fe8d55ae1870b6c7c3725f7ac7cc6b1d1ac0d39e,tensorforce/models/q_model.py,QModel,tf_q_delta,#QModel#Any#Any#Any#Any#,59

Before Change

        next_q_value = tf.where(condition=terminal, x=zeros, y=(self.discount * next_q_value))

        delta = reward + next_q_value - q_value  // tf.stop_gradient(q_target)
        collapsed_size = util.prod(util.shape(delta)[1:])
        return tf.reshape(tensor=delta, shape=(-1, collapsed_size))

    def tf_loss_per_instance(self, states, internals, actions, terminal, reward):
        embedding = self.network.apply(

After Change

            terminal = tf.expand_dims(input=terminal, axis=1)
            reward = tf.expand_dims(input=reward, axis=1)

        multiples = (1,) + util.shape(q_value)[1:]
        terminal = tf.tile(input=terminal, multiples=multiples)
        reward = tf.tile(input=reward, multiples=multiples)

        zeros = tf.zeros_like(tensor=next_q_value)
        next_q_value = tf.where(condition=terminal, x=zeros, y=(self.discount * next_q_value))

Instances