da6b040e254710f0df907d838db51f4f2cc9bf93,social_rl/multiagent_tfagents/multiagent_ppo.py,MultiagentPPO,get_single_agent_specs,#MultiagentPPO#Any#Any#,217

Before Change

    for k in time_step_spec.observation.keys():
      if k == "direction":
        shape = [1]
      elif k == "image":
        // Remove agent dimension
        shape = time_step_spec.observation[k].shape[1:]
      else:

After Change

          maximum=spec.maximum,
          dtype=spec.dtype)

    single_obs_spec = tf.nest.map_structure(make_single_agent_spec,
                                            time_step_spec.observation)
    single_reward_spec = tensor_spec.TensorSpec(
        shape=(), dtype=time_step_spec.reward.dtype, name="reward")
    single_time_step_spec = ts_lib.TimeStep(time_step_spec.step_type,
                                            single_reward_spec,

Instances