Coming soon

---

Major features planned for Spring 2023 release, in order of priority:

• Hoeffding's concentration inequality bounding method. This will enable running true Seldonian algorithms (as opposed to quasi-Seldonian algorithms) with the Seldonian Engine.
• Multiclass classification - This was implemented in the Engine in version 0.0.8, but is not yet integrated into the Experiments library.
• Multiple label columns in a dataset (supervised learning).
• Importance sampling variants, such as weighted and per-decision importance sampling. In version alpha, the standard importance sampling estimator is the only primary objective function available for Seldonian reinforcement algorithms.
• An automated method for determining optimal data split between candidate data and safety data.
• Automatic differentiation using JAX. JAX will be a big upgrade to autograd and should provide significant improvements for many problems. Autograd will remain an option for users who do not wish to install or use JAX.