Project structure of: openai/Video-Pre-Training

• Video-Pre-Training Minecraft AI training, competition, resources, and models.
  • agent.py Configure Minecraft agent for reinforcement learning
  • behavioural_cloning.py Trains actor-critic model for behavioral cloning.
  • data_loader.py Data loader for batch processing with worker threads
  • inverse_dynamics_model.py Minecraft action predictor with IDM model.
  • lib Transformer models, RL agents, and utility functions.
    • action_head.py ActionHead: Reinforcement learning action heads abstract base class
    • action_mapping.py Action mapping for video game inputs and management.
    • actions.py Minecraft actions, quantization, and transformer.
    • impala_cnn.py ImpalaCNN: Customizable, Multi-Stack Classification Model
    • masked_attention.py Masked Attention for time series data.
    • minecraft_util.py Compute entropy from categorical head outputs.
    • misc.py Data processing tasks and function design discussed.
    • mlp.py MLP class: Neural network with specified layers, normed linear.
    • normalize_ewma.py Normalizes data, maintains running mean & variance.
    • policy.py MinecraftAgentPolicy with reinforcement learning, image preprocessing
    • scaled_mse_head.py Scaled MSE head for linear output layer and MSE loss.
    • torch_util.py Torch utilities: libraries, device defaults, tensor functions.
    • tree_util.py PyTree API utility functions and data type registration.
    • util.py Neural network utilities for data processing
    • xf.py Attention mechanism, layers for transformer models.
  • README.md Minecraft AI training, competition, resource data, 10-min house building.
  • requirements.txt Installs essential libraries for Python.
  • run_agent.py Run MineRL model with pre-trained weights.
  • run_inverse_dynamics_model.py Inverse dynamics model with game inputs and video streaming.