- Strategic classification: the problem of learning in settings where users can strategically modify their features to improve outcomes.
- The contribution of the original author: a flexible, practical and socially-aware framework for learning in strategic environments.
- Our addition: A combined dataset where some users are gaming and some are not.
Strategic classification regards the problem of learning in settings where users can strategically modify their features to improve outcomes. This setting applies broadly and has received much recent attention. But despite its practical significance, work in this space has so far been predominantly theoretical. The original authors presented a learning framework for strategic classification that is practical. Our approach directly minimizes the "strategic" empirical risk, achieved by differentiating through the strategic response of users. This provides flexibility that allows us to extend beyond the original problem formulation and towards more realistic learning scenarios.
The reproducibilty was mainly focussed on the vanilla notebook with editing. The other notebooks were already in the original codebase and were used to test their results for these experiments. Our extension is the semi_strategic_vanilla.ipynb notebook which can be run from start to finish.
Setup the environment using
conda create --name "name" python=3.8
conda activate "name"
pip install -r requirements.txt
pip install jupyter
jupyter notebook
Download the card fraud dataset from https://www.kaggle.com/mlg-ulb/creditcardfraud, rename it to "creditcard.csv" and put it in the data folder
Use the Reproduction plots.ipynb to create the plots with the saved models. Exception is the card fraud dataset, since that one was too big for github to allow. To allow card fraud dataset first
run strategicClassification-Vanilla.ipynb then
run semi_strategic_vanilla.ipynb finally
run Reproduction plots.ipynb for card fraud
Run a notebook from start to finish to train the model and get the evaluation saved.
Sagi Levanon
@SagiLevanon
Nir Rosenfeld
@NirRosenfeld
- If the environment file doesn't work you need to manually install the modules