.assertTrue(_as_scalar(T.DimShuffle([False, False],
[0, 'x', 1])(a)) is None)
class T_real_matrix(TestCase):
......@@ -541,8 +541,8 @@ def test_gemm_opt_double_gemm():
ishapes = [(4, 3), (3, 5), (4, 5), (), (), (5, 9), (9, 4), ()]
i = [X, Y, Z, a, b, R, S, c]
o = [(a * T.dot(X, Y)
+ gemm_inplace(Z, b, S.T, R.T, T.constant(1.0).astype(config.floatX)))]
o = [(a * T.dot(X, Y) +
gemm_inplace(Z, b, S.T, R.T, T.constant(1.0).astype(config.floatX)))]
try:
f = inplace_func([In(ii, mutable=True) for ii in i], o,
mode='FAST_RUN', on_unused_input='ignore')
......@@ -579,8 +579,7 @@ def test_gemm_opt_double_gemm():
def test_gemm_canonicalize():
X, Y, Z, a, b = T.matrix('X'), T.matrix('Y'), T.matrix('Z'), T.scalar(
'a'), T.scalar('b')
R, S, U, c, d = T.matrix('R'), T.matrix('S'), T.matrix('U'), T.scalar(
'c'), T.scalar('d')
c, d = T.scalar('c'), T.scalar('d')
u = T.row('u')
v = T.vector('v')
w = T.col('w')
......@@ -631,10 +630,7 @@ def test_gemm_canonicalize():
def test_gemm_factor():
X, Y, Z, a, b = T.matrix('X'), T.matrix('Y'), T.matrix('Z'), T.scalar(
'a'), T.scalar('b')
R, S, U, c, d = T.matrix('R'), T.matrix('S'), T.matrix('U'), T.scalar(
'c'), T.scalar('d')
X, Y = T.matrix('X'), T.matrix('Y')
assert [(1.0, X), (1.0, Y)] == _factor_canonicalized([(1.0, X), (1.0, Y)])
assert [(2.0, X)] == _factor_canonicalized([(1.0, X), (1.0, X)])
......@@ -653,7 +649,7 @@ def test_upcasting_scalar_nogemm():
f = theano.function([w, v, t, alpha], rval)
t = f.maker.fgraph.toposort()
assert np.sum([isinstance(n.op, Gemm) for n in t]) == 0
#theano.printing.debugprint(f, print_type=True)
# theano.printing.debugprint(f, print_type=True)
v = T.fmatrix('v')
w = T.fmatrix('w')
......@@ -670,7 +666,7 @@ def test_upcasting_scalar_nogemm():
t = f.maker.fgraph.toposort()
assert np.sum([isinstance(n.op, Gemm) for n in t]) == 0
#theano.printing.debugprint(f, print_type=True)
# theano.printing.debugprint(f, print_type=True)
def test_gemm_nested():
......@@ -681,20 +677,20 @@ def test_gemm_nested():
just_gemm([X, Y, Z, R, S, U, a, b, c, d],
[a * Z - b * (c * T.dot(X, Y) + d * Z)],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4), (
2, 4), (), (), (), ()],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4),
(2, 4), (), (), (), ()],
max_graphlen=1)
# print "---------------------"
just_gemm([X, Y, Z, R, S, U, a, b, c, d],
[a * Z - b * (c * T.dot(X, Y) + d * Z + c * Z)],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4), (
2, 4), (), (), (), ()],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4),
(2, 4), (), (), (), ()],
max_graphlen=1)
# print "---------------------"
just_gemm([X, Y, Z, R, S, U, a, b, c, d],
[a * Z - b * (c * T.dot(X, Y) + d * Z + c * U)],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4), (
2, 4), (), (), (), ()],
ishapes=[(2, 3), (3, 4), (2, 4), (2, 3), (3, 4),
(2, 4), (), (), (), ()],
max_graphlen=3)
......@@ -720,7 +716,7 @@ def test_gemm_with_vector():
def my_just_gemm(o):
i = [X, Y, Z, a, b, v]
ishapes = [(4, 3), (3, 5), (4, 5), (), (), (5, )]
rval = just_gemm(i, o, ishapes=ishapes)
just_gemm(i, o, ishapes=ishapes)
my_just_gemm([v + T.dot(X, Y) * a + Z * b])
my_just_gemm([v + a * T.dot(X, Y) + b * Z])
......@@ -741,7 +737,7 @@ def test_gemm_with_vector():
def test_gemm_opt_vector_stuff():
X, Y, Z, a, b = T.matrix(), T.matrix(), T.matrix(), T.scalar(), T.scalar()
X, Y, a = T.matrix(), T.matrix(), T.scalar()
u, v = T.vector(), T.vector()
f = inplace_func([a, u, v], a + T.dot(u, v), mode='FAST_RUN')
......@@ -771,8 +767,6 @@ def test_gemm_unrolled():
H = sharedX(np.zeros((batch_size, rep_size)), name='H')
G = sharedX(np.zeros((batch_size, rep_size)), name='G')
init_V = sharedX(rng.uniform(0, 1, (batch_size, rep_size)), name='init_V')
init_H = sharedX(rng.uniform(0, 1, (batch_size, rep_size)), name='init_H')
cur_V = V
cur_H = H
......@@ -817,7 +811,7 @@ def test_inplace0():
f = inplace_func([X, Y, Z, a, b, R, S, c],
[Z * (c * Z + a * T.dot(X, Y) + b * T.dot(R, S).T)],
mode='FAST_RUN')
if (not gemm_inplace in [n.op for n in f.maker.fgraph.apply_nodes]):
if (gemm_inplace not in [n.op for n in f.maker.fgraph.apply_nodes]):
theano.printing.debugprint(f)
raise Failure('no gemm_inplace in graph')
......@@ -866,7 +860,7 @@ def test_dot22scalar():
