polish(pu): rename muzero_gpt to unizero, add muzero_context variant

lzero/entry/__init__.py

@@ -1,10 +1,12 @@
 from .train_alphazero import train_alphazero
 from .eval_alphazero import eval_alphazero
 from .train_muzero import train_muzero
+from .train_muzero_context import train_muzero_context
+
 from .train_muzero_multi_task import train_muzero_multi_task 
 
-from .train_muzero_gpt import train_muzero_gpt
-from .train_muzero_gpt_multi_task import train_muzero_gpt_multi_task 
+from .train_unizero import train_unizero
+from .train_unizero_multi_task import train_unizero_multi_task 
 
 from .train_muzero_with_reward_model import train_muzero_with_reward_model
 from .eval_muzero import eval_muzero

lzero/entry/eval_muzero.py

@@ -13,7 +13,7 @@
 from ding.utils import set_pkg_seed
 from ding.worker import BaseLearner
 from lzero.worker import MuZeroEvaluator
-from lzero.entry.train_muzero_gpt import initialize_zeros_batch
+from lzero.entry.train_unizero import initialize_zeros_batch
 
 def eval_muzero(
         input_cfg: Tuple[dict, dict],
@@ -38,7 +38,7 @@ def eval_muzero(
         - policy (:obj:`Policy`): Converged policy.
     """
     cfg, create_cfg = input_cfg
-    assert create_cfg.policy.type in ['efficientzero', 'muzero', 'muzero_gpt', 'stochastic_muzero', 'gumbel_muzero', 'sampled_efficientzero'], \
+    assert create_cfg.policy.type in ['efficientzero', 'muzero', 'unizero', 'stochastic_muzero', 'gumbel_muzero', 'sampled_efficientzero'], \
         "LightZero now only support the following algo.: 'efficientzero', 'muzero', 'stochastic_muzero', 'gumbel_muzero', 'sampled_efficientzero'"
 
     if cfg.policy.cuda and torch.cuda.is_available():

lzero/entry/train_muzero.py

@@ -47,10 +47,10 @@ def train_muzero(
     """
 
     cfg, create_cfg = input_cfg
-    assert create_cfg.policy.type in ['efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
+    assert create_cfg.policy.type in ['efficientzero', 'muzero', 'muzero_context','sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
         "train_muzero entry now only support the following algo.: 'efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'"
 
-    if create_cfg.policy.type == 'muzero':
+    if create_cfg.policy.type in ['muzero', 'muzero_context']:
         from lzero.mcts import MuZeroGameBuffer as GameBuffer
     elif create_cfg.policy.type == 'efficientzero':
         from lzero.mcts import EfficientZeroGameBuffer as GameBuffer

