search

nrontsis / PILCO

Bayesian Reinforcement Learning in Tensorflow
MIT License
313 stars 84 forks source link

Make trigonometric augmentation user friendly #10

Open nrontsis opened 6 years ago

nrontsis commented 6 years ago

Matlab's implementation augments the variables that represents angles to two new states that are simply their cos and sin.

This is important for the performance of the algorithm and is possible to be done in the current implementation by simply passing an augmented dataset to the PILCO object.

Investigate how to do this in a more user friendly way.

RicardoDominguez commented 6 years ago

Hey Nikitas,

I can address this issue if you haven't started working on it.

I was thinking about passing to the PILCO object a set of indices similar to the ones used in the MATLAB PILCO version (angi, dyni, dyno...)

nrontsis commented 6 years ago

Sure, PRs are welcome!

While coding, keep in mind that the purpose of this package is to provide researchers/experts of the area of gaussian process dynamical systems with a basic & clean implementation for PILCO. So we usually opt for code readability/conciseness instead of user friendliness [that is, when we can't have both :) ].