# TODO: exclude other optimizations in BlasOpt?
# m = theano.compile.get_default_mode().including('local_dot_to_dot22',
# 'local_dot22_to_dot22scalar','specialize')
#m = theano.compile.get_default_mode().including('BlasOpt', 'specialize')
# m = theano.compile.get_default_mode().including('BlasOpt', 'specialize')
rng = np.random.RandomState(unittest_tools.fetch_seed())
for dtype1 in ['complex64', 'complex128']:
a = T.matrix('a', dtype=dtype1)
......@@ -1087,7 +1081,7 @@ def test_dot_w_self():
class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
def test_dot_vv(self):
''' Currently we generate a gemv for that case'''
# Currently we generate a gemv for that case
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'))
......@@ -1101,11 +1095,10 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
assert np.allclose(f(), np.dot(v.get_value(), w.get_value()))
def test_dot_vm(self):
''' Test vector dot matrix '''
# Test vector dot matrix
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'))
m = theano.shared(np.array(rng.uniform(size=(2, 3)), dtype='float32'))
f = theano.function([], theano.dot(v, m), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
......@@ -1115,17 +1108,14 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
# Assert they produce the same output
assert np.allclose(f(), np.dot(v.get_value(), m.get_value()))
# Assert it works when m has no contiguous dimension
m.set_value(
m.get_value(borrow=True)[::-1, ::-1],
borrow=True)
m.set_value(m.get_value(borrow=True)[::-1, ::-1], borrow=True)
assert np.allclose(f(), np.dot(v.get_value(), m.get_value()))
def test_dot_mv(self):
''' Test matrix dot vector '''
# Test matrix dot vector
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'))
m = theano.shared(np.array(rng.uniform(size=(3, 2)), dtype='float32'))
f = theano.function([], theano.dot(m, v), mode=mode_blas_opt)
# Assert that the dot was optimized somehow
......@@ -1135,31 +1125,27 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
# Assert they produce the same output
assert np.allclose(f(), np.dot(m.get_value(), v.get_value()))
# Assert it works when m has no contiguous dimension
m.set_value(
m.get_value(borrow=True)[::-1, ::-1],
borrow=True)
m.set_value(m.get_value(borrow=True)[::-1, ::-1], borrow=True)
assert np.allclose(f(), np.dot(m.get_value(), v.get_value()))
@staticmethod
def t_gemv1(m_shp):
''' test vector2+dot(matrix,vector1) '''
# test vector2+dot(matrix,vector1)
rng = np.random.RandomState(unittest_tools.fetch_seed())
v1 = theano.shared(np.array(rng.uniform(size=(m_shp[1],)
), dtype='float32'))
v1 = theano.shared(np.array(rng.uniform(size=(m_shp[1],)),
dtype='float32'))
v2_orig = np.array(rng.uniform(size=(m_shp[0],)), dtype='float32')
v2 = theano.shared(v2_orig)
m = theano.shared(np.array(rng.uniform(size=m_shp),
dtype='float32'))
m = theano.shared(np.array(rng.uniform(size=m_shp), dtype='float32'))
f = theano.function([], v2 + theano.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)
assert np.allclose(f(), np.dot(m.get_value(), v1.get_value()) + v2_orig)
topo = f.maker.fgraph.toposort()
assert len(topo) == 1
assert isinstance(topo[0].op, Gemv)
assert topo[0].op.inplace == False
assert topo[0].op.inplace is False
