State after ray.tune with Trajectory View API

RLlib
1

Hello, I am experimenting Trajectory View API and using the following example:

I have tried to run ray.tune and after that restore from checkpoint and run some episodes again.
The following is my code:

import argparse
import numpy as np

import ray
from ray import tune
from ray.rllib.examples.env.stateless_cartpole import StatelessCartPole
from ray.rllib.examples.models.trajectory_view_utilizing_models import \
    FrameStackingCartPoleModel, TorchFrameStackingCartPoleModel
from ray.rllib.models.catalog import ModelCatalog
from ray.rllib.utils.framework import try_import_tf
from ray.rllib.utils.test_utils import check_learning_achieved
from ray.rllib import agents

tf1, tf, tfv = try_import_tf()

parser = argparse.ArgumentParser()
parser.add_argument(
    "--run",
    type=str,
    default="PPO",
    help="The RLlib-registered algorithm to use.")
parser.add_argument(
    "--framework",
    choices=["tf", "tf2", "tfe", "torch"],
    default="tf",
    help="The DL framework specifier.")
parser.add_argument(
    "--as-test",
    action="store_true",
    help="Whether this script should be run as a test: --stop-reward must "
    "be achieved within --stop-timesteps AND --stop-iters.")
parser.add_argument(
    "--stop-iters",
    type=int,
    default=2,
    help="Number of iterations to train.")
parser.add_argument(
    "--stop-timesteps",
    type=int,
    default=200000,
    help="Number of timesteps to train.")
parser.add_argument(
    "--stop-reward",
    type=float,
    default=150.0,
    help="Reward at which we stop training.")

if __name__ == "__main__":
    args = parser.parse_args()
    ray.init(num_cpus=3)

    ModelCatalog.register_custom_model(
        "frame_stack_model", FrameStackingCartPoleModel
        if args.framework != "torch" else TorchFrameStackingCartPoleModel)
    tune.register_env("stateless_cartpole", lambda c: StatelessCartPole())

    config = {
        "env": "stateless_cartpole",
        "model": {
            # "vf_share_layers": True,
            # "custom_model": "frame_stack_model",
            # "custom_model_config": {
            #     "num_frames": 16,
            # },

            # To compare against a simple LSTM:
            "use_lstm": True,
            "lstm_cell_size": 128,
            "lstm_use_prev_action": True,
            "lstm_use_prev_reward": True,

            # To compare against a simple attention net:
            # "use_attention": True,
            # "attention_use_n_prev_actions": 1,
            # "attention_use_n_prev_rewards": 1,
        },
        "num_sgd_iter": 5,
        "vf_loss_coeff": 0.0001,
        "framework": args.framework,
    }

    stop = {
        "training_iteration": args.stop_iters,
        "timesteps_total": args.stop_timesteps,
        "episode_reward_mean": args.stop_reward,
    }
    results = tune.run(args.run,
        config=config,
        stop=stop,
        checkpoint_freq = 200,
        local_dir = 'lstm_trajectory',
        checkpoint_at_end = True,
        verbose=2)

    checkpoints = results.get_trial_checkpoints_paths(trial=results.get_best_trial('episode_reward_mean', mode='max'),
                                                                                    metric='episode_reward_mean')

    checkpoint_path = checkpoints[0][0]
    agent = agents.ppo.PPOTrainer(config, env="stateless_cartpole")
    agent.restore(checkpoint_path)

    env = StatelessCartPole()

    # run until episode ends
    for _ in range(10):
        episode_reward = 0
        done = False
        obs = env.reset()
        cell_size=128
        state=[np.zeros(cell_size, np.float32),
               np.zeros(cell_size, np.float32)]
        while not done:
            action, state, logits = agent.compute_action(obs, state)
            obs, reward, done, info = env.step(action)
            episode_reward += reward

        print("reward: {}".format(episode_reward))
        env.render()


    if args.as_test:
        check_learning_achieved(results, args.stop_reward)
    ray.shutdown()

This code fails with the following error:

