Code for Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions

Last update: Nov 25, 2022

Related tags

Deep Learning EMC-COLS-recourse

Overview

EMS-COLS-recourse

Initial Code for Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions

Folder structure:

data folder contains raw and final preprocessed data, along with the pre-processing script.
Src folder contain the code for our method.
trained_model contains the trained black box model checkpoint.

Making the environment

conda create -n rec_gen python=3.8.1
conda activate rec_gen
pip install -r requirements.txt

Steps for running experiments.

change current working directory to src

cd ./src/

Run data_io.py to dump mcmc cost samples.

python ./utils/data_io.py --save_data --data_name adult_binary --dump_negative_data --num_mcmc 1000

python ./utils/data_io.py --save_data --data_name compas_binary --dump_negative_data --num_mcmc 1000

run main experiments on COLS and P-COLS.

python run.py --data_name adult_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_main --budget 5000
python run.py --data_name compas_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_main --budget 5000

python run.py --data_name adult_binary --num_mcmc 1000 --model pls --num_cfs 10 --project_name exp_main --budget 5000
python run.py --data_name compas_binary --num_mcmc 1000 --model pls --num_cfs 10 --project_name exp_main --budget 5000

Run ablation Experiments

python run.py --data_name adult_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_ablation --budget 3000 --eval cost
python run.py --data_name adult_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_ablation --budget 3000 --eval cost_simple
python run.py --data_name adult_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_ablation --budget 3000 --eval proximity
python run.py --data_name adult_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_ablation --budget 3000 --eval sparsity
python run.py --data_name adult_binary --num_mcmc 1000 --model ls --num_cfs 10 --project_name exp_ablation --budget 3000 --eval diversity

Run experiments with budget

python run.py --data_name adult_binary --model ls --num_cfs 10 --num_users 100 --project_name exp_budget --budget 500
python run.py --data_name adult_binary --model ls --num_cfs 10 --num_users 100 --project_name exp_budget --budget 1000
python run.py --data_name adult_binary --model ls --num_cfs 10 --num_users 100 --project_name exp_budget --budget 2000
python run.py --data_name adult_binary --model ls --num_cfs 10 --num_users 100 --project_name exp_budget --budget 3000
python run.py --data_name adult_binary --model ls --num_cfs 10 --num_users 100 --project_name exp_budget --budget 5000
python run.py --data_name adult_binary --model ls --num_cfs 10 --num_users 100 --project_name exp_budget --budget 10000

Run experiments with number of counterfactuals

python run.py --data_name adult_binary --model model_name --num_cfs 1 --num_users 100 --project_name exp_cfs --budget 5000
python run.py --data_name adult_binary --model model_name --num_cfs 2 --num_users 100 --project_name exp_cfs --budget 5000
python run.py --data_name adult_binary --model model_name --num_cfs 3 --num_users 100 --project_name exp_cfs --budget 5000
python run.py --data_name adult_binary --model model_name --num_cfs 5 --num_users 100 --project_name exp_cfs --budget 5000
python run.py --data_name adult_binary --model model_name --num_cfs 10 --num_users 100 --project_name exp_cfs --budget 5000
python run.py --data_name adult_binary --model model_name --num_cfs 20 --num_users 100 --project_name exp_cfs --budget 5000
python run.py --data_name adult_binary --model model_name --num_cfs 30 --num_users 100 --project_name exp_cfs --budget 5000

Experiment with respect to Monte Carlo samples

Run these commands for different num_mcmc values. Default set to 5 in commands.

python ./utils/data_io.py --save_data --data_name adult_binary --dump_negative_data --num_mcmc 5

python run.py --data_name adult_binary --num_mcmc 5 --model model_name --num_cfs 10 --project_name exp_mcmc --budget 5000 --num_users 100

To train a new blackbox model

Run this right after preprocessing the data.

python train_model.py --data_name adult --max_epochs 1000 --check_val_every_n_epoch=1 --learning_rate=0.0001

Code for Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions

Related tags

Overview

EMS-COLS-recourse

Initial Code for Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions

Folder structure:

Making the environment

Steps for running experiments.

To train a new blackbox model

Owner

Prateek Yadav

Code accompanying the paper "ProxyFL: Decentralized Federated Learning through Proxy Model Sharing"

Conservative Q Learning for Offline Reinforcement Reinforcement Learning in JAX

TOOD: Task-aligned One-stage Object Detection, ICCV2021 Oral

A Kitti Road Segmentation model implemented in tensorflow.

Some methods for comparing network representations in deep learning and neuroscience.

Waymo motion prediction challenge 2021: 3rd place solution

A platform for intelligent agent learning based on a 3D open-world FPS game developed by Inspir.AI.

Explicable Reward Design for Reinforcement Learning Agents [NeurIPS'21]

Fast and Easy Infinite Neural Networks in Python

Official PyTorch implementation of "Improving Face Recognition with Large AgeGaps by Learning to Distinguish Children" (BMVC 2021)

Deep-Learning-Image-Captioning - Implementing convolutional and recurrent neural networks in Keras to generate sentence descriptions of images

Code for Fold2Seq paper from ICML 2021

Conditional Generative Adversarial Networks (CGAN) for Mobility Data Fusion

TensorFlow Ranking is a library for Learning-to-Rank (LTR) techniques on the TensorFlow