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

Before Change

        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(
            x={name: state[:-1] for name, state in states.items()},

After Change

        :return: A list of deltas per action
        
        for _ in range(util.rank(q_value) - 1):
            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)

Instances