2021-06-15 13:18:16,108	ERROR tf_run_builder.py:46 -- Error fetching: [<tf.Tensor 'default_policy/cond_1/Merge:0' shape=(?,) dtype=int64>, <tf.Tensor 'default_policy/model_1/lstm/while/Exit_3:0' shape=(?, 128) dtype=float32>, <tf.Tensor 'default_policy/model_1/lstm/while/Exit_4:0' shape=(?, 128) dtype=float32>, {'action_prob': <tf.Tensor 'default_policy/Exp:0' shape=(?,) dtype=float32>, 'action_logp': <tf.Tensor 'default_policy/cond_2/Merge:0' shape=(?,) dtype=float32>, 'action_dist_inputs': <tf.Tensor 'default_policy/Reshape_3:0' shape=(?, 2) dtype=float32>, 'vf_preds': <tf.Tensor 'default_policy/Reshape_5:0' shape=(?,) dtype=float32>}], feed_dict={<tf.Tensor 'default_policy/obs:0' shape=(?, 2) dtype=float32>: [array([-0.00997954, -0.01892092])], <tf.Tensor 'default_policy/seq_lens:0' shape=(?,) dtype=int32>: array([1.]), 
<tf.Tensor 'default_policy/Placeholder:0' shape=(?, 128) dtype=float32>: array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]],
      dtype=float32), <tf.Tensor 'default_policy/Placeholder_1:0' shape=(?, 128) dtype=float32>: array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]],
      dtype=float32), <tf.Tensor 'default_policy/is_training:0' shape=() dtype=bool>: False, <tf.Tensor 'default_policy/is_exploring:0' shape=() dtype=bool>: True, <tf.Tensor 'default_policy/timestep:0' shape=() dtype=int64>: 0}
Traceback (most recent call last):
  File "/home/condauser/.local/lib/python3.8/site-packages/tensorflow/python/client/session.py", line 1375, in _do_call
    return fn(*args)
  File "/home/condauser/.local/lib/python3.8/site-packages/tensorflow/python/client/session.py", line 1359, in _run_fn
    return self._call_tf_sessionrun(options, feed_dict, fetch_list,
  File "/home/condauser/.local/lib/python3.8/site-packages/tensorflow/python/client/session.py", line 1451, in _call_tf_sessionrun
    return tf_session.TF_SessionRun_wrapper(self._session, options, feed_dict,
tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'default_policy/prev_action' with dtype int64 and shape [?]
	 [[{{node default_policy/prev_action}}]]

Can anyone help me to understand how to create an initial state?

Thank you.

2

I think my question is not properly posted.
I am trying to use LSTM instead of Attention with Trajectory API.

3

This is more of an rllib question - @sven1977 take a look?

4

Hi @mg64ve,

In your config you set lstm_use_prev_{action,reward} to True which means you need to pass these in to compute actions.

You can find the api documentation here:
https://docs.ray.io/en/stable/rllib-package-ref.html#ray.rllib.agents.Agent.compute_action

5

Thanks @mannyv .
It seems the following code works:

import argparse
import numpy as np

import ray
from ray import tune, workers
from ray.rllib.examples.env.stateless_cartpole import StatelessCartPole
from ray.rllib.examples.models.trajectory_view_utilizing_models import \
    FrameStackingCartPoleModel, TorchFrameStackingCartPoleModel
from ray.rllib.models.catalog import ModelCatalog
from ray.rllib.utils.framework import try_import_tf
from ray.rllib.utils.test_utils import check_learning_achieved
from ray.rllib import agents

tf1, tf, tfv = try_import_tf()

parser = argparse.ArgumentParser()
parser.add_argument(
    "--run",
    type=str,
    default="PPO",
    help="The RLlib-registered algorithm to use.")
parser.add_argument(
    "--framework",
    choices=["tf", "tf2", "tfe", "torch"],
    default="tf",
    help="The DL framework specifier.")
parser.add_argument(
    "--as-test",
    action="store_true",
    help="Whether this script should be run as a test: --stop-reward must "
    "be achieved within --stop-timesteps AND --stop-iters.")
parser.add_argument(
    "--stop-iters",
    type=int,
    default=20,
    help="Number of iterations to train.")
parser.add_argument(
    "--stop-timesteps",
    type=int,
    default=200000,
    help="Number of timesteps to train.")
parser.add_argument(
    "--stop-reward",
    type=float,
    default=150.0,
    help="Reward at which we stop training.")

if __name__ == "__main__":
    args = parser.parse_args()
    ray.init(num_cpus=3)

    ModelCatalog.register_custom_model(
        "frame_stack_model", FrameStackingCartPoleModel
        if args.framework != "torch" else TorchFrameStackingCartPoleModel)
    tune.register_env("stateless_cartpole", lambda c: StatelessCartPole())

    config = {
        "env": "stateless_cartpole",
        "model": {
            # "vf_share_layers": True,
            # "custom_model": "frame_stack_model",
            # "custom_model_config": {
            #     "num_frames": 16,
            # },

            # To compare against a simple LSTM:
            "use_lstm": True,
            "lstm_cell_size": 128,
            "lstm_use_prev_action": True,
            "lstm_use_prev_reward": True,

            # To compare against a simple attention net:
            # "use_attention": True,
            # "attention_use_n_prev_actions": 1,
            # "attention_use_n_prev_rewards": 1,
        },
        "num_sgd_iter": 5,
        "vf_loss_coeff": 0.0001,
        "framework": args.framework,
    }

    stop = {
        "training_iteration": args.stop_iters,
        "timesteps_total": args.stop_timesteps,
        "episode_reward_mean": args.stop_reward,
    }

    results = tune.run(args.run,
        config=config,
        stop=stop,
        checkpoint_freq = 200,
        local_dir = 'lstm_trajectory',
        checkpoint_at_end = True,
        verbose=2)

    checkpoints = results.get_trial_checkpoints_paths(trial=results.get_best_trial('episode_reward_mean', mode='max'),
                                                                                    metric='episode_reward_mean')

    checkpoint_path = checkpoints[0][0]
    agent = agents.ppo.PPOTrainer(config, env="stateless_cartpole")
    agent.restore(checkpoint_path)

    env = StatelessCartPole()

