M.Sc. thesis by Ludwig Bald
Supervised by Nicole Ludwig, 2022-2023 at Uni Tübingen, Germany
This repository contains all code, experiments, results, figures, and necessary sources for my master's thesis. It contains modified data and code from the 2022 CityLearn challenge and from the research paper Estimating Risk and Uncertainty in Deep Reinforcement Learning, which proposes the UA-DQN algorithm.
In order to run an experiment, you need to install the packages from requirements.txt.
By default, the script local_evaluation.py runs the Rule-Based agent on the 5 public buildings from the 2022 CityLearn challenge.
The script run_experiment.py can be used to run DQN and UA-DQN. Hyperparameters and the Dataset can be specified as command line arguments or in the script.
The schema file data/citylearn_challenge_2022/test_schema.json specifies the setup used throughout the thesis.
Reward definitions can be changed in rewards/get_reward.py.
Explorative Data analysis is in the exploration folder, and experiments, along with their results and analysis in code are in the experiments folder.
LaTeX sources for my thesis are in the thesis folder.