Error in _compute_target_reward_value

[ArshamSol]

Hi, I am trying to use stochastic muZero to play a game. Here is my config file:

from easydict import EasyDict
#from zoo.matchthree.envs.matchthree_wrapper import branch_size

#env_id = 'match3_lightzero'
action_space_size = 127
chance_space_size = 1

# ==============================================================
# begin of the most frequently changed config specified by the user
# ==============================================================
collector_env_num = 1
n_episode = 2
evaluator_env_num = 1
num_simulations = 25
update_per_collect = 100
batch_size = 256
max_env_step = 500
reanalyze_ratio = 0.
use_ture_chance_label_in_chance_encoder = True
#num_unroll_steps = 2

# ==============================================================
# end of the most frequently changed config specified by the user
# ==============================================================

matchthree_stochastic_muzero_config = dict(
    exp_name=f'data_stochastic_mz/match3_stochastic_muzero_ns{num_simulations}_upc{update_per_collect}_rer{reanalyze_ratio}_chance{chance_space_size}_seed0',
    env=dict(
        stop_value=int(1e6),
        #env_id=env_id,
        obs_shape=(9, 64, 64) ,
        collector_env_num=collector_env_num,
        evaluator_env_num=evaluator_env_num,
        n_evaluator_episode=evaluator_env_num,
        manager=dict(shared_memory=False, ),   
        channel_last=False,
    ),
    policy=dict(
        model=dict(
            observation_shape=(9, 64, 64),
            frame_stack_num=1,
            image_channel=9,
            action_space_size=action_space_size,
            chance_space_size=chance_space_size,
            downsample=True,
            self_supervised_learning_loss=True,  # default is False
            discrete_action_encoding_type='one_hot',
            norm_type='BN',
            support_scale = 70,
            value_support_size = 141,
            reward_support_size = 141,
            #device='cpu',
            #categorical_distribution = False,
        ),
        model_path=None,
        cuda=True,
        gumbel_algo=False,
        mcts_ctree=True,
        env_type='not_board_games',
        game_segment_length=400,
        use_augmentation=True,
        update_per_collect=update_per_collect,
        batch_size=batch_size,
        optim_type='Adam',
        piecewise_decay_lr_scheduler=False,
        learning_rate=3e-3,
        discount_factor=0.997,
        num_simulations=num_simulations,
        reanalyze_ratio=reanalyze_ratio,
        ssl_loss_weight=2, 
        n_episode=n_episode,
        eval_freq=int(2e3),
        replay_buffer_size=int(1e6),
        collector_env_num=collector_env_num,
        evaluator_env_num=evaluator_env_num,
    ),
)
matchthree_stochastic_muzero_config = EasyDict(matchthree_stochastic_muzero_config)
main_config = matchthree_stochastic_muzero_config

matchthree_stochastic_muzero_create_config = dict(
    env=dict(
        type='match3_lightzero',
        import_names=['zoo.matchthree.envs.matchthree_lightzero_env'],
    ),
    env_manager=dict(type='subprocess'),
    policy=dict(
        type='stochastic_muzero',
        import_names=['lzero.policy.stochastic_muzero'],
    ),
)  
matchthree_stochastic_muzero_create_config = EasyDict(matchthree_stochastic_muzero_create_config)
create_config = matchthree_stochastic_muzero_create_config


if __name__ == "__main__":
    from lzero.entry import train_muzero
    train_muzero([main_config, create_config], seed=0, model_path=main_config.policy.model_path, max_env_step=max_env_step)

And then I receive following error from game_buffer_muzero script:
value_list = value_list.reshape(-1) * (
ValueError: operands could not be broadcast together with shapes (6144,) (1536,)

Here is the Traceback:

Traceback (most recent call last):
  File "/home/arsham/miniconda3/envs/mlagents/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/home/arsham/miniconda3/envs/mlagents/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/home/arsham/.vscode/extensions/ms-python.debugpy-2025.0.1-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/__main__.py", line 71, in <module>
    cli.main()
  File "/home/arsham/.vscode/extensions/ms-python.debugpy-2025.0.1-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 501, in main
    run()
  File "/home/arsham/.vscode/extensions/ms-python.debugpy-2025.0.1-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 351, in run_file
    runpy.run_path(target, run_name="__main__")
  File "/home/arsham/.vscode/extensions/ms-python.debugpy-2025.0.1-linux-x64/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 310, in run_path
    return _run_module_code(code, init_globals, run_name, pkg_name=pkg_name, script_name=fname)
  File "/home/arsham/.vscode/extensions/ms-python.debugpy-2025.0.1-linux-x64/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 127, in _run_module_code
    _run_code(code, mod_globals, init_globals, mod_name, mod_spec, pkg_name, script_name)
  File "/home/arsham/.vscode/extensions/ms-python.debugpy-2025.0.1-linux-x64/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 118, in _run_code
    exec(code, run_globals)
  File "/home/arsham/LightZero/zoo/matchthree/config/matchthree_stochastic_muzero_config.py", line 104, in <module>
    train_muzero([main_config, create_config], seed=0, model_path=main_config.policy.model_path, max_env_step=max_env_step)
  File "/home/arsham/LightZero/lzero/entry/train_muzero.py", line 202, in train_muzero
    train_data = replay_buffer.sample(batch_size, policy)
  File "/home/arsham/LightZero/lzero/mcts/buffer/game_buffer_muzero.py", line 141, in sample
    batch_rewards, batch_target_values = self._compute_target_reward_value(
  File "/home/arsham/LightZero/lzero/mcts/buffer/game_buffer_muzero.py", line 485, in _compute_target_reward_value
    value_list = value_list.reshape(-1) * (
ValueError: operands could not be broadcast together with shapes (6144,) (1536,) 

The problem is value_list have shape 4x transition_batch_size.
I would be grateful for any guidance on how to resolve the issue.

[puyuan1996]

Hello,

Could you confirm whether any modifications have been made to components other than config and the environment?

Additionally, could you provide detailed debugging information within _compute_target_reward_value()? Specifically, details such as m_output.latent_state.shape, transition_batch_size, and value_list.shape would be helpful.

By default, transition_batch_size is calculated as game_segment_batch_size * (num_unroll_steps + 1), where num_unroll_steps defaults to 5, and game_segment_batch_size corresponds to batch_size in config.

Thank you for your attention! If you have any further questions, please feel free to ask.