lzero/entry/train_muzero_context.py

@@ -0,0 +1,255 @@
+import logging
+import os
+from functools import partial
+from typing import Optional, Tuple
+
+import torch
+from ding.config import compile_config
+from ding.envs import create_env_manager
+from ding.envs import get_vec_env_setting
+from ding.policy import create_policy
+from ding.utils import set_pkg_seed, get_rank
+from ding.rl_utils import get_epsilon_greedy_fn
+from ding.worker import BaseLearner
+from tensorboardX import SummaryWriter
+
+from lzero.entry.utils import log_buffer_memory_usage
+from lzero.policy import visit_count_temperature
+from lzero.policy.random_policy import LightZeroRandomPolicy
+from lzero.worker import MuZeroCollector as Collector
+from lzero.worker import MuZeroEvaluator as Evaluator
+from .utils import random_collect
+
+def initialize_zeros_batch(observation_shape, batch_size, device):
+    """Initialize a zeros tensor for batch observations based on the shape."""
+    if isinstance(observation_shape, list):
+        shape = [batch_size, *observation_shape]
+    elif isinstance(observation_shape, int):
+        shape = [batch_size, observation_shape]
+    else:
+        raise TypeError("observation_shape must be either an int or a list")
+
+    return torch.zeros(shape).to(device)
+
+def train_muzero_context(
+        input_cfg: Tuple[dict, dict],
+        seed: int = 0,
+        model: Optional[torch.nn.Module] = None,
+        model_path: Optional[str] = None,
+        max_train_iter: Optional[int] = int(1e10),
+        max_env_step: Optional[int] = int(1e10),
+) -> 'Policy':  # noqa
+    """
+    Overview:
+        The train entry for MCTS+RL algorithms, including MuZero, EfficientZero, Sampled EfficientZero, Gumbel Muzero.
+    Arguments:
+        - input_cfg (:obj:`Tuple[dict, dict]`): Config in dict type.
+            ``Tuple[dict, dict]`` type means [user_config, create_cfg].
+        - seed (:obj:`int`): Random seed.
+        - model (:obj:`Optional[torch.nn.Module]`): Instance of torch.nn.Module.
+        - model_path (:obj:`Optional[str]`): The pretrained model path, which should
+            point to the ckpt file of the pretrained model, and an absolute path is recommended.
+            In LightZero, the path is usually something like ``exp_name/ckpt/ckpt_best.pth.tar``.
+        - max_train_iter (:obj:`Optional[int]`): Maximum policy update iterations in training.
+        - max_env_step (:obj:`Optional[int]`): Maximum collected environment interaction steps.
+    Returns:
+        - policy (:obj:`Policy`): Converged policy.
+    """
+
+    cfg, create_cfg = input_cfg
+    assert create_cfg.policy.type in ['efficientzero', 'muzero', 'muzero_context','sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
+        "train_muzero entry now only support the following algo.: 'efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'"
+
+    if create_cfg.policy.type in ['muzero', 'muzero_context']:
+        from lzero.mcts import MuZeroGameBuffer as GameBuffer
+    elif create_cfg.policy.type == 'efficientzero':
+        from lzero.mcts import EfficientZeroGameBuffer as GameBuffer
+    elif create_cfg.policy.type == 'sampled_efficientzero':
+        from lzero.mcts import SampledEfficientZeroGameBuffer as GameBuffer
+    elif create_cfg.policy.type == 'gumbel_muzero':
+        from lzero.mcts import GumbelMuZeroGameBuffer as GameBuffer
+    elif create_cfg.policy.type == 'stochastic_muzero':
+        from lzero.mcts import StochasticMuZeroGameBuffer as GameBuffer
+
+    if cfg.policy.cuda and torch.cuda.is_available():
+        cfg.policy.device = 'cuda'
+    else:
+        cfg.policy.device = 'cpu'
+
+    cfg = compile_config(cfg, seed=seed, env=None, auto=True, create_cfg=create_cfg, save_cfg=True)
+    # Create main components: env, policy
+    env_fn, collector_env_cfg, evaluator_env_cfg = get_vec_env_setting(cfg.env)
+
+    collector_env = create_env_manager(cfg.env.manager, [partial(env_fn, cfg=c) for c in collector_env_cfg])
+    evaluator_env = create_env_manager(cfg.env.manager, [partial(env_fn, cfg=c) for c in evaluator_env_cfg])
+
+    collector_env.seed(cfg.seed)
+    evaluator_env.seed(cfg.seed, dynamic_seed=False)
+    set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda)
+
+    if cfg.policy.eval_offline:
+        cfg.policy.learn.learner.hook.save_ckpt_after_iter = cfg.policy.eval_freq
+
+    policy = create_policy(cfg.policy, model=model, enable_field=['learn', 'collect', 'eval'])
+
+    # load pretrained model
+    if model_path is not None:
+        policy.learn_mode.load_state_dict(torch.load(model_path, map_location=cfg.policy.device))
+
+    # Create worker components: learner, collector, evaluator, replay buffer, commander.
+    tb_logger = SummaryWriter(os.path.join('./{}/log/'.format(cfg.exp_name), 'serial')) if get_rank() == 0 else None
+    learner = BaseLearner(cfg.policy.learn.learner, policy.learn_mode, tb_logger, exp_name=cfg.exp_name)
+
+    # ==============================================================
+    # MCTS+RL algorithms related core code
+    # ==============================================================
+    policy_config = cfg.policy
+    batch_size = policy_config.batch_size
+    # specific game buffer for MCTS+RL algorithms
+    replay_buffer = GameBuffer(policy_config)
+    collector = Collector(
+        env=collector_env,
+        policy=policy.collect_mode,
+        tb_logger=tb_logger,
+        exp_name=cfg.exp_name,
+        policy_config=policy_config
+    )
+    evaluator = Evaluator(
+        eval_freq=cfg.policy.eval_freq,
+        n_evaluator_episode=cfg.env.n_evaluator_episode,
+        stop_value=cfg.env.stop_value,
+        env=evaluator_env,
+        policy=policy.eval_mode,
+        tb_logger=tb_logger,
+        exp_name=cfg.exp_name,
+        policy_config=policy_config
+    )
+
+    # ==============================================================
+    # Main loop
+    # ==============================================================
+    # Learner's before_run hook.
+    learner.call_hook('before_run')
+
+    if cfg.policy.update_per_collect is not None:
+        update_per_collect = cfg.policy.update_per_collect
+