# test the inplace version
g = theano.function([], [], updates=[(v2, v2 + theano.dot(m, v1))],
......@@ -1167,17 +1153,16 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
# Assert they produce the same output
g()
assert np.allclose(v2.get_value(),
np.dot(m.get_value(), v1.get_value()) + v2_orig)
assert np.allclose(v2.get_value(), np.dot(m.get_value(),
v1.get_value()) + v2_orig)
topo = g.maker.fgraph.toposort()
assert len(topo) == 1
assert isinstance(topo[0].op, Gemv)
if config.mode != 'FAST_COMPILE':
assert topo[0].op.inplace == True
assert topo[0].op.inplace is True
# Do the same tests with a matrix with strides in both dimensions
m.set_value(
m.get_value(borrow=True)[::-1, ::-1],
m.set_value(m.get_value(borrow=True)[::-1, ::-1],
borrow=True)
v2.set_value(v2_orig)
assert np.allclose(f(),
......@@ -1194,7 +1179,7 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
self.t_gemv1((0, 0))
def test_gemv2(self):
''' test vector2+dot(vector1,matrix) '''
# test vector2+dot(vector1,matrix)
rng = np.random.RandomState(unittest_tools.fetch_seed())
v1 = theano.shared(np.array(rng.uniform(size=(2,)),
dtype='float32'))
......@@ -1207,10 +1192,11 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
# Assert they produce the same output
assert np.allclose(f(),
np.dot(v1.get_value(), m.get_value()) + v2.get_value())
np.dot(v1.get_value(), m.get_value()) +
v2.get_value())
topo = f.maker.fgraph.toposort()
assert sum(isinstance(node.op, Gemv) for node in topo) == 1
assert topo[-1].op.inplace == False
assert topo[-1].op.inplace is False
# test the inplace version
g = theano.function([], [], updates=[(v2, v2 + theano.dot(v1, m))],
......@@ -1223,21 +1209,21 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
topo = g.maker.fgraph.toposort()
assert sum(isinstance(node.op, Gemv) for node in topo) == 1
if config.mode != 'FAST_COMPILE':
assert topo[-1].op.inplace == True
assert topo[-1].op.inplace is True
# Do the same tests with a matrix with strides in both dimensions
m.set_value(
m.get_value(borrow=True)[::-1, ::-1],
m.set_value(m.get_value(borrow=True)[::-1, ::-1],
borrow=True)
v2.set_value(v2_orig)
assert np.allclose(f(),
np.dot(v1.get_value(), m.get_value()) + v2.get_value())
np.dot(v1.get_value(), m.get_value()) +
v2.get_value())
g()
assert np.allclose(v2.get_value(),
np.dot(v1.get_value(), m.get_value()) + v2_orig)
def test_gemv_broadcast(self):
''' test gemv with some broadcasted input '''
# test gemv with some broadcasted input
rng = np.random.RandomState(unittest_tools.fetch_seed())
v1 = theano.shared(np.array(rng.uniform(size=(2,)),
dtype='float32'))
......@@ -1261,7 +1247,7 @@ class TestGemv(TestCase, unittest_tools.TestOptimizationMixin):
f = theano.function([], o, mode=mode_blas_opt)
assert np.allclose(
f(),
0.5*np.dot(m.get_value(), v1.get_value()) + 0.25*v2.get_value())
0.5 * np.dot(m.get_value(), v1.get_value()) + 0.25 * v2.get_value())
topo = f.maker.fgraph.toposort()
assert sum(isinstance(node.op, Gemv) for node in topo) == 1
......@@ -1336,8 +1322,8 @@ class BaseGemv(object):
def test_simple(self):
alpha, beta, a, x, y = [self.shared(value)
for value in self.get_data()]
desired_oy = alpha.get_value() * matrixmultiply(a.
