Workflow

The main PyEPO workflow has four steps: define an optimization model, build an optimization-aware dataset, choose and train with a PyEPO method, and evaluate decision quality.

Core Steps

• New to PyEPO: read these pages in order.

  1. Optimization Models - define the optimization model

  2. Data and Datasets - generate data and build the dataset

  3. Training Methods - choose a training method and train the predictor

  4. Evaluation - decision-quality metrics

• Want to pick a method: the Choosing a Method section of Training Methods groups the methods by whether they return a loss or a solution, with a summary table of return types and inputs.

• Training in JAX/Flax: JAX Frontend follows the PyTorch loss API for jax.grad-based training. MPAX runs natively; non-JAX backends run through jax.pure_callback.

• Notebooks: runnable Colab examples are listed in Notebooks.