def test_bad_shape(self):
# Test that at run time we raise an exception when the shape
# is not the one specified
specify_shape = SpecifyShape()
x = vector()
xval = np.random.rand(2).astype(config.floatX)
f = theano.function([x], specify_shape(x, [2]), mode=self.mode)
f(xval)
xval = np.random.rand(3).astype(config.floatX)
with pytest.raises(AssertionError):
f(xval)
assert isinstance(
[n for n in f.maker.fgraph.toposort() if isinstance(n.op, SpecifyShape)][0]
.inputs[0]
.type,
self.input_type,
)
x = matrix()
xval = np.random.rand(2, 3).astype(config.floatX)
f = theano.function([x], specify_shape(x, [2, 3]), mode=self.mode)
assert isinstance(
[n for n in f.maker.fgraph.toposort() if isinstance(n.op, SpecifyShape)][0]
.inputs[0]
.type,
self.input_type,
)
f(xval)
for shape_ in [(1, 3), (2, 2), (5, 5)]:
xval = np.random.rand(*shape_).astype(config.floatX)
with pytest.raises(AssertionError):
f(xval)
def check_force_gemv_init():
if check_force_gemv_init._force_init_beta is None:
"""
Test issue 1569.
Namely when evaulating
beta*aa + alpha*dot(xx, yy)
where we set aa = betas = zeros of the correct dimensions we do not
actually set aa = zeros and instead let the BLAS perform beta*aa with
uninitialized memory for speed. Occasionally the memory contains values
that are equivalent to NaN in which case the product beta*aa contains
NaN's for correctly implemented BLAS libraries. In this situation, since
we are introducing the NaN's, we need to test whether the BLAS performs
correctly. If it *does*, i.e. it actually performs the multiplication
beta*aa which will result in NaN's in the result, then we need intialize
the memory to zeros.
"""
tv = theano.config.compute_test_value
tvo = theano.config.compute_test_value_opt
theano.config.compute_test_value = 'off'
theano.config.compute_test_value_opt = 'off'
try:
aa = T.vector('aa')
yy = T.vector('yy')
xx = T.matrix('xx')
f = theano.function(
[aa, yy, xx],
gemv_no_inplace(aa, 1., xx, yy, 0.),
theano.compile.Mode(optimizer='fast_compile').excluding('gpu',
'gpuarray'),
profile=False
)
finally:
theano.config.compute_test_value = tv
theano.config.compute_test_value_opt = tvo
# Here we introduce NaNs into the data, if they are returned by the BLAS
# then we want gemv_c_code to initiliaze the memory to 0 so that we
# don't inadvertantly introduce NaNs to the users data.
aa_data = numpy.array(
float('NaN')*numpy.ones((2,)),
dtype=theano.config.floatX
)
yy_data = numpy.array(
numpy.ones((2,))*2,
dtype=theano.config.floatX
)
xx_data = numpy.array(
numpy.ones((2, 2)),
dtype=theano.config.floatX
)
zz = f(aa_data, yy_data, xx_data)
check_force_gemv_init._force_init_beta = numpy.isnan(zz).any()
return check_force_gemv_init._force_init_beta
def check_force_gemv_init():
if check_force_gemv_init._force_init_beta is None:
"""
Test issue 1569.
Namely when evaulating
beta*aa + alpha*dot(xx, yy)
where we set aa = betas = zeros of the correct dimensions we do not
actually set aa = zeros and instead let the BLAS perform beta*aa with
uninitialized memory for speed. Occasionally the memory contains values
that are equivalent to NaN in which case the product beta*aa contains
NaN's for correctly implemented BLAS libraries. In this situation, since
we are introducing the NaN's, we need to test whether the BLAS performs
correctly. If it *does*, i.e. it actually performs the multiplication
beta*aa which will result in NaN's in the result, then we need intialize
the memory to zeros.
"""
tv = theano.config.compute_test_value
tvo = theano.config.compute_test_value_opt
theano.config.compute_test_value = 'off'
theano.config.compute_test_value_opt = 'off'
try:
aa = T.vector('aa')
yy = T.vector('yy')
xx = T.matrix('xx')
f = theano.function(
[aa, yy, xx],
gemv_no_inplace(aa, 1., xx, yy, 0.),
theano.compile.Mode(optimizer='fast_compile').excluding(
'gpu', 'gpuarray'),
profile=False)
finally:
