Bachelor Thesis

An evaluation of Interpretation-Nets applied to Logistic Regression for explaining Neural Networks

Appendix A: Program Code

Setup

This code was tested on python==3.9.12 using the dependencies of the requirements.txt file provided.

Reproduce results of thesis

To reproduce the results of the thesis, run the following jupyter notebook scripts:

[instance]_1_generateData_n[numFeatures][addNoise]
[instance]_2_lambda_n[numFeatures][addNoise]
[instance]_3_inet_n[numFeatures][addNoise]
LR_4_eval_n[numFeatures][addNoise] (only for instance = 'LR')

Specify the experiment parameters using:

• instance = {'DT', 'LR'}, use either 'DT' (for inets for Decision Tree) or 'LR' (for inets for Logistic Regression, Plain Logistic Regression and Plain Decision Trees)
• numFeatures = {'5', '10', '20'}
• addNoise = {'-noise', ''}

For example:

LR_1_generateData_n10
LR_2_lambda_n10
LR_3_inet_n10
LR_4_eval_n10

DT_1_generateData_n5-noise
DT_2_lambda_n5-noise
DT_3_inet_n5-noise

Afterwards, view the results in "05-BA/data_LR" (if instance=LR) or "05-BA/data" (if instance=DT).