def test_fit_to_less_width(self):
"""Fit a tensor to a smalles width (i.e. trimming).
Given a 3D tensor of shape [batch, length, width], apply the
`ops.fit()` operator to it with the a smaller `width` as the
target one and check that the last axis of the tensor have been
deleted.
"""
batch = 2
length = 5
width = 4
fit_width = 3
delta = width - fit_width
shape = [None, None, None]
input_ = tf.placeholder(dtype=tf.float32, shape=shape)
output = ops.fit(input_, fit_width)
input_actual = np.random.rand(batch, length, width) # pylint: disable=I0011,E1101
delete_idx = [width - (i + 1) for i in range(delta)]
output_expected = np.delete(input_actual, delete_idx, axis=2) # pylint: disable=I0011,E1101
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
output_actual = sess.run(output, {input_: input_actual})
self.assertAllClose(output_expected, output_actual)
def test_fit_to_greater_width(self):
"""Fit a tensor to a larger width (i.e. padding).
Given a 3D tensor of shape [batch, length, width], apply the
`ops.fit()` operator to it with the a larger `width` as the
target one and check that the tensor has now padded values
(with 0 as the padding value).
"""
batch = 2
length = 5
width = 4
fit_width = 5
shape = [None, None, None]
input_ = tf.placeholder(dtype=tf.float32, shape=shape)
output = ops.fit(input_, fit_width)
input_actual = np.random.rand(batch, length, width) # pylint: disable=I0011,E1101
output_expected = np.pad( # pylint: disable=I0011,E1101
input_actual,
((0, 0), (0, 0), (0, fit_width - width)),
mode='constant')
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
output_actual = sess.run(output, {input_: input_actual})
self.assertAllClose(output_expected, output_actual)
def next_inp(self, time, output):
"""Returns the next input.
Arguments:
time: a `int` or unit `Tensor` representing the current timestep.
output: a `2D Tensor` of shape `[batch_size, output_size]` representing
the current output.
*NOTE* that at time `t+1` the desired decoder input is the output
from the previous step, `t`, it means that at timestep `t` the next
input is the desired output for the very same timestep, if decoder
inputs have been provided -- otherwise is just the current output.
"""
if self._inputs_ta:
output = tf.cond(time < self._inputs_ta.size(),
lambda: self._inputs_ta.read(time),
lambda: self.zero_output()) # pylint: disable=W0108
next_inp = ops.fit(output, self._inp_size)
return next_inp
def test_fit_same_width(self):
"""Fit a tensor to a dimension which is the actual one.
Given a 3D tensor of shape [batch, length, width], apply the
`ops.fit()` operator to it with the same `width` as the target one
and check that the output tensor is the same of the input one.
"""
batch = 2
length = 5
width = 4
shape = [None, None, None]
input_ = tf.placeholder(dtype=tf.float32, shape=shape)
output = ops.fit(input_, width)
self.assertEqual(output.get_shape().as_list()[-1], width)
input_actual = np.random.rand(batch, length, width) # pylint: disable=I0011,E1101
output_expected = input_actual
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
output_actual = sess.run(output, {input_: input_actual})
self.assertAllClose(output_expected, output_actual)