--- language: en license: apache-2.0 library_name: pytorch tags: - deep-reinforcement-learning - reinforcement-learning - DI-engine - QbertNoFrameskip-v4 benchmark_name: OpenAI/Gym/Atari task_name: QbertNoFrameskip-v4 pipeline_tag: reinforcement-learning model-index: - name: C51 results: - task: type: reinforcement-learning name: reinforcement-learning dataset: name: OpenAI/Gym/Atari-QbertNoFrameskip-v4 type: OpenAI/Gym/Atari-QbertNoFrameskip-v4 metrics: - type: mean_reward value: 16750.0 +/- 0.0 name: mean_reward --- # Play **QbertNoFrameskip-v4** with **C51** Policy ## Model Description This is a simple **C51** implementation to OpenAI/Gym/Atari **QbertNoFrameskip-v4** using the [DI-engine library](https://github.com/opendilab/di-engine) and the [DI-zoo](https://github.com/opendilab/DI-engine/tree/main/dizoo). **DI-engine** is a python library for solving general decision intelligence problems, which is based on implementations of reinforcement learning framework using PyTorch or JAX. This library aims to standardize the reinforcement learning framework across different algorithms, benchmarks, environments, and to support both academic researches and prototype applications. Besides, self-customized training pipelines and applications are supported by reusing different abstraction levels of DI-engine reinforcement learning framework. ## Model Usage ### Install the Dependencies
(Click for Details) ```shell # install huggingface_ding git clone https://github.com/opendilab/huggingface_ding.git pip3 install -e ./huggingface_ding/ # install environment dependencies if needed pip3 install DI-engine[common_env] ```
### Git Clone from Huggingface and Run the Model
(Click for Details) ```shell # running with trained model python3 -u run.py ``` **run.py** ```python from ding.bonus import C51Agent from ding.config import Config from easydict import EasyDict import torch # Pull model from files which are git cloned from huggingface policy_state_dict = torch.load("pytorch_model.bin", map_location=torch.device("cpu")) cfg = EasyDict(Config.file_to_dict("policy_config.py").cfg_dict) # Instantiate the agent agent = C51Agent( env_id="QbertNoFrameskip-v4", exp_name="QbertNoFrameskip-v4-C51", cfg=cfg.exp_config, policy_state_dict=policy_state_dict ) # Continue training agent.train(step=5000) # Render the new agent performance agent.deploy(enable_save_replay=True) ```
### Run Model by Using Huggingface_ding
(Click for Details) ```shell # running with trained model python3 -u run.py ``` **run.py** ```python from ding.bonus import C51Agent from huggingface_ding import pull_model_from_hub # Pull model from Hugggingface hub policy_state_dict, cfg = pull_model_from_hub(repo_id="OpenDILabCommunity/QbertNoFrameskip-v4-C51") # Instantiate the agent agent = C51Agent( env_id="QbertNoFrameskip-v4", exp_name="QbertNoFrameskip-v4-C51", cfg=cfg.exp_config, policy_state_dict=policy_state_dict ) # Continue training agent.train(step=5000) # Render the new agent performance agent.deploy(enable_save_replay=True) ```
