ppo+lstm train continuous environments

1900360 commented 1 year ago

I need to apply ppo+lstm in continuous environment recently, can you provide this kind of functionality :)

vwxyzjn commented 1 year ago

Hey, CleanRL is pretty hackable. Feel free to combine https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/ppo_continuous_action.py And https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/ppo_atari_lstm.py

1900360 commented 1 year ago

Hi @vwxyzjn! Thank you very much, I will refer to these two files to make changes, please keep in touch :)

1900360 commented 1 year ago

Hi @vwxyzjn! I made some changes to these two codes, but I get the following error, do you know where the error is?

# docs and experiment results can be found at https://docs.cleanrl.dev/rl-algorithms/ppo/#ppo_atari_lstmpy
import argparse
import os
import random
import time
from distutils.util import strtobool

import gym
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.distributions.categorical import Categorical
from torch.utils.tensorboard import SummaryWriter

from stable_baselines3.common.atari_wrappers import (  # isort:skip
    ClipRewardEnv,
    EpisodicLifeEnv,
    FireResetEnv,
    MaxAndSkipEnv,
    NoopResetEnv,
)

def parse_args():
    # fmt: off
    parser = argparse.ArgumentParser()
    parser.add_argument("--exp-name", type=str, default=os.path.basename(__file__).rstrip(".py"),
        help="the name of this experiment")
    parser.add_argument("--seed", type=int, default=2,
        help="seed of the experiment")
    parser.add_argument("--torch-deterministic", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
        help="if toggled, `torch.backends.cudnn.deterministic=False`")
    parser.add_argument("--cuda", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
        help="if toggled, cuda will be enabled by default")
    parser.add_argument("--track", type=lambda x: bool(strtobool(x)), default=False, nargs="?", const=True,
        help="if toggled, this experiment will be tracked with Weights and Biases")
    parser.add_argument("--wandb-project-name", type=str, default="cleanRL",
        help="the wandb's project name")
    parser.add_argument("--wandb-entity", type=str, default=None,
        help="the entity (team) of wandb's project")
    parser.add_argument("--capture-video", type=lambda x: bool(strtobool(x)), default=False, nargs="?", const=True,
        help="whether to capture videos of the agent performances (check out `videos` folder)")

    # Algorithm specific arguments
    parser.add_argument("--env-id", type=str, default="Pendulum-v1",
        help="the id of the environment")
    parser.add_argument("--total-timesteps", type=int, default=1000000,
        help="total timesteps of the experiments")
    parser.add_argument("--learning-rate", type=float, default=3e-4,
        help="the learning rate of the optimizer")
    parser.add_argument("--num-envs", type=int, default=10,
        help="the number of parallel game environments")
    parser.add_argument("--num-steps", type=int, default=200,
        help="the number of steps to run in each environment per policy rollout")
    parser.add_argument("--anneal-lr", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
        help="Toggle learning rate annealing for policy and value networks")
    parser.add_argument("--gamma", type=float, default=0.99,
        help="the discount factor gamma")
    parser.add_argument("--gae-lambda", type=float, default=0.95,
        help="the lambda for the general advantage estimation")
    parser.add_argument("--num-minibatches", type=int, default=32,
        help="the number of mini-batches")
    parser.add_argument("--update-epochs", type=int, default=10,
        help="the K epochs to update the policy")
    parser.add_argument("--norm-adv", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
        help="Toggles advantages normalization")
    parser.add_argument("--clip-coef", type=float, default=0.2,
        help="the surrogate clipping coefficient")
    parser.add_argument("--clip-vloss", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
        help="Toggles whether or not to use a clipped loss for the value function, as per the paper.")
    parser.add_argument("--ent-coef", type=float, default=0.0,
        help="coefficient of the entropy")
    parser.add_argument("--vf-coef", type=float, default=0.5,
        help="coefficient of the value function")
    parser.add_argument("--max-grad-norm", type=float, default=0.5,
        help="the maximum norm for the gradient clipping")
    parser.add_argument("--target-kl", type=float, default=None,
        help="the target KL divergence threshold")
    args = parser.parse_args()
    args.batch_size = int(args.num_envs * args.num_steps)
    args.minibatch_size = int(args.batch_size // args.num_minibatches)
    # fmt: on
    return args

def make_env(env_id, seed, idx, capture_video, run_name,gamma):
    def thunk():
        env = gym.make(env_id)
        env = gym.wrappers.RecordEpisodeStatistics(env)
        if capture_video:
            if idx == 0:
                env = gym.wrappers.RecordVideo(env, f"videos/{run_name}")

