Support for Supervised Learning

While the focus of OPF-Gym is on reinforcement learning and its environments, it also enables comparability with other machine learning approaches like supervised or unsupervised learning.

For that, we provide a convenience function to create a labeled dataset for supervised learning from any given OPF-Gym environment.

from opfgym.util import create_labeled_dataset
from opfgym.envs import QMarket  # Or some other environment

env = QMarket()
# Set `store_to_path = path` to store the dataset to a directory instead
inputs, outputs, optimal_objs = create_labeled_dataset(env, num_samples=10)

# Train your supervised learning algorithm with inputs and outputs
# supervised_model = train(inputs, outputs)

# Assuming no data re-scaling, the outputs can directly be fed back to `env.step()`.
# obs, info = env.reset()
# action = supervised_model.predict(obs)
# env.step(action)

Disclaimer: The dataset creation works only for environments that are solvable with the pandapower conventional OPF to generate ground-truth labels. That is the case for all provided Benchmarks. However, it might not be the case for custom environments, especially when implementing advanced OPF concepts like multi-stage OPF or stochastic OPF. These are not solvable with the pandapower OPF. In that case, you also have to provide your own OPF solver to your custom environment, as described in Create Custom Environments.