python expert for RL and Gaussian Processes -- 2

I need to replace the deep learning model in the FCQ code on (https://github.com/jcborges/FCQ#reinforcement-learning-with-fully-connected-q-function) with MGGP from (https://github.com/andrewcharlesjones/multi-group-GP).

MULTI-GROUP REINFORCEMENT LEARNING

Neural Fitted-Q (NFQ, Riedmiller [2005]) utilizes the neural networks as regressors to fitted Q-iteration by updating network weights per iteration, rather than rebuilding the trees entirely. We will now modify NFQ by introducing Gaussian process regression in place of Neural networks.

We next use MGGPs as the engine of off-policy RL methods. Viewing the process of learning patient-specific policies as supervised learning, we may use MGGP regression to share strength across large patient cohorts, and to study differences in the treatment objectives between patient subgroups. Specifically, we replace the Q-value function with the MGGP to allow nonlinear group-based inference of the Q-value function. We can then get a full posterior distribution for any categorical subset of patients.

Deliverables :

1. A working model of Gaussian process regression and RL should be delivered.
2. The model should be able to classify the right birth method (Ceaseran delivery and Vaginal birth) for a woman based on their ethnic group.
3. Slides that show the result from different rewards functions in the project.
4. The right reward function should be built.
5. There should be a written report of the analysis of the result gotten from the project.


Thanks.