    # run until episode ends
    for _ in range(10):
        episode_reward = 0
        reward = 0.
        action = 0
        done = False
        obs = env.reset()
        cell_size=128
        state=[np.zeros(cell_size, np.float32),
               np.zeros(cell_size, np.float32)]
        while not done:
            action, state, logits = agent.compute_action(obs, state, prev_action=action, prev_reward=reward)
            obs, reward, done, info = env.step(action)
            episode_reward += reward

        print("reward: {}".format(episode_reward))


    if args.as_test:
        check_learning_achieved(results, args.stop_reward)
    ray.shutdown()

I am still learning Ray and this is another little brick in my knowledge.
Thanks @mannyv

1 Like
6

@mannyv I am not sure if you can help me with the following code:

import argparse
import gym
import numpy as np

import ray
from ray import tune
from ray.rllib.examples.env.stateless_cartpole import StatelessCartPole
from ray.rllib.examples.models.trajectory_view_utilizing_models import \
    FrameStackingCartPoleModel, TorchFrameStackingCartPoleModel
from ray.rllib.models.catalog import ModelCatalog
from ray.rllib.utils.framework import try_import_tf
from ray.rllib.utils.test_utils import check_learning_achieved
from ray.rllib import agents

tf1, tf, tfv = try_import_tf()

parser = argparse.ArgumentParser()
parser.add_argument(
    "--run",
    type=str,
    default="PPO",
    help="The RLlib-registered algorithm to use.")
parser.add_argument(
    "--framework",
    choices=["tf", "tf2", "tfe", "torch"],
    default="tf",
    help="The DL framework specifier.")
parser.add_argument(
    "--as-test",
    action="store_true",
    help="Whether this script should be run as a test: --stop-reward must "
    "be achieved within --stop-timesteps AND --stop-iters.")
parser.add_argument(
    "--stop-iters",
    type=int,
    default=5,
    help="Number of iterations to train.")
parser.add_argument(
    "--stop-timesteps",
    type=int,
    default=200000,
    help="Number of timesteps to train.")
parser.add_argument(
    "--stop-reward",
    type=float,
    default=150.0,
    help="Reward at which we stop training.")

if __name__ == "__main__":
    args = parser.parse_args()
    ray.init(num_cpus=3)

    ModelCatalog.register_custom_model(
        "frame_stack_model", FrameStackingCartPoleModel
        if args.framework != "torch" else TorchFrameStackingCartPoleModel)
    tune.register_env("stateless_cartpole", lambda c: StatelessCartPole())

    config = {
        "env": "stateless_cartpole",
        "model": {
            "vf_share_layers": True,
            "custom_model": "frame_stack_model",
            "custom_model_config": {
                "num_frames": 16,
            },

            # To compare against a simple LSTM:
            # "use_lstm": True,
            # "lstm_use_prev_action": True,
            # "lstm_use_prev_reward": True,

            # To compare against a simple attention net:
            # "use_attention": True,
            # "attention_use_n_prev_actions": 1,
            # "attention_use_n_prev_rewards": 1,
        },
        "num_sgd_iter": 5,
        "vf_loss_coeff": 0.0001,
        "framework": args.framework,
    }

    stop = {
        "training_iteration": args.stop_iters,
        "timesteps_total": args.stop_timesteps,
        "episode_reward_mean": args.stop_reward,
    }

    results = tune.run(args.run,
        config=config,
        stop=stop,
        checkpoint_freq = 200,
        local_dir = 'attnet_trajectory',
        checkpoint_at_end = True,
        verbose=2)

    checkpoints = results.get_trial_checkpoints_paths(trial=results.get_best_trial('episode_reward_mean', mode='max'),
                                                                                    metric='episode_reward_mean')

    checkpoint_path = checkpoints[0][0]
    agent = agents.ppo.PPOTrainer(config, env="stateless_cartpole")
    agent.restore(checkpoint_path)

    env = StatelessCartPole()

    # run until episode ends
    for _ in range(10):
        episode_reward = 0
        reward = 0.
        action = 0
        done = False
        obs = env.reset()
        state=np.zeros(5*256, np.float32).reshape(5,256)
        while not done:
            action, state, logits = agent.compute_action(obs, state)
            obs, reward, done, info = env.step(action)
            episode_reward += reward

        print("reward: {}".format(episode_reward))


    if args.as_test:
        check_learning_achieved(results, args.stop_reward)
    ray.shutdown()

I do not know what dimension is correct for initial state.
Do you know how can I properly size state ?