ed172723345439f2636f2c8b5ed6634dfdb4cfb8,tf_agents/networks/value_network.py,ValueNetwork,call,#ValueNetwork#Any#Any#Any#,103

Before Change

                                           self.input_tensor_spec)
    batch_squash = utils.BatchSquash(outer_rank)

    states = tf.cast(tf.nest.flatten(observation)[0], tf.float32)
    states = batch_squash.flatten(states)
    for layer in self._postprocessing_layers:
      states = layer(states)

    value = tf.reshape(states, [-1])
    value = batch_squash.unflatten(value)
    return value, network_state

After Change

            minval=-0.03, maxval=0.03))

  def call(self, observation, step_type=None, network_state=()):
    state, network_state = self._encoder(
        observation, step_type=step_type, network_state=network_state)
    value = self._postprocessing_layers(state)
    return tf.squeeze(value, -1), network_state

Instances