## Model Training ### Train the Model and Push to Huggingface_hub
(Click for Details) ```shell #Training Your Own Agent python3 -u train.py ``` **train.py** ```python from ding.bonus import C51Agent from huggingface_ding import push_model_to_hub # Instantiate the agent agent = C51Agent(env_id="QbertNoFrameskip-v4", exp_name="QbertNoFrameskip-v4-C51") # Train the agent return_ = agent.train(step=int(20000000)) # Push model to huggingface hub push_model_to_hub( agent=agent.best, env_name="OpenAI/Gym/Atari", task_name="QbertNoFrameskip-v4", algo_name="C51", wandb_url=return_.wandb_url, github_repo_url="https://github.com/opendilab/DI-engine", github_doc_model_url="https://di-engine-docs.readthedocs.io/en/latest/12_policies/c51.html", github_doc_env_url="https://di-engine-docs.readthedocs.io/en/latest/13_envs/atari.html", installation_guide="pip3 install DI-engine[common_env]", usage_file_by_git_clone="./c51/qbert_c51_deploy.py", usage_file_by_huggingface_ding="./c51/qbert_c51_download.py", train_file="./c51/qbert_c51.py", repo_id="OpenDILabCommunity/QbertNoFrameskip-v4-C51", create_repo=False ) ```
**Configuration**
(Click for Details) ```python exp_config = { 'env': { 'manager': { 'episode_num': float("inf"), 'max_retry': 1, 'retry_type': 'reset', 'auto_reset': True, 'step_timeout': None, 'reset_timeout': None, 'retry_waiting_time': 0.1, 'cfg_type': 'BaseEnvManagerDict' }, 'stop_value': 30000, 'n_evaluator_episode': 8, 'collector_env_num': 8, 'evaluator_env_num': 8, 'env_id': 'QbertNoFrameskip-v4', 'frame_stack': 4, 'env_wrapper': 'atari_default' }, 'policy': { 'model': { 'encoder_hidden_size_list': [128, 128, 512], 'v_min': -10, 'v_max': 10, 'n_atom': 51, 'obs_shape': [4, 84, 84], 'action_shape': 6 }, 'learn': { 'learner': { 'train_iterations': 1000000000, 'dataloader': { 'num_workers': 0 }, 'log_policy': True, 'hook': { 'load_ckpt_before_run': '', 'log_show_after_iter': 100, 'save_ckpt_after_iter': 10000, 'save_ckpt_after_run': True }, 'cfg_type': 'BaseLearnerDict' }, 'update_per_collect': 10, 'batch_size': 32, 'learning_rate': 0.0001, 'target_update_freq': 500, 'target_theta': 0.005, 'ignore_done': False }, 'collect': { 'collector': {}, 'n_sample': 100, 'unroll_len': 1 }, 'eval': { 'evaluator': { 'eval_freq': 4000, 'render': { 'render_freq': -1, 'mode': 'train_iter' }, 'figure_path': None, 'cfg_type': 'InteractionSerialEvaluatorDict', 'stop_value': 30000, 'n_episode': 8 } }, 'other': { 'replay_buffer': { 'replay_buffer_size': 400000 }, 'eps': { 'type': 'exp', 'start': 1.0, 'end': 0.05, 'decay': 1000000 } }, 'on_policy': False, 'cuda': True, 'multi_gpu': False, 'bp_update_sync': True, 'traj_len_inf': False, 'type': 'c51', 'priority': True, 'priority_IS_weight': False, 'discount_factor': 0.99, 'nstep': 3, 'cfg_type': 'C51PolicyDict' }, 'exp_name': 'QbertNoFrameskip-v4-C51', 'seed': 0, 'wandb_logger': { 'gradient_logger': True, 'video_logger': True, 'plot_logger': True, 'action_logger': True, 'return_logger': False } } ```
**Training Procedure** - **Weights & Biases (wandb):** [monitor link](https://wandb.ai/zjowowen/QbertNoFrameskip-v4-C51) ## Model Information - **Github Repository:** [repo link](https://github.com/opendilab/DI-engine) - **Doc**: [DI-engine-docs Algorithm link](https://di-engine-docs.readthedocs.io/en/latest/12_policies/c51.html) - **Configuration:** [config link](https://huggingface.co/OpenDILabCommunity/QbertNoFrameskip-v4-C51/blob/main/policy_config.py) - **Demo:** [video](https://huggingface.co/OpenDILabCommunity/QbertNoFrameskip-v4-C51/blob/main/replay.mp4) - **Parameters total size:** 55276.2 KB - **Last Update Date:** 2023-09-21 ## Environments - **Benchmark:** OpenAI/Gym/Atari - **Task:** QbertNoFrameskip-v4 - **Gym version:** 0.25.1 - **DI-engine version:** v0.4.9 - **PyTorch version:** 2.0.1+cu117 - **Doc**: [DI-engine-docs Environments link](https://di-engine-docs.readthedocs.io/en/latest/13_envs/atari.html)