def bottleneck_layer(targets_c, hparams):
"""Compute latents from compressed targets."""
latents_discrete_hot, extra_loss = discretization.parametrized_bottleneck(
targets_c, hparams)
latents_dense = discretization.parametrized_unbottleneck(
latents_discrete_hot, hparams.hidden_size, hparams)
latents_discrete = tf.argmax(latents_discrete_hot, axis=-1)
if DO_SUMMARIES:
tf.summary.histogram("b0", tf.reshape(latents_discrete, [-1]))
return latents_dense, latents_discrete, extra_loss
def bottleneck_layer(targets_c, hparams):
"""Compute latents from compressed targets."""
latents_discrete_hot, extra_loss = discretization.parametrized_bottleneck(
targets_c, hparams)
latents_dense = discretization.parametrized_unbottleneck(
latents_discrete_hot, hparams.hidden_size, hparams)
latents_dense = targets_c + tf.stop_gradient(latents_dense - targets_c)
latents_discrete = tf.argmax(latents_discrete_hot, axis=-1)
if DO_SUMMARIES:
tf.summary.histogram("b0", tf.reshape(latents_discrete, [-1]))
return latents_dense, latents_discrete_hot, extra_loss
def unbottleneck(self, x, res_size, reuse=None):
with tf.variable_scope("unbottleneck", reuse=reuse):
return discretization.parametrized_unbottleneck(x, res_size, self.hparams)
def unbottleneck(self, x, res_size): return discretization.parametrized_unbottleneck(x, res_size, self.hparams)
