def test_case_not_tensor(self):
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
for init_num_dims in range(4):
shape = list(range(1, 1 + init_num_dims))
size = 1
if len(shape) > 0:
for d in shape:
size *= d
if init_num_dims == 0:
values = np.array(1.)
else:
values = np.arange(0, size, 1)
for _ in range(init_num_dims + 1):
shape = shape[-1:] + shape[:-1]
np_tens = np.reshape(values, shape)
axes = tf.range(init_num_dims)
num_dims = tf.constant(init_num_dims)
expanded = tensor_ops.expand_multiple_dims(
np_tens, num_dims, axes)
msg_values = "tensor == {}\n" \
"num_dims == {}\n" \
"axes == {}".format(np_tens, init_num_dims, list(range(init_num_dims)))
assert expanded is not None, \
"expand_multiple_dims() returned `None` on\n" + msg_values
r = sess.run(expanded)
assert (r == np_tens).all(
), "expand_multiple_dims() returned\n{} on\n".format(
r) + msg_values
def test_insert_inside_and_permute(self):
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
tensor_numdims_axes_output = [
(
[[[1, 2], [3, 4]], [[5, 6], [7, 8]]],
5,
[2, 0, 4],
[[[[[1, 2]], [[5, 6]]]], [[[[3, 4]], [[7, 8]]]]],
),
([[[1, 2], [3, 4]], [[5, 6], [7, 8]]], 5, [0, 4,
2], [[[[[1, 3]],
[[2, 4]]]],
[[[[5, 7]],
[[6, 8]]]]]),
([[[1, 2], [3, 4]], [[5, 6], [7, 8]]], 5, [4, 2,
0], [[[[[1, 5]],
[[3, 7]]]],
[[[[2, 6]],
[[4, 8]]]]])
]
for tensor, num_dims, axes, output in tensor_numdims_axes_output:
tensor = np.array(tensor)
expanded = tensor_ops.expand_multiple_dims(
tensor, num_dims, axes)
r = sess.run(expanded)
assert (r == np.array(output)).all(), \
'failed on tensor={}, num_dims={} and axes={}\noutput={}\nexpected={}'.format(
tensor, num_dims, axes, r, np.array(output))
def test_permutation_with_negative_axes(self):
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
tensor_axes_output = [
([[1, 2], [3, 4]], [-1, -2], [[1, 3], [2, 4]]),
(
[[[1, 2], [3, 4]], [[5, 6], [7, 8]]],
[-2, -3, -1],
[[[1, 2], [5, 6]], [[3, 4], [7, 8]]],
),
([[[1, 2], [3, 4]], [[5, 6], [7, 8]]], [-3, -1, -2], [[[1, 3],
[2, 4]],
[[5, 7],
[6,
8]]]),
([[[1, 2], [3, 4]], [[5, 6], [7, 8]]], [-1, -2, -3], [[[1, 5],
[3, 7]],
[[2, 6],
[4,
8]]])
]
for tensor, axes, output in tensor_axes_output:
tensor = np.array(tensor)
expanded = tensor_ops.expand_multiple_dims(
tensor, tensor.ndim, axes)
r = sess.run(expanded)
assert (r == np.array(output)
).all(), 'failed on tensor={} and axes={}'.format(
tensor, axes)
def test_expansion_of_no_shape(self):
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
tensor = 6.
num_dims = 3
axes = []
expected = [[[6.]]]
expanded = tensor_ops.expand_multiple_dims(tensor, num_dims, axes)
r = sess.run(expanded)
assert (r == np.array(expected)).all(), \
'failed on tensor={}, num_dims={} and axes={}\noutput={}\nexpected={}'.format(
tensor, num_dims, axes, r, np.array(expected)
)
def test_raises_if_num_dims_is_too_small(self):
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
num_dims = tf.placeholder(tf.int32)
tensor = tf.placeholder(tf.float32)
axes = tf.placeholder(tf.int32)
expanded = tensor_ops.expand_multiple_dims(tensor, num_dims, axes)
with pytest.raises(tf.errors.InvalidArgumentError):
_ = sess.run(expanded,
feed_dict={
num_dims: 1,
tensor: [[2]],
axes: [1, 2]
})
def test_insert_dims_after(self):
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
tensor_numdims_axes_output = [
([1, 2], 2, [0], [[1], [2]]),
([[1, 2]], 3, [0, 1], [[[1], [2]]]),
([1, 2], 3, [0], [[[1]], [[2]]]),
]
for tensor, num_dims, axes, output in tensor_numdims_axes_output:
tensor = np.array(tensor)
expanded = tensor_ops.expand_multiple_dims(
tensor, num_dims, axes)
r = sess.run(expanded)
assert (r == np.array(output)).all(), \
'failed on tensor={}, num_dims={} and axes={}\noutput={}\nexpected={}'.format(
tensor, num_dims, axes, r, np.array(output))