search

MichelangeloConserva / oblax

OBLax is a collection of online Bayesian learning algorithms implemented in jax.
MIT License
3 stars 0 forks source link

Base classes for environments #3

Open MichelangeloConserva opened 1 year ago

MichelangeloConserva commented 1 year ago

How should we structure the base classes for the sequential environments?

I don't particularly like the approach in the sequential-neural-testbed repository. They assume that the dataset is fixed and simply iterates over it. I believe that a more flexible approach that allows for the data to be sampled (and maybe batched to improve sampling speed) would be more convenient. In other words, the SequentialEnvironment base class should take a data generator function in input and iteratively sample data from it.

MichelangeloConserva commented 1 year ago

The base.py file contains the current base class for sequential environment. It is vaguely inspired by the sequential-neural-testbed base class.