This repository contains the implementation of the Acquisition Conditioned Oracle (AACO), which was used in the paper "Acquisition Conditioned Oracle for Nongreedy Active Feature Acquisition" published at ICML 2024.

This repository contains the code used to implement and evaluate AACO. The main files are as follows:

• aaco_rollout.py: Main script that runs the AACO on a given dataset and configuration.
• src/classifier.py: Contains different classifier classes.
• src/mask_generator.py: Defines the two mask generation strategies (an exhaustive search or random masks).
• config.yaml: Configuration file used to specify the dataset, hyperparameters, and other settings.

The repository uses the following key libraries:

• numpy
• torch
• xgboost
• yaml

To run the AACO, make sure that the configuration file (config.yaml) is properly set up. Then, execute the following command:

python src/aaco_rollout.py

Results will be saved in the results/ directory as a .pt file.