This repository is part of the Bachelor's thesis project of Matej Havelka for CSE3000 in Q4 2022. Other bachelor projects can be found here. This project studies the effect of honesty on causal forests in different situations and tries to conclude whether using honesty in general cases is beneficial or not. To access the bachelors thesis you might require TU Delft login, the paper can be found in the TU Delft repository TODO: provide link.

To add run the experiment you can run the main script. It might take quite a while (at least 2 hours on my setup). Afterwards you should be able to find the results in newly generated directories, most importantly in the parameterization directory. Each experiment is a separate function, it automatically saves all figures in an appropriate folder. To save all intermediate results make sure to go to the session class and enable saving tables. This is automatically disabled by default as it makes the running time double. Default number of threads is equal to 6, if your setup supports more feel free to update this in the session class as well.

To add a new model you need to extend the CausalMethod class in the appropriate class. Then add a new function to the experiment builder that adds the causal method. Afterwards you can construct the experiment with whatever data generators there are.

To add a new generator you need to create a new function in experiment builder where you define the necessary functions to generate that data. With that you can add it to any experiment as you would with other generators.

For further examples refer to the main script.

Matej Havelka - [email protected]