+    # The purpose of collecting random data before training:
+    # Exploration: Collecting random data helps the agent explore the environment and avoid getting stuck in a suboptimal policy prematurely.
+    # Comparison: By observing the agent's performance during random action-taking, we can establish a baseline to evaluate the effectiveness of reinforcement learning algorithms.
+    if cfg.policy.random_collect_episode_num > 0:
+        random_collect(cfg.policy, policy, LightZeroRandomPolicy, collector, collector_env, replay_buffer)
+    if cfg.policy.eval_offline:
+        eval_train_iter_list = []
+        eval_train_envstep_list = []
+
+    import copy
+    num_unroll_steps = copy.deepcopy(replay_buffer._cfg.num_unroll_steps)
+    collect_cnt = -1
+    # Usage
+    policy.last_batch_obs = initialize_zeros_batch(
+        cfg.policy.model.observation_shape,
+        len(evaluator_env_cfg),
+        cfg.policy.device
+    )
+    policy.last_batch_action = [-1 for _ in range(len(evaluator_env_cfg))]
+    # TODO: comment if debugging
+    # stop, reward = evaluator.eval(learner.save_checkpoint, learner.train_iter, collector.envstep)
+
+
+    while True:
+        log_buffer_memory_usage(learner.train_iter, replay_buffer, tb_logger)
+        collect_kwargs = {}
+        # set temperature for visit count distributions according to the train_iter,
+        # please refer to Appendix D in MuZero paper for details.
+        collect_kwargs['temperature'] = visit_count_temperature(
+            policy_config.manual_temperature_decay,
+            policy_config.fixed_temperature_value,
+            policy_config.threshold_training_steps_for_final_temperature,
+            trained_steps=learner.train_iter
+        )
+
+        if policy_config.eps.eps_greedy_exploration_in_collect:
+            epsilon_greedy_fn = get_epsilon_greedy_fn(
+                start=policy_config.eps.start,
+                end=policy_config.eps.end,
+                decay=policy_config.eps.decay,
+                type_=policy_config.eps.type
+            )
+            collect_kwargs['epsilon'] = epsilon_greedy_fn(collector.envstep)
+        else:
+            collect_kwargs['epsilon'] = 0.0
+
+        # Evaluate policy performance.
+        if evaluator.should_eval(learner.train_iter):
+            policy.last_batch_obs = initialize_zeros_batch(
+                cfg.policy.model.observation_shape,
+                len(evaluator_env_cfg),
+                cfg.policy.device
+            )
+            policy.last_batch_action = [-1 for _ in range(len(evaluator_env_cfg))]
+            if cfg.policy.eval_offline:
+                eval_train_iter_list.append(learner.train_iter)
+                eval_train_envstep_list.append(collector.envstep)
+            else:
+                stop, reward = evaluator.eval(learner.save_checkpoint, learner.train_iter, collector.envstep)
+                if stop:
+                    break
+
+        policy.last_batch_obs = initialize_zeros_batch(
+            cfg.policy.model.observation_shape,
+            len(collector_env_cfg),
+            cfg.policy.device
+        )
+        policy.last_batch_action = [-1 for _ in range(len(collector_env_cfg))]
+        # Collect data by default config n_sample/n_episode.
+        new_data = collector.collect(train_iter=learner.train_iter, policy_kwargs=collect_kwargs)
+        if cfg.policy.update_per_collect is None:
+            # update_per_collect is None, then update_per_collect is set to the number of collected transitions multiplied by the model_update_ratio.
+            collected_transitions_num = sum([len(game_segment) for game_segment in new_data[0]])
+            update_per_collect = int(collected_transitions_num * cfg.policy.model_update_ratio)
+        # save returned new_data collected by the collector
+        replay_buffer.push_game_segments(new_data)
+        # remove the oldest data if the replay buffer is full.
+        replay_buffer.remove_oldest_data_to_fit()
+
+        if replay_buffer.get_num_of_transitions() > 2000:
+            # Learn policy from collected data.
+            for i in range(update_per_collect):
+                # Learner will train ``update_per_collect`` times in one iteration.
+                if replay_buffer.get_num_of_transitions() > batch_size:
+                    train_data = replay_buffer.sample(batch_size, policy)
+                else:
+                    logging.warning(
+                        f'The data in replay_buffer is not sufficient to sample a mini-batch: '
+                        f'batch_size: {batch_size}, '
+                        f'{replay_buffer} '
+                        f'continue to collect now ....'
+                    )
+                    break
+
+                # The core train steps for MCTS+RL algorithms.
+                log_vars = learner.train(train_data, collector.envstep)
+
+                if cfg.policy.use_priority:
+                    replay_buffer.update_priority(train_data, log_vars[0]['value_priority_orig'])
+
+        if collector.envstep >= max_env_step or learner.train_iter >= max_train_iter:
+            if cfg.policy.eval_offline:
+                logging.info(f'eval offline beginning...')
+                ckpt_dirname = './{}/ckpt'.format(learner.exp_name)
+                # Evaluate the performance of the pretrained model.
+                for train_iter, collector_envstep in zip(eval_train_iter_list, eval_train_envstep_list):
+                    ckpt_name = 'iteration_{}.pth.tar'.format(train_iter)
+                    ckpt_path = os.path.join(ckpt_dirname, ckpt_name)
+                    # load the ckpt of pretrained model
+                    policy.learn_mode.load_state_dict(torch.load(ckpt_path, map_location=cfg.policy.device))
+                    stop, reward = evaluator.eval(learner.save_checkpoint, train_iter, collector_envstep)
+                    logging.info(
+                        f'eval offline at train_iter: {train_iter}, collector_envstep: {collector_envstep}, reward: {reward}')
+                logging.info(f'eval offline finished!')
+            break
+
+    # Learner's after_run hook.
+    learner.call_hook('after_run')
+    return policy