get_value(), x.get_value()) + beta.get_value() * y.get_value()
desired_oy = alpha.get_value() *
matrixmultiply(a.get_value(), x.get_value()) + beta.get_value() * y.get_value()
oy = alpha * T.dot(a, x) + beta * y
......@@ -1406,8 +1392,8 @@ class BaseGemv(object):
alpha_v, beta_v, a_v, x_v, y_v = vs
alpha, beta, a, x, y = [self.shared(v) for v in vs]
desired_oy = alpha_v * matrixmultiply(transpose(a_v), x_v[::
2]) + beta_v * y_v
desired_oy = alpha_v * matrixmultiply(transpose(a_v), x_v[::2]) +
beta_v * y_v
oy = alpha * T.dot(a.T, x[::2]) + beta * y
......@@ -1456,8 +1442,7 @@ class BaseGemv(object):
alpha_v, beta_v, a_v, x_v, y_v = vs
alpha, beta, a, x, y = [self.shared(v) for v in vs]
a_v = a_v[::-1, ::-1]
a.set_value(
a.get_value(borrow=True,
a.set_value(a.get_value(borrow=True,
return_internal_type=True)[::-1, ::-1],
borrow=True)
......@@ -1477,8 +1462,7 @@ class BaseGemv(object):
alpha_v, beta_v, a_v, x_v, y_v = vs
alpha, beta, a, x, y = [self.shared(v) for v in vs]
a_v = a_v[::-1, ::-1]
a.set_value(
a.get_value(borrow=True,
a.set_value(a.get_value(borrow=True,
return_internal_type=True)[::-1, ::-1],
borrow=True)
......@@ -1517,7 +1501,7 @@ class BaseGemv(object):
# done inplace on a temporarily allocated-buffer, which is
# then scaled by alpha and to t with a fused elemwise.
n_gemvs = 0
#theano.printing.debugprint(f, print_type=True)
# theano.printing.debugprint(f, print_type=True)
for node in f.maker.fgraph.toposort():
if node.op == self.gemv_inplace:
n_gemvs += 1
......@@ -1602,7 +1586,8 @@ class TestGer_make_node(TestCase):
ger, self.fm, self.fv1, self.fv, self.fv_2)
# actually doing the aforementioned dimshuffle makes it work
self.assertEqual(self.fm.type,
ger(self.fm, self.fv1.dimshuffle(), self.fv, self.fv_2).type)
ger(self.fm, self.fv1.dimshuffle(), self.fv,
self.fv_2).type)
def test_fails_for_nonmatrix_A(self):
self.assertRaises(TypeError,
......@@ -1610,9 +1595,11 @@ class TestGer_make_node(TestCase):
def test_fails_for_nonvector_x_or_y(self):
self.assertRaises(TypeError,
ger, self.fm, self.fa, self.fv.dimshuffle('x', 0), self.fv_2)
ger, self.fm, self.fa,
self.fv.dimshuffle('x', 0), self.fv_2)
self.assertRaises(TypeError,
ger, self.fm, self.fa, self.fv, self.fv_2.dimshuffle('x', 0))
ger, self.fm, self.fa,
self.fv, self.fv_2.dimshuffle('x', 0))
def test_fails_for_mixed_dtypes(self):
self.assertRaises(TypeError, ger, self.dm, self.fa, self.fv, self.fv_2)
......@@ -1657,29 +1644,25 @@ class TestGer(TestCase, unittest_tools.TestOptimizationMixin):
def test_b_0_triggers_ger(self):
""" test local_gemm_to_ger opt"""
assert T.blas.local_gemm_to_ger.transform(
gemm_no_inplace(
self.A, self.a, self.x.dimshuffle(0, 'x'),
gemm_no_inplace(self.A, self.a, self.x.dimshuffle(0, 'x'),
self.y.dimshuffle('x', 0), self.b(0)).owner)
def test_b_1_triggers_ger(self):
""" test local_gemm_to_ger opt"""
assert T.blas.local_gemm_to_ger.transform(
gemm_no_inplace(
self.A, self.a, self.x.dimshuffle(0, 'x'),
gemm_no_inplace(self.A, self.a, self.x.dimshuffle(0, 'x'),
self.y.dimshuffle('x', 0), self.b(1)).owner)
def test_b_other_does_not_triggers_ger(self):
""" test local_gemm_to_ger opt"""
assert not T.blas.local_gemm_to_ger.transform(
gemm_no_inplace(
self.A, self.a, self.x.dimshuffle(0, 'x'),
gemm_no_inplace(self.A, self.a, self.x.dimshuffle(0, 'x'),
self.y.dimshuffle('x', 0), self.b(1.5)).owner)
def test_b_nonconst_does_not_triggers_ger(self):
""" test local_gemm_to_ger opt"""
assert not T.blas.local_gemm_to_ger.transform(
