sm64_env_cleanrl_curiosity_topdown.py

# Heavily based on this example https://docs.cleanrl.dev/rl-algorithms/ppo/#ppo_atari_lstmpy

import argparse
import importlib
import os
import random
import time
from distutils.util import strtobool

import gymnasium
import numpy as np
import supersuit as ss
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 env.sm64_env_curiosity import SM64_ENV_CURIOSITY
from tqdm import tqdm

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=1,
        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=True, 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="sm64",
        help="the id of the environment")
    parser.add_argument("--total-timesteps", type=int, default=50000000,
        help="total timesteps of the experiments")
    parser.add_argument("--learning-rate", type=float, default=3e-5,
        help="the learning rate of the optimizer")
    parser.add_argument("--num-envs", type=int, default=1,
        help="the number of parallel game environments")
    parser.add_argument("--num-steps", type=int, default=500,
        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=10,
        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=0.1,
        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=0.001,
        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.num_envs = env.MAX_PLAYERS

    # fmt: on
    return args


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(
            # No frame stack so just 1 as the first number
            layer_init(nn.Conv2d(1, 256, 8, stride=2)),
            nn.MaxPool2d(kernel_size=4, stride=2),
            nn.LeakyReLU(),
            layer_init(nn.Conv2d(256, 128, 4, stride=2)),
            nn.LeakyReLU(),
            layer_init(nn.Conv2d(128, 128, 2, stride=1)),
            nn.LeakyReLU(),
            nn.Flatten(),

            # 6272 calculated from torch_layer_size_test.py, given 4 channels and 128x72 input
            layer_init(nn.Linear(6272, 4096)),
            nn.LeakyReLU(),
            layer_init(nn.Linear(4096, 4096)),
            nn.LeakyReLU(),
        )
        self.lstm = nn.LSTM(4096, 2048)
        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)

        self.actor = nn.Sequential(
            layer_init(nn.Linear(2048,1024), std=0.01),
            nn.LeakyReLU(),
            layer_init(nn.Linear(1024,512), std=0.01),
            nn.LeakyReLU(),
            layer_init(nn.Linear(512, envs.single_action_space.n), std=0.01)
        )
        
        self.critic = nn.Sequential(
            layer_init(nn.Linear(2048,1024), std=0.01),
            nn.LeakyReLU(),
            layer_init(nn.Linear(1024,512), std=0.01),
            nn.LeakyReLU(),
            layer_init(nn.Linear(512, 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):
        x = x.clone()
        hidden, _ = self.get_states(x.permute((0, 3, 1, 2)), lstm_state, done)
        return self.critic(hidden)

    def get_action_and_value(self, x, lstm_state, done, action=None):
        # the /255 is done in get_states
        x = x.clone()
        hidden, lstm_state = self.get_states(x.permute((0, 3, 1, 2)), 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__":
    # extra moves added for the more complicated model 
    ACTION_BOOK = [
        # angle, A, B, Z
        # angle 0 is vertical
        # 8 directions
        [90,False,False,False],
        [-90,False,False,False],
        [0,False,False,False],
        [180,False,False,False],
        [45,False,False,False],
        [-45,False,False,False],
        [135,False,False,False],
        [-135,False,False,False],
        # 8 directions + jump
        [90,True,False,False],
        [-90,True,False,False],
        [0,True,False,False],
        [180,True,False,False],
        [45,True,False,False],
        [-45,True,False,False],
        [135,True,False,False],
        [-135,True,False,False],
    ]
    env = SM64_ENV_CURIOSITY(FRAME_SKIP=4, ACTION_BOOK=ACTION_BOOK,
                             NODES_MAX=3000, NODE_RADIUS= 400, NODES_MAX_VISITS=40, NODE_MAX_HEIGHT_ABOVE_GROUND=2000,
                            #  IMPORTANT
                            TOP_DOWN_CAMERA=True,
                            MAKE_OTHER_PLAYERS_INVISIBLE=False,
                            IMG_HEIGHT=84,
                            IMG_WIDTH=84
                            )
    envs = ss.black_death_v3(env)
    envs = ss.clip_reward_v0(envs, lower_bound=-1, upper_bound=1)
    envs = ss.color_reduction_v0(envs, mode="full")

    envs = ss.frame_stack_v1(envs, 1)
    envs = ss.pettingzoo_env_to_vec_env_v1(envs)

    envs = ss.concat_vec_envs_v1(envs, 1, num_cpus=99999, base_class="gymnasium")
    envs.single_observation_space = envs.observation_space
    envs.single_action_space = envs.action_space
    envs.is_vector_env = True

    args = parse_args()
    args.num_envs = env.MAX_PLAYERS


    args.batch_size = int(args.num_envs * args.num_steps)
    args.minibatch_size = int(args.batch_size // (args.num_minibatches) )

    args.num_iterations = args.total_timesteps // args.batch_size
    run_name = f"CURIOS_TOPDOWN_LSTMPPO_{int(time.time())}__LENGTH_{args.num_steps}_ACTIONS_{env.N_ACTIONS}_{env.IMG_WIDTH}x{env.IMG_HEIGHT}_PLAYERS_{env.MAX_PLAYERS}"

    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(wandb.run.dir)
    else:
        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")


    agent = Agent(envs).to(device)
    # if you want to load a model
    # agent.load_state_dict(torch.load(f"trained_models/agentCuriosityTopdown_14.8+19.6h.pt", map_location=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, _ = envs.reset(seed=args.seed)
    # next_obs = torch.Tensor(next_obs).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)

    for iteration in tqdm(range(1, args.num_iterations + 1)):
        # reset LSTM hidden units when episode begins
        # 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),
        # )
        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 - (iteration - 1.0) / args.num_iterations
            lrnow = frac * args.learning_rate
            optimizer.param_groups[0]["lr"] = lrnow

        ### Uncomment this if you want to reset every iteration
        o, infos = envs.reset()
        next_obs = torch.Tensor(o).to(device)
        # env.reset_nodes()

        next_done = torch.zeros(args.num_envs).to(device)
        for step in tqdm(range(0, args.num_steps),leave=False):
            global_step += 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.
            tmp = envs.step(action.cpu().numpy())
            next_obs, reward, done, truncations, infos = tmp[0], tmp[1], tmp[2], tmp[3], tmp[4]
  
            rewards[step] = torch.tensor(reward).to(device).view(-1)
            next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(done).to(device)
            

            #DO NOT RESET LSTM AFTER DEATH THIS DOESN'T WORK WELL AT ALL
            # for i in range(env.MAX_PLAYERS):
            #     if infos[i]["died"]:
            #         # when a player dies, reset the lstm state
            #         next_lstm_state[0][:,i,:] = 0
            #         next_lstm_state[1][:,i,:] = 0

        # 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 and 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)
        writer.add_scalar("charts/SPS", int(global_step / (time.time() - start_time)), global_step)
        writer.add_scalar("charts/avg_reward", torch.mean(torch.mean(rewards)), global_step)
        writer.add_scalar("charts/n_nodes", infos[0]["node_index"], global_step)

        # make sure to tune `CHECKPOINT_FREQUENCY` 
        # so models are not saved too frequently
        if args.track:
            if iteration % 60 == 0:
                t = time.time() - start_time
                t_hours = round(t / 3600, 1)
                torch.save(agent.state_dict(), f"{wandb.run.dir}/agentCuriosity_{t_hours}h.pt")
                wandb.save(f"{wandb.run.dir}/agentCuriosity_{t_hours}h.pt", policy="now", base_path=wandb.run.dir)
    envs.close()
    writer.close()