        # env = NoopResetEnv(env, noop_max=30)
        # env = MaxAndSkipEnv(env, skip=4)
        # env = EpisodicLifeEnv(env)
        # if "FIRE" in env.unwrapped.get_action_meanings():
        #     env = FireResetEnv(env)
        # env = ClipRewardEnv(env)
        # env = gym.wrappers.ResizeObservation(env, (84, 84))
        # env = gym.wrappers.GrayScaleObservation(env)
        # env = gym.wrappers.FrameStack(env, 1)
        env = gym.wrappers.ClipAction(env)
        env = gym.wrappers.NormalizeObservation(env)
        env = gym.wrappers.TransformObservation(env, lambda obs: np.clip(obs, -10, 10))
        env = gym.wrappers.NormalizeReward(env, gamma=gamma)
        env = gym.wrappers.TransformReward(env, lambda reward: np.clip(reward, -10, 10))

        env.seed(seed)
        env.action_space.seed(seed)
        env.observation_space.seed(seed)
        return env

    return thunk

def layer_init(layer, std=np.sqrt(2), bias_const=0.0):
    torch.nn.init.orthogonal_(layer.weight, std)
    torch.nn.init.constant_(layer.bias, bias_const)
    return layer

class Agent(nn.Module):
    def __init__(self, envs):
        super().__init__()
        self.network = nn.Sequential(
            layer_init(nn.Conv2d(1, 32, 8, stride=4)),
            nn.ReLU(),
            layer_init(nn.Conv2d(32, 64, 4, stride=2)),
            nn.ReLU(),
            layer_init(nn.Conv2d(64, 64, 3, stride=1)),
            nn.ReLU(),
            nn.Flatten(),
            layer_init(nn.Linear(64 * 7 * 7, 512)),
            nn.ReLU(),
        )
        self.lstm = nn.LSTM(512, 128)
        for name, param in self.lstm.named_parameters():
            if "bias" in name:
                nn.init.constant_(param, 0)
            elif "weight" in name:
                nn.init.orthogonal_(param, 1.0)
        # print(np.prod(envs.single_action_space.shape))
        self.actor = layer_init(nn.Linear(128, np.prod(envs.single_action_space.shape)), std=0.01)
        self.critic = layer_init(nn.Linear(128, 1), std=1)

    def get_states(self, x, lstm_state, done):
        hidden = self.network(x / 255.0)

        # LSTM logic
        batch_size = lstm_state[0].shape[1]
        hidden = hidden.reshape((-1, batch_size, self.lstm.input_size))
        done = done.reshape((-1, batch_size))
        new_hidden = []
        for h, d in zip(hidden, done):
            h, lstm_state = self.lstm(
                h.unsqueeze(0),
                (
                    (1.0 - d).view(1, -1, 1) * lstm_state[0],
                    (1.0 - d).view(1, -1, 1) * lstm_state[1],
                ),
            )
            new_hidden += [h]
        new_hidden = torch.flatten(torch.cat(new_hidden), 0, 1)
        return new_hidden, lstm_state

    def get_value(self, x, lstm_state, done):
        hidden, _ = self.get_states(x, lstm_state, done)
        return self.critic(hidden)

    def get_action_and_value(self, x, lstm_state, done, action=None):
        hidden, lstm_state = self.get_states(x, lstm_state, done)
        logits = self.actor(hidden)
        probs = Categorical(logits=logits)
        if action is None:
            action = probs.sample()
        return action, probs.log_prob(action), probs.entropy(), self.critic(hidden), lstm_state

if __name__ == "__main__":
    args = parse_args()
    run_name = f"{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}"
    if args.track:
        import wandb

        wandb.init(
            project=args.wandb_project_name,
            entity=args.wandb_entity,
            sync_tensorboard=True,
            config=vars(args),
            name=run_name,
            monitor_gym=True,
            save_code=True,
        )
    writer = SummaryWriter(f"runs/{run_name}")
    writer.add_text(
        "hyperparameters",
        "|param|value|\n|-|-|\n%s" % ("\n".join([f"|{key}|{value}|" for key, value in vars(args).items()])),
    )