gemm_no_inplace(
self.A, self.a, self.x.dimshuffle(0, 'x'),
gemm_no_inplace(self.A, self.a, self.x.dimshuffle(0, 'x'),
self.y.dimshuffle('x', 0), self.a).owner)
def test_outer(self):
......@@ -1954,28 +1937,36 @@ class TestBlasStrides(TestCase):
bt_dev = b_t.get_value(borrow=False, return_internal_type=True)
ct_dev = c_t.get_value(borrow=False, return_internal_type=True)
f_nnn = theano.function([], [],
f_nnn = theano.function(
[], [],
updates=[(a, (l * a + tensor.dot(b, c)))],
mode=self.mode)
f_nnt = theano.function([], [],
f_nnt = theano.function(
[], [],
updates=[(a, (l * a + tensor.dot(b, c_t.T)))],
mode=self.mode)
f_ntn = theano.function([], [],
f_ntn = theano.function(
[], [],
updates=[(a, (l * a + tensor.dot(b_t.T, c)))],
mode=self.mode)
f_ntt = theano.function([], [],
f_ntt = theano.function(
[], [],
updates=[(a, (l * a + tensor.dot(b_t.T, c_t.T)))],
mode=self.mode)
f_tnn = theano.function([], [],
f_tnn = theano.function(
[], [],
updates=[(a_t, (l * a_t + tensor.dot(b, c).T))],
mode=self.mode)
f_tnt = theano.function([], [],
f_tnt = theano.function(
[], [],
updates=[(a_t, (l * a_t + tensor.dot(b, c_t.T).T))],
mode=self.mode)
f_ttn = theano.function([], [],
f_ttn = theano.function(
[], [],
updates=[(a_t, (l * a_t + tensor.dot(b_t.T, c).T))],
mode=self.mode)
f_ttt = theano.function([], [],
f_ttt = theano.function(
[], [],
updates=[(a_t, (l * a_t + tensor.dot(b_t.T, c_t.T).T))],
mode=self.mode)
......@@ -1991,11 +1982,11 @@ class TestBlasStrides(TestCase):
c_t.set_value(ct_dev.copy()[::c_step2, ::c_step1], borrow=True)
# Numpy results
a_n = (l * av[::a_step1, ::a_step2]
+ np.dot(bv[::b_step1, ::b_step2],
a_n = (l * av[::a_step1, ::a_step2] +
np.dot(bv[::b_step1, ::b_step2],
cv[::c_step1, ::c_step2]))
at_n = (l * av[::a_step1, ::a_step2].T
+ np.dot(bv[::b_step1, ::b_step2],
at_n = (l * av[::a_step1, ::a_step2].T +
np.dot(bv[::b_step1, ::b_step2],
cv[::c_step1, ::c_step2]).T)
# a's value is updated, so we need to reinitialize it each time
......@@ -2085,8 +2076,8 @@ class TestBlasStrides(TestCase):
borrow=True)
c.set_value(c_dev.copy()[::c_step], borrow=True)
a_n = (av[::a_step]
+ l * np.dot(bv[::b_step1, ::b_step2],
a_n = (av[::a_step] +
l * np.dot(bv[::b_step1, ::b_step2],
cv[::c_step]))
f_n()
assert np.allclose(a.get_value(), a_n), (a.get_value(), a_n)
......@@ -2120,11 +2111,13 @@ class TestBlasStrides(TestCase):
b_dev = b.get_value(borrow=False, return_internal_type=True)
c_dev = c.get_value(borrow=False, return_internal_type=True)
f_n = theano.function([], [],
f_n = theano.function(
[], [],
updates=[(a, (a + l * tensor.outer(b, c)))],
mode=self.mode)
f_t = theano.function([], [],
f_t = theano.function(
[], [],
updates=[(a_t, (a_t + l * tensor.outer(b, c).T))],
mode=self.mode)
......@@ -2141,15 +2134,14 @@ class TestBlasStrides(TestCase):
c.set_value(c_dev.copy()[::c_step], borrow=True)
f_n()
n_n = (av[::a_step1, ::a_step2]
+ l * np.outer(bv[::b_step], cv[::c_step]))
n_n = (av[::a_step1, ::a_step2] +
l * np.outer(bv[::b_step], cv[::c_step]))
assert np.allclose(a.get_value(), n_n), (a.get_value(), n_n)
f_t()
n_t = (av.T[::a_step1, ::a_step2]
+ l * np.outer(bv[::b_step], cv[::c_step]).T)
assert np.allclose(a_t.get_value(), n_t),\
(a_t.get_value(), n_t)
n_t = (av.T[::a_step1, ::a_step2] +
l * np.outer(bv[::b_step], cv[::c_step]).T)
assert np.allclose(a_t.get_value(), n_t), (a_t.get_value(), n_t)
def test_ger_strides(self):
self.cmp_ger((3, 5), 3, 5)
......
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