pendulum

Pendulum

Both state and action space are continuous.

Data Collection

Skip if you want to use the given expert data.

• Install stable-baselines3

    pip install stable-baselines3[extra]

• Generate data: data_gen_pendulum.ipynb

Train Regression policy

• predict action directly from state
• MSE loss Using Scikit-learn : bc_pendulum_sklearn.ipynb
  Using Pytorch : bc_pendulum_torch.ipynb

run pytorch training on pendulum in colab

Train gaussian policy

• predict mean and variance
• Negative log likelihood loss

bc_pendulum_torch_gaussian.ipynb