lzero/entry/train_muzero_gpt_same-data.py

@@ -21,7 +21,7 @@
 from .utils import random_collect
 
 
-def train_muzero_gpt(
+def train_unizero(
         input_cfg: Tuple[dict, dict],
         seed: int = 0,
         model: Optional[torch.nn.Module] = None,
@@ -47,10 +47,10 @@ def train_muzero_gpt(
     """
 
     cfg, create_cfg = input_cfg
-    assert create_cfg.policy.type in ['efficientzero', 'muzero_gpt', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
-        "train_muzero_gpt entry now only support the following algo.: 'efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero'"
+    assert create_cfg.policy.type in ['efficientzero', 'unizero', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
+        "train_unizero entry now only support the following algo.: 'efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero'"
 
-    if create_cfg.policy.type == 'muzero_gpt':
+    if create_cfg.policy.type == 'unizero':
         from lzero.mcts import MuZeroGameBuffer as GameBuffer
     elif create_cfg.policy.type == 'efficientzero':
         from lzero.mcts import EfficientZeroGameBuffer as GameBuffer

lzero/entry/train_muzero_gpt.py → lzero/entry/train_unizero.py

@@ -33,7 +33,7 @@ def initialize_zeros_batch(observation_shape, batch_size, device):
     return torch.zeros(shape).to(device)
 
