decyphir / rlrom

A tool set of methods combining Reinforcement Learning with somewhat Robust and possibly Online Monitoring.

RLRom

This module integrates Robust Online Monitoring methods with Reinforcement Learning stuff. The first motivation is testing/monitoring RL models.

Install

Those are needed for building some of the required python modules:

• CMake
• swig

Then run the following:

pip install -r requirements.txt

Running

Run python run_app.py, then open browser.

Features

• Select an environment among a list of supported ones.
• To load a model, choose between
  • Random: random actions
  • Local: Upload model zip files created with stable-baselines-3, then choose one
  • Hugging Face: Fetch the list of models available on Hugging Face, then choose one
• Choose between running with or without human render
• Runs from a list of seeds and store traces
• Compute total rewards
• Plots observation, reward, actions, individually or together of any trace, with flexible layout
• Evaluate (monitor) and plot quantitative and Boolean satisfaction of any Signal Temporal Logic formula (STL)
• Sort runs against STL formula robustness