Notes and limits

Things worth knowing before you depend on it.

• CPU only. No GPU support.
• Dense layers only. No convolutional or recurrent layers.
• float32 everywhere. Python floats are converted on the way in and out.
• No shuffling between epochs.
• No architecture check when loading weights.
• Output size tracking follows the last add_dense call.

Reinforcement learning

The C library also ships a REINFORCE agent (rl_agent_create, rl_draw_action, rl_remember, rl_policy_fit). It is not wrapped in Python. See the Snake example in the repository.

License

MIT.