theano.config.compute_test_value = tv
theano.config.compute_test_value_opt = tvo
# Here we introduce NaNs into the data, if they are returned by the BLAS
# then we want gemv_c_code to initiliaze the memory to 0 so that we
# don't inadvertantly introduce NaNs to the users data.
aa_data = numpy.array(float('NaN') * numpy.ones((2, )),
dtype=theano.config.floatX)
yy_data = numpy.array(numpy.ones((2, )) * 2,
dtype=theano.config.floatX)
xx_data = numpy.array(numpy.ones((2, 2)), dtype=theano.config.floatX)
zz = f(aa_data, yy_data, xx_data)
check_force_gemv_init._force_init_beta = numpy.isnan(zz).any()
return check_force_gemv_init._force_init_beta
def test_local_hard_sigmoid(self):
x = tensor.matrix('x')
s = sigmoid(x)
mode = self.get_mode('local_hard_sigmoid')
f = theano.function([x], s, mode=mode)
topo = f.maker.fgraph.toposort()
assert topo[0].op == sigmoid
assert len(topo) == 1
mode = self.get_mode().including('local_hard_sigmoid')
f = theano.function([x], s, mode=mode)
topo = f.maker.fgraph.toposort()
assert len(topo) > 1
assert not any([n.op == sigmoid for n in topo])
ux_v = f([[-50, -10, -4, -1, 0, 1, 4, 10, 50]])
def test_local_ultra_fast_sigmoid(self):
x = tensor.matrix('x')
s = sigmoid(x)
mode = self.get_mode('local_ultra_fast_sigmoid')
f = theano.function([x], s, mode=mode)
topo = f.maker.fgraph.toposort()
assert len(topo) == 1
assert topo[0].op == sigmoid
mode = self.get_mode().including('local_ultra_fast_sigmoid')
f = theano.function([x], s, mode=mode)
topo = f.maker.fgraph.toposort()
assert topo[0].op == ultra_fast_sigmoid
assert len(topo) == 1
ux_v = f([[-50, -10, -4, -1, 0, 1, 4, 10, 50]])
def test_local_ultra_fast_sigmoid(self):
x = tensor.matrix("x")
s = sigmoid(x)
mode = self.get_mode("local_ultra_fast_sigmoid")
f = theano.function([x], s, mode=mode)
assert check_stack_trace(f, ops_to_check=sigmoid)
topo = f.maker.fgraph.toposort()
assert len(topo) == 1
assert topo[0].op == sigmoid
mode = self.get_mode().including("local_ultra_fast_sigmoid")
f = theano.function([x], s, mode=mode)
assert check_stack_trace(f, ops_to_check=ultra_fast_sigmoid)
topo = f.maker.fgraph.toposort()
assert topo[0].op == ultra_fast_sigmoid
assert len(topo) == 1
f([[-50, -10, -4, -1, 0, 1, 4, 10, 50]])
def test_bad_number_of_shape(self):
# Test that the number of dimensions provided is good
specify_shape = SpecifyShape()
x = vector()
shape_vec = ivector()
xval = np.random.rand(2).astype(config.floatX)
with pytest.raises(AssertionError):
specify_shape(x, [])
with pytest.raises(AssertionError):
specify_shape(x, [2, 2])
f = theano.function([x, shape_vec],
specify_shape(x, shape_vec),
mode=self.mode)
assert isinstance(
[
n for n in f.maker.fgraph.toposort()
if isinstance(n.op, SpecifyShape)
][0].inputs[0].type,
self.input_type,
)
with pytest.raises(AssertionError):
f(xval, [])
with pytest.raises(AssertionError):
f(xval, [2, 2])
x = matrix()
xval = np.random.rand(2, 3).astype(config.floatX)
for shape_ in [(), (1, ), (2, 3, 4)]:
with pytest.raises(AssertionError):
specify_shape(x, shape_)
f = theano.function([x, shape_vec],
specify_shape(x, shape_vec),
mode=self.mode)
assert isinstance(
[
n for n in f.maker.fgraph.toposort()
if isinstance(n.op, SpecifyShape)
][0].inputs[0].type,
self.input_type,
)
with pytest.raises(AssertionError):
f(xval, shape_)
def test_grad_log1msigm(self):
# 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')
s = sigmoid(x)
l = T.log(1 - s)
c = l.mean()
ux = x - lr * theano.grad(c, x)
# Before the optimization, inf and NaN will be produced in the graph,
# and DebugMode will complain. Everything is fine afterwards.
mode = self.get_mode()
if not isinstance(mode, theano.compile.DebugMode):
f = theano.function([x, lr], ux, mode=mode)
ux_v = f([[50]], 0.1)
assert not np.isnan(ux_v)
def test_grad_log1msigm(self):
# 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')
s = sigmoid(x)
l = T.log(1 - s)
c = l.mean()
ux = x - lr * theano.grad(c, x)
# Before the optimization, inf and NaN will be produced in the graph,
# and DebugMode will complain. Everything is fine afterwards.
mode = self.get_mode()
if not isinstance(mode, theano.compile.DebugMode):
f = theano.function([x, lr], ux, mode=mode)
ux_v = f([[50]], 0.1)
assert not numpy.isnan(ux_v)
def test_local_hard_sigmoid(self):
x = tensor.matrix('x')
s = sigmoid(x)
mode = self.get_mode('local_hard_sigmoid')
f = theano.function([x], s, mode=mode)
assert check_stack_trace(f, ops_to_check=sigmoid)
topo = f.maker.fgraph.toposort()
assert topo[0].op == sigmoid
assert len(topo) == 1
mode = self.get_mode().including('local_hard_sigmoid')
f = theano.function([x], s, mode=mode)
topo = f.maker.fgraph.toposort()
assert not any([n.op == sigmoid for n in topo])
f([[-50, -10, -4, -1, 0, 1, 4, 10, 50]])
mode2 = mode.excluding('fusion').excluding('inplace')
f2 = theano.function([x], s, mode=mode2)
self.assertTrue(check_stack_trace(f2, ops_to_check=theano.tensor.clip))
def test_local_hard_sigmoid(self):
x = tensor.matrix('x')
s = sigmoid(x)
mode = self.get_mode('local_hard_sigmoid')
f = theano.function([x], s, mode=mode)
assert check_stack_trace(f, ops_to_check=sigmoid)
topo = f.maker.fgraph.toposort()
assert topo[0].op == sigmoid
assert len(topo) == 1
mode = self.get_mode().including('local_hard_sigmoid')
f = theano.function([x], s, mode=mode)
topo = f.maker.fgraph.toposort()
assert not any([n.op == sigmoid for n in topo])
f([[-50, -10, -4, -1, 0, 1, 4, 10, 50]])
mode2 = mode.excluding('fusion').excluding('inplace')
f2 = theano.function([x], s, mode=mode2)
self.assertTrue(check_stack_trace(f2, ops_to_check=theano.tensor.clip))