This repo contains the code for my specialization project at NTNU and part of the code for my master thesis. This is because they are two ways of solving the same problem.

The experiments can be found in the /experiments folder. The experiments/approx_phi directory contains the experiments for the specialization project. In these examples a reparametrization is approximated directly. The experiments/dqn contains the experiments for the master thesis. In these examples the value function is approximated using deep Q-networks.

Also note that the code specifically for the specialization project is located in neural_reparm/reparametrization.py.

The code written specifically for the master thesis is located in two files. The code that implements the altered DQN optimization algorithm is located in neural_reparm/reinforcement_learning. The code that implements the reparametrization environment is located in neural_reparm/reparam_env.

To run the first experiments, install the packages in requirements.txt. For instance with

pip3 install -r requirements.txt 

To run experiments using motion capture data. Set up the motion capture database as explained in alexarntzen/signatureshape. This is a copy of paalel/Signatures-in-Shape-Analysis that works with python3.