def test_fail(self):
# Test that conv2d fails for dimensions other than 2 or 3.
with pytest.raises(Exception):
conv.conv2d(dtensor4(), dtensor3())
with pytest.raises(Exception):
conv.conv2d(dtensor3(), dvector())
def test_norm(self, axis):
x = dtensor3()
a = np.random.random((3, 2, 4)).astype(aesara.config.floatX)
mode = Mode(optimizer="fast_compile", linker="py")
f = function(
[x],
[
x.norm(L=1, axis=axis, keepdims=True),
self.makeKeepDims_local(
x, x.norm(L=1, axis=axis, keepdims=False), axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
g = function(
[x],
[
x.norm(L=2, axis=axis, keepdims=True),
self.makeKeepDims_local(
x, x.norm(L=2, axis=axis, keepdims=False), axis),
],
mode=mode,
)
ans1, ans2 = g(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
def test_max_pool_3d_3D_deprecated_interface(self):
rng = np.random.RandomState(utt.fetch_seed())
maxpoolshps = ((1, 1, 1), (3, 2, 1))
imval = rng.rand(4, 5, 6)
images = dtensor3()
for maxpoolshp, ignore_border, mode in product(
maxpoolshps,
[True, False],
["max", "sum", "average_inc_pad", "average_exc_pad"],
):
# print 'maxpoolshp =', maxpoolshp
# print 'ignore_border =', ignore_border
numpy_output_val = self.numpy_max_pool_nd(imval,
maxpoolshp,
ignore_border,
mode=mode)
output = pool_3d(
input=images,
ds=maxpoolshp,
ignore_border=ignore_border,
st=maxpoolshp,
padding=(0, 0, 0),
mode=mode,
)
output_val = function([images], output)(imval)
utt.assert_allclose(output_val, numpy_output_val)
def mp(input):
return pool_3d(input, maxpoolshp, ignore_border, mode=mode)
def test_wrong_input(self):
# 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=dmatrix())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", filters=dvector())
with pytest.raises(Exception):
self.validate((3, 2, 8, 8), (4, 2, 5, 5), "valid", input=dtensor3())
def test_norm(self):
x = dtensor3()
a = np.random.rand(3, 2, 4).astype(aesara.config.floatX)
mode = Mode(optimizer="fast_compile", linker="py")
for axis in [
0,
1,
2,
[0],
[1],
[2],
None,
[0, 1],
[1, 2],
[0, 1, 2],
[-1],
[-2],
[-3],
[-1, -2],
[-1, -2, -3],
[0, -2, 2],
]:
f = function(
[x],
[
x.norm(L=1, axis=axis, keepdims=True),
self.makeKeepDims_local(
x, x.norm(L=1, axis=axis, keepdims=False), axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
g = function(
[x],
[
x.norm(L=2, axis=axis, keepdims=True),
self.makeKeepDims_local(
x, x.norm(L=2, axis=axis, keepdims=False), axis),
],
mode=mode,
)
ans1, ans2 = g(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
def test_infer_shape(self):
z = dtensor3()
x = dmatrix()
y = dscalar()
self._compile_and_check(
[x, y],
[self.op(x, y)],
[np.random.random((8, 5)), np.random.random()],
self.op_class,
)
self._compile_and_check(
[z, y],
[self.op(z, y)],
# must be square when nd>2
[np.random.random((8, 8, 8)), np.random.random()],
self.op_class,
warn=False,
)
def test_max_pool_2d_3D(self):
rng = np.random.RandomState(utt.fetch_seed())
maxpoolshps = [(1, 2)]
imval = rng.rand(2, 3, 4)
images = dtensor3()
for maxpoolshp, ignore_border, mode in product(
maxpoolshps,
[True, False],
["max", "sum", "average_inc_pad", "average_exc_pad"],
):
# print 'maxpoolshp =', maxpoolshp
# print 'ignore_border =', ignore_border
numpy_output_val = self.numpy_max_pool_2d(imval, maxpoolshp,
ignore_border, mode)
output = pool_2d(images, maxpoolshp, ignore_border, mode=mode)
output_val = function([images], output)(imval)
utt.assert_allclose(output_val, numpy_output_val)
def test_single_or_any_axis(self, axis, op):
# the following ops can be specified with either a single axis or every
# axis:
x = dtensor3()
a = np.random.random((3, 2, 4))
# We don't need to test all opt and C code, as this is tested
# by the ops tests.
mode = Mode(optimizer="fast_compile", linker="py")
f = function(
[x],
[
op(x, axis=axis, keepdims=True),
self.makeKeepDims_local(x, op(x, axis=axis, keepdims=False),
axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
def test_max_and_argmax(self, axis, op):
x = dtensor3()
a = np.random.random((3, 2, 4))
# We don't need to test all opt and C code, as this is tested
# by the ops tests.
mode = Mode(optimizer="fast_compile", linker="py")
# 'max_and_argmax' has two outputs and can be specified with either
# a single or every axis:
f = function(
[x],
[
op(x, axis=axis, keepdims=True)[0],
self.makeKeepDims_local(x,
op(x, axis=axis, keepdims=False)[0],
axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
f = function(
[x],
[
op(x, axis=axis, keepdims=True)[1],
self.makeKeepDims_local(x,
op(x, axis=axis, keepdims=False)[1],
axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
def test_keepdims(self):
x = dtensor3()
a = np.random.rand(3, 2, 4)
# We don't need to test all opt and C code, as this is tested
# by the ops tests.
mode = Mode(optimizer="fast_compile", linker="py")
# 'max_and_argmax' has two outputs and can be specified with either
# a single or every axis:
for axis in [
0,
1,
2,
[0],
[1],
[2],
None,
[0, 1, 2],
[-1],
[-2],
[-3],
[-1, -2, -3],
[0, -1, -2],
[-2, -3, 2],
]:
op = max_and_argmax
f = function(
[x],
[
op(x, axis=axis, keepdims=True)[0],
self.makeKeepDims_local(
x,
op(x, axis=axis, keepdims=False)[0], axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
f = function(
[x],
[
op(x, axis=axis, keepdims=True)[1],
self.makeKeepDims_local(
x,
op(x, axis=axis, keepdims=False)[1], axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
# the following ops can be specified with either a single axis or every
# axis:
for op in [argmax, argmin]:
for axis in [
0,
1,
2,
[0],
[1],
[2],
None,
[0, 1, 2],
[-1],
[-2],
[-3],
[-1, -2, -3],
[0, -2, 2],
]:
f = function(
[x],
[
op(x, axis=axis, keepdims=True),
self.makeKeepDims_local(
x, op(x, axis=axis, keepdims=False), axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
# the following ops can be specified with a freely specified axis
# parameter
for op in [
tt_sum,
prod,
mean,
var,
std,
tt_all,
tt_any,
tt_max,
tt_min,
]:
for axis in [
0,
1,
2,
[0],
[1],
[2],
None,
[0, 1],
[1, 2],
[0, 1, 2],
[-1],
[-2],
[-3],
[-1, -2],
[-1, -2, -3],
[0, -2, 2],
]:
f = function(
[x],
[
op(x, axis=axis, keepdims=True),
self.makeKeepDims_local(
x, op(x, axis=axis, keepdims=False), axis),
],
mode=mode,
)
ans1, ans2 = f(a)
assert np.allclose(ans1, ans2)
assert ans1.shape == ans2.shape
