b5ae1c6bde39f5130c975992198f1f8ee5200f9a,tf_agents/agents/dqn/dqn_agent.py,DqnAgent,_check_action_spec,#DqnAgent#Any#,265

Before Change

    // TODO(b/119321125): Disable this once index_with_actions supports
    // negative-valued actions.
    if not all(spec.minimum == 0 for spec in flat_action_spec):
      raise ValueError(
          "Action specs should have minimum of 0, but saw: {0}".format(
              [spec.minimum for spec in flat_action_spec]))

  def _setup_policy(self, time_step_spec, action_spec,
                    boltzmann_temperature, emit_log_probability):

    policy = q_policy.QPolicy(

After Change

          "Only scalar actions are supported now, but action spec is: {}"
          .format(action_spec))

    spec = flat_action_spec[0]

    // TODO(b/119321125): Disable this once index_with_actions supports
    // negative-valued actions.
    if spec.minimum != 0:
      raise ValueError(
          "Action specs should have minimum of 0, but saw: {0}".format(spec))

    self._num_actions = spec.maximum - spec.minimum + 1

  def _check_network_output(self, net, label):
    Check outputs of q_net and target_q_net against expected shape.

Instances