 
-def train_muzero_gpt(
+def train_unizero(
         input_cfg: Tuple[dict, dict],
         seed: int = 0,
         model: Optional[torch.nn.Module] = None,
@@ -59,10 +59,10 @@ def train_muzero_gpt(
     """
 
     cfg, create_cfg = input_cfg
-    assert create_cfg.policy.type in ['efficientzero', 'muzero_gpt', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
-        "train_muzero_gpt entry now only support the following algo.: 'efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero'"
+    assert create_cfg.policy.type in ['efficientzero', 'unizero', 'sampled_efficientzero', 'gumbel_muzero', 'stochastic_muzero'], \
+        "train_unizero entry now only support the following algo.: 'efficientzero', 'muzero', 'sampled_efficientzero', 'gumbel_muzero'"
 
-    if create_cfg.policy.type == 'muzero_gpt':
+    if create_cfg.policy.type == 'unizero':
         from lzero.mcts import MuZeroGameBufferGPT as GameBuffer
     elif create_cfg.policy.type == 'efficientzero':
         from lzero.mcts import EfficientZeroGameBuffer as GameBuffer
@@ -249,11 +249,16 @@ def train_muzero_gpt(
         policy._target_model.world_model.precompute_pos_emb_diff_kv() # 非常重要，kv更新后需要重新计算
 
         policy._target_model.world_model.past_keys_values_cache_init_infer.clear()
+        for kv_cache_dict_env in policy._target_model.world_model.past_keys_values_cache_init_infer_envs:
+            kv_cache_dict_env.clear() 
+
         policy._target_model.world_model.past_keys_values_cache_recurrent_infer.clear()
         policy._target_model.world_model.keys_values_wm_list.clear() # TODO: 只适用于recurrent_inference() batch_pad
         print('sample target_model past_keys_values_cache.clear()')
 
         policy._collect_model.world_model.past_keys_values_cache_init_infer.clear() # very important
+        for kv_cache_dict_env in policy._collect_model.world_model.past_keys_values_cache_init_infer_envs:
+            kv_cache_dict_env.clear() 
         policy._collect_model.world_model.past_keys_values_cache_recurrent_infer.clear() # very important
         policy._collect_model.world_model.keys_values_wm_list.clear()  # TODO: 只适用于recurrent_inference() batch_pad
         torch.cuda.empty_cache() # TODO: NOTE

lzero/entry/train_muzero_gpt_multi_task.py → lzero/entry/train_unizero_multi_task.py

@@ -19,10 +19,10 @@
 from lzero.worker import MuZeroCollector as Collector
 from lzero.worker import MuZeroEvaluator as Evaluator
 from .utils import random_collect
-from lzero.mcts import MuZeroGameBufferGPT as GameBuffer
+from lzero.mcts import UniZeroGameBuffer as GameBuffer
 
 
-def train_muzero_gpt_multi_task(
+def train_unizero_multi_task(
         input_cfg_list: List[Tuple[dict, dict]],
         seed: int = 0,
         model: Optional[torch.nn.Module] = None,

lzero/mcts/buffer/__init__.py

@@ -1,5 +1,5 @@
 from .game_buffer_muzero import MuZeroGameBuffer
-from .game_buffer_muzero_gpt import MuZeroGameBufferGPT
+from .game_buffer_unizero import UniZeroGameBuffer
 from .game_buffer_efficientzero import EfficientZeroGameBuffer
 from .game_buffer_sampled_efficientzero import SampledEfficientZeroGameBuffer
 from .game_buffer_gumbel_muzero import GumbelMuZeroGameBuffer

lzero/mcts/buffer/game_buffer_muzero_gpt.py → lzero/mcts/buffer/game_buffer_unizero.py

@@ -14,8 +14,8 @@
     from lzero.policy import MuZeroPolicy, EfficientZeroPolicy, SampledEfficientZeroPolicy
 
 
-@BUFFER_REGISTRY.register('game_buffer_muzero_gpt')
-class MuZeroGameBufferGPT(GameBuffer):
+@BUFFER_REGISTRY.register('game_buffer_unizero')
+class UniZeroGameBuffer(GameBuffer):
     """
     Overview:
         The specific game buffer for MuZero policy.