def sample(self,
n,
max_length=None,
z=None,
temperature=1.0,
start_inputs=None,
end_fn=None):
"""Sample from decoder with an optional conditional latent vector `z`.
Args:
n: Scalar number of samples to return.
max_length: (Optional) Scalar maximum sample length to return. Required if
data representation does not include end tokens.
z: (Optional) Latent vectors to sample from. Required if model is
conditional. Sized `[n, z_size]`.
temperature: (Optional) The softmax temperature to use when sampling, if
applicable.
start_inputs: (Optional) Initial inputs to use for batch.
Sized `[n, output_depth]`.
end_fn: (Optional) A callable that takes a batch of samples (sized
`[n, output_depth]` and emits a `bool` vector
shaped `[batch_size]` indicating whether each sample is an end token.
Returns:
samples: Sampled sequences. Sized `[n, max_length, output_depth]`.
final_state: The final states of the decoder.
Raises:
ValueError: If `z` is provided and its first dimension does not equal `n`.
"""
if z is not None and z.shape[0].value != n:
raise ValueError(
'`z` must have a first dimension that equals `n` when given. '
'Got: %d vs %d' % (z.shape[0].value, n))
# Use a dummy Z in unconditional case.
z = tf.zeros((n, 0), tf.float32) if z is None else z
# If not given, start with zeros.
start_inputs = start_inputs if start_inputs is not None else tf.zeros(
[n, self._output_depth], dtype=tf.float32)
# In the conditional case, also concatenate the Z.
start_inputs = tf.concat([start_inputs, z], axis=-1)
sample_fn = lambda x: self._sample(x, temperature)
end_fn = end_fn or (lambda x: False)
# In the conditional case, concatenate Z to the sampled value.
next_inputs_fn = lambda x: tf.concat([x, z], axis=-1)
sampler = seq2seq.InferenceHelper(sample_fn,
sample_shape=[self._output_depth],
sample_dtype=tf.float32,
start_inputs=start_inputs,
end_fn=end_fn,
next_inputs_fn=next_inputs_fn)
decoder_outputs, final_state = self._decode(z,
helper=sampler,
max_length=max_length)
return decoder_outputs.sample_id, final_state
def sample(self,
n,
max_length=None,
z=None,
temperature=1.0,
start_inputs=None,
end_fn=None):
if z is not None and z.shape[0].value != n:
raise ValueError(
'`z` must have a first dimension that equals `n` when given. '
'Got: %d vs %d' % (z.shape[0].value, n))
# Use a dummy Z in unconditional case.
z = tf.zeros((n, 0), tf.float32) if z is None else z
# If not given, start with zeros.
start_inputs = start_inputs if start_inputs is not None else tf.zeros(
[n, self._output_depth], dtype=tf.float32)
# In the conditional case, also concatenate the Z.
start_inputs = tf.concat([start_inputs, z], axis=-1)
sample_fn = lambda x: self._sample(x, temperature)
end_fn = end_fn or (lambda x: False)
# In the conditional case, concatenate Z to the sampled value.
next_inputs_fn = lambda x: tf.concat([x, z], axis=-1)
sampler = seq2seq.InferenceHelper(sample_fn,
sample_shape=[self._output_depth],
sample_dtype=tf.float32,
start_inputs=start_inputs,
end_fn=end_fn,
next_inputs_fn=next_inputs_fn)
decoder_outputs = self._decode(n,
helper=sampler,
z=z,
max_length=max_length)
return decoder_outputs.sample_id
# 7. Output layer
out = tf.matmul(h_flat_drop, W_fc1_) + b_fc1_
if clas:
return tf.reshape(out, [b_size, seq_size, -1])
else:
return tf.reshape(out, [b_size, -1])
# In[18]:
# Helper without embedding, can add param: 'next_inputs_fn'
helper_infer = seq2seq.InferenceHelper(
sample_fn=(lambda x: x),
sample_shape=[letter_size],
sample_dtype=tf.float32,
start_inputs=tf.cast(tf.fill([batch_size, letter_size], LETTER_PAD), tf.float32), # PAD <- EOS, need flaot32
end_fn=(lambda sample_ids:
tf.equal(tf.argmax(CNN_2(sample_ids, False), axis=-1, output_type=tf.int32), 1)))
decoder_infer = seq2seq.BasicDecoder(
decoder_cell, helper_infer, decoder_initial_state,
output_layer=projection_layer)
# Dynamic decoding
outputs_infer, final_context_state, final_seq_lengths = seq2seq.dynamic_decode(
decoder_infer,
impute_finished=True,
maximum_iterations=output_length)
