README.md

lightRaven -- Offline RL with Maximum Speed

This library provides convenient tools for people to create their own seldonian algorithms with optimum performance. A detailed example is also included in dynamic_training.ipynb. Performance test is in ci_performance.ipynb.

Dependencies

• gym==0.17.3
• numpy==1.19.1
• scipy==1.5.2
• numba == 0.51.2

Supplementary Materials

• Definition of Seldonian Framework
  • Preventing undesirable behavior of intelligent machines
  • High Confidence Policy Improvement
• Definition of different Importance Sampling estimators
  • High Confidence Off-Policy Evaluation
• Definition of the new concentration bound
  • A New Confidence Interval for the Mean of a Bounded Random Variable