    # TRY NOT TO MODIFY: seeding
    random.seed(args.seed)
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    torch.backends.cudnn.deterministic = args.torch_deterministic

    device = torch.device("cuda" if torch.cuda.is_available() and args.cuda else "cpu")

    # env setup
    envs = gym.vector.SyncVectorEnv(
        [make_env(args.env_id, args.seed + i, i, args.capture_video, run_name,args.gamma) for i in range(args.num_envs)]
    )
    # assert isinstance(envs.single_action_space, gym.spaces.Discrete), "only discrete action space is supported"
    assert isinstance(envs.single_action_space, gym.spaces.Box), "only continuous action space is supported"

    agent = Agent(envs).to(device)
    optimizer = optim.Adam(agent.parameters(), lr=args.learning_rate, eps=1e-5)

    # ALGO Logic: Storage setup
    obs = torch.zeros((args.num_steps, args.num_envs) + envs.single_observation_space.shape).to(device)
    actions = torch.zeros((args.num_steps, args.num_envs) + envs.single_action_space.shape).to(device)
    logprobs = torch.zeros((args.num_steps, args.num_envs)).to(device)
    rewards = torch.zeros((args.num_steps, args.num_envs)).to(device)
    dones = torch.zeros((args.num_steps, args.num_envs)).to(device)
    values = torch.zeros((args.num_steps, args.num_envs)).to(device)

    # TRY NOT TO MODIFY: start the game
    global_step = 0
    start_time = time.time()
    next_obs = torch.Tensor(envs.reset()).to(device)
    next_done = torch.zeros(args.num_envs).to(device)
    next_lstm_state = (
        torch.zeros(agent.lstm.num_layers, args.num_envs, agent.lstm.hidden_size).to(device),
        torch.zeros(agent.lstm.num_layers, args.num_envs, agent.lstm.hidden_size).to(device),
    )  # hidden and cell states (see https://youtu.be/8HyCNIVRbSU)
    num_updates = args.total_timesteps // args.batch_size

    for update in range(1, num_updates + 1):
        initial_lstm_state = (next_lstm_state[0].clone(), next_lstm_state[1].clone())
        # Annealing the rate if instructed to do so.
        if args.anneal_lr:
            frac = 1.0 - (update - 1.0) / num_updates
            lrnow = frac * args.learning_rate
            optimizer.param_groups[0]["lr"] = lrnow

        for step in range(0, args.num_steps):
            global_step += 1 * args.num_envs
            obs[step] = next_obs
            dones[step] = next_done

            # ALGO LOGIC: action logic
            with torch.no_grad():
                action, logprob, _, value, next_lstm_state = agent.get_action_and_value(next_obs, next_lstm_state, next_done)
                values[step] = value.flatten()
            actions[step] = action
            logprobs[step] = logprob

            # TRY NOT TO MODIFY: execute the game and log data.
            next_obs, reward, done, info = envs.step(action.cpu().numpy())
            rewards[step] = torch.tensor(reward).to(device).view(-1)
            next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(done).to(device)

            for item in info:
                if "episode" in item.keys():
                    print(f"global_step={global_step}, episodic_return={item['episode']['r']}")
                    writer.add_scalar("charts/episodic_return", item["episode"]["r"], global_step)
                    writer.add_scalar("charts/episodic_length", item["episode"]["l"], global_step)
                    break

        # bootstrap value if not done
        with torch.no_grad():
            next_value = agent.get_value(
                next_obs,
                next_lstm_state,
                next_done,
            ).reshape(1, -1)
            advantages = torch.zeros_like(rewards).to(device)
            lastgaelam = 0
            for t in reversed(range(args.num_steps)):
                if t == args.num_steps - 1:
                    nextnonterminal = 1.0 - next_done
                    nextvalues = next_value
                else:
                    nextnonterminal = 1.0 - dones[t + 1]
                    nextvalues = values[t + 1]
                delta = rewards[t] + args.gamma * nextvalues * nextnonterminal - values[t]
                advantages[t] = lastgaelam = delta + args.gamma * args.gae_lambda * nextnonterminal * lastgaelam
            returns = advantages + values

        # flatten the batch
        b_obs = obs.reshape((-1,) + envs.single_observation_space.shape)
        b_logprobs = logprobs.reshape(-1)
        b_actions = actions.reshape((-1,) + envs.single_action_space.shape)
        b_dones = dones.reshape(-1)
        b_advantages = advantages.reshape(-1)
        b_returns = returns.reshape(-1)
        b_values = values.reshape(-1)

        # Optimizing the policy and value network
        assert args.num_envs % args.num_minibatches == 0
        envsperbatch = args.num_envs // args.num_minibatches
        envinds = np.arange(args.num_envs)
        flatinds = np.arange(args.batch_size).reshape(args.num_steps, args.num_envs)
        clipfracs = []
        for epoch in range(args.update_epochs):
            np.random.shuffle(envinds)
            for start in range(0, args.num_envs, envsperbatch):
                end = start + envsperbatch
                mbenvinds = envinds[start:end]
                mb_inds = flatinds[:, mbenvinds].ravel()  # be really careful about the index

                _, newlogprob, entropy, newvalue, _ = agent.get_action_and_value(
                    b_obs[mb_inds],
                    (initial_lstm_state[0][:, mbenvinds], initial_lstm_state[1][:, mbenvinds]),
                    b_dones[mb_inds],
                    b_actions.long()[mb_inds],
                )
                logratio = newlogprob - b_logprobs[mb_inds]
                ratio = logratio.exp()

                with torch.no_grad():
                    # calculate approx_kl http://joschu.net/blog/kl-approx.html
                    old_approx_kl = (-logratio).mean()
                    approx_kl = ((ratio - 1) - logratio).mean()
                    clipfracs += [((ratio - 1.0).abs() > args.clip_coef).float().mean().item()]

                mb_advantages = b_advantages[mb_inds]
                if args.norm_adv:
                    mb_advantages = (mb_advantages - mb_advantages.mean()) / (mb_advantages.std() + 1e-8)

                # Policy loss
                pg_loss1 = -mb_advantages * ratio
                pg_loss2 = -mb_advantages * torch.clamp(ratio, 1 - args.clip_coef, 1 + args.clip_coef)
                pg_loss = torch.max(pg_loss1, pg_loss2).mean()

                # Value loss
                newvalue = newvalue.view(-1)
                if args.clip_vloss:
                    v_loss_unclipped = (newvalue - b_returns[mb_inds]) ** 2
                    v_clipped = b_values[mb_inds] + torch.clamp(
                        newvalue - b_values[mb_inds],
                        -args.clip_coef,
                        args.clip_coef,
                    )
                    v_loss_clipped = (v_clipped - b_returns[mb_inds]) ** 2
                    v_loss_max = torch.max(v_loss_unclipped, v_loss_clipped)
                    v_loss = 0.5 * v_loss_max.mean()
                else:
                    v_loss = 0.5 * ((newvalue - b_returns[mb_inds]) ** 2).mean()

                entropy_loss = entropy.mean()
                loss = pg_loss - args.ent_coef * entropy_loss + v_loss * args.vf_coef

                optimizer.zero_grad()
                loss.backward()
                nn.utils.clip_grad_norm_(agent.parameters(), args.max_grad_norm)
                optimizer.step()

            if args.target_kl is not None:
                if approx_kl > args.target_kl:
                    break

        y_pred, y_true = b_values.cpu().numpy(), b_returns.cpu().numpy()
        var_y = np.var(y_true)
        explained_var = np.nan if var_y == 0 else 1 - np.var(y_true - y_pred) / var_y

        # TRY NOT TO MODIFY: record rewards for plotting purposes
        writer.add_scalar("charts/learning_rate", optimizer.param_groups[0]["lr"], global_step)
        writer.add_scalar("losses/value_loss", v_loss.item(), global_step)
        writer.add_scalar("losses/policy_loss", pg_loss.item(), global_step)
        writer.add_scalar("losses/entropy", entropy_loss.item(), global_step)
        writer.add_scalar("losses/old_approx_kl", old_approx_kl.item(), global_step)
        writer.add_scalar("losses/approx_kl", approx_kl.item(), global_step)
        writer.add_scalar("losses/clipfrac", np.mean(clipfracs), global_step)
        writer.add_scalar("losses/explained_variance", explained_var, global_step)
        print("SPS:", int(global_step / (time.time() - start_time)))
        writer.add_scalar("charts/SPS", int(global_step / (time.time() - start_time)), global_step)

    envs.close()
    writer.close()

vwxyzjn commented 1 year ago

Hi @1900360, thanks for giving it a try. Some high-level comments:

The NN should be taken from ppo_continuous_action.py, but the storage and training logic should be taken ppo_atari_lstm.py
in the make_env function you don't need the Atari preprocessing wrappers.

You may also find our blog post pretty helpful. Doing a file diff should help you identify what is needed to implement LSTM and you can transfer that knowledge to ppo_continuous_action.py.

1900360 commented 1 year ago

Hi @vwxyzjn ! I still don't get the gist, the changed Agent function looks like this:

class Agent(nn.Module):
    def __init__(self, envs):
        super().__init__()
        # self.network = nn.Sequential(
        #     layer_init(nn.Conv2d(1, 32, 8, stride=4)),
        #     nn.ReLU(),
        #     layer_init(nn.Conv2d(32, 64, 4, stride=2)),
        #     nn.ReLU(),
        #     layer_init(nn.Conv2d(64, 64, 3, stride=1)),
        #     nn.ReLU(),
        #     nn.Flatten(),
        #     layer_init(nn.Linear(64 * 7 * 7, 512)),
        #     nn.ReLU(),
        # )
        self.network = nn.Sequential(
            # layer_init(nn.Conv2d(1, 32, 8, stride=4)),
            # nn.ReLU(),
            # layer_init(nn.Conv2d(32, 64, 4, stride=2)),
            # nn.ReLU(),
            layer_init(nn.Linear(np.array(envs.single_observation_space.shape).prod(), 64)),
            nn.Tanh(),
            layer_init(nn.Linear(64, 64)),
            nn.Tanh(),
            layer_init(nn.Linear(64, 1), std=1.0),
        )

        # print(self.network)
        self.lstm = nn.LSTM(512, 128)
        for name, param in self.lstm.named_parameters():
            if "bias" in name:
                nn.init.constant_(param, 0)
            elif "weight" in name:
                nn.init.orthogonal_(param, 1.0)
        # # print(np.prod(envs.single_action_space.shape))
        # self.actor = layer_init(nn.Linear(128, np.prod(envs.single_action_space.shape)), std=0.01)
        # self.critic = layer_init(nn.Linear(128, 1), std=1)
        self.critic = nn.Sequential(
            layer_init(nn.Linear(np.array(envs.single_observation_space.shape).prod(), 64)),
            nn.Tanh(),
            layer_init(nn.Linear(64, 64)),
            nn.Tanh(),
            layer_init(nn.Linear(64, 1), std=1.0),
        )
        self.actor_mean = nn.Sequential(
            layer_init(nn.Linear(np.array(envs.single_observation_space.shape).prod(), 64)),
            nn.Tanh(),
            layer_init(nn.Linear(64, 64)),
            nn.Tanh(),
            layer_init(nn.Linear(64, np.prod(envs.single_action_space.shape)), std=0.01),
        )
        self.actor_logstd = nn.Parameter(torch.zeros(1, np.prod(envs.single_action_space.shape)))

    def get_states(self, x, lstm_state, done):
        # hidden = self.network(x / 255.0)
        hidden = self.network(x)
        print(hidden)

        # LSTM logic
        batch_size = lstm_state[0].shape[1]
        hidden = hidden.reshape((-1, batch_size, self.lstm.input_size))
        done = done.reshape((-1, batch_size))
        new_hidden = []
        for h, d in zip(hidden, done):
            h, lstm_state = self.lstm(
                h.unsqueeze(0),
                (
                    (1.0 - d).view(1, -1, 1) * lstm_state[0],
                    (1.0 - d).view(1, -1, 1) * lstm_state[1],
                ),
            )
            new_hidden += [h]
        new_hidden = torch.flatten(torch.cat(new_hidden), 0, 1)
        return new_hidden, lstm_state

    def get_value(self, x, lstm_state, done):
        hidden, _ = self.get_states(x, lstm_state, done)
        return self.critic(hidden)

    def get_action_and_value(self, x, lstm_state, done, action=None):
        hidden, lstm_state = self.get_states(x, lstm_state, done)
        # logits = self.actor(hidden)
        # probs = Categorical(logits=logits)

        # action_mean = self.actor_mean(x)
        action_mean = self.actor_mean(hidden)
        action_logstd = self.actor_logstd.expand_as(action_mean)
        action_std = torch.exp(action_logstd)
        probs = Normal(action_mean, action_std)
        if action is None:
            action = probs.sample()
        return action, probs.log_prob(action).sum(1), probs.entropy().sum(1), self.critic(hidden), lstm_state

But the error is as follows:

Traceback (most recent call last):
  File "D:\desktop\lunwen_dabao\xinsuanfa0912\cleanrl-master\cleanrl\ppo_atari_lstm.py", line 292, in <module>
    action, logprob, _, value, next_lstm_state = agent.get_action_and_value(next_obs, next_lstm_state, next_done)
  File "D:\desktop\lunwen_dabao\xinsuanfa0912\cleanrl-master\cleanrl\ppo_atari_lstm.py", line 204, in get_action_and_value
    hidden, lstm_state = self.get_states(x, lstm_state, done)
  File "D:\desktop\lunwen_dabao\xinsuanfa0912\cleanrl-master\cleanrl\ppo_atari_lstm.py", line 184, in get_states
    hidden = hidden.reshape((-1, batch_size, self.lstm.input_size))
RuntimeError: shape '[-1, 10, 512]' is invalid for input of size 10

It seems that there is a problem with the NN settings, causing hidden not to be converted correctly, can you help me modify it :)

1900360 commented 1 year ago

Hi @vwxyzjn ! Could you give me more detail about LSTM? I'm not so familiar with it:)

vwxyzjn commented 1 year ago

You were on the right track! I'd suggest printing out the hidden layers to see what their shapes look like and see if you can connect them. The hidden = self.network(x) gives you a shape of (batch_size, 1) which could not be plugged into self.lstm = nn.LSTM(512, 128), where self.lstm is expecting an input of size (batch_size, 512)

In the shared network's case, the LSTM core is used after the output of the shared layers. Using LSTM in the case of a separate network (e.g., ppo_continuous_action.py) has multiple implementations

use lstm core for both the actor and the critic
use lstm core for the actor only
use lstm core for the critic only

I suggest considering these design choices and thinking about how to implement them, each of which will have different performance considerations (e.g., SPS, sample efficiency)

Here is a prototype that uses lstm core for the actor only, which may be helpful to you after you have tried to play with your own implementation a bit more and understand the design decisions. The performance is not great (see tracked run here), but it might give you a good starting point.

If you really want to have a good LSTM implementation for robotics, it's worth doing extensive benchmarks and experiments on the three implementation variants and comparing them to the non-lstm baselines. If you are interested in this, the process is basically listed out in our contribution guideline.

Good luck.

vwxyzjn commented 1 year ago

Closing this issue now, but feel free to open a PR if you are interested in exploring this — having a rigorous benchmark and experiment design will help ensure the implementation is correct.

vwxyzjn / cleanrl

ppo+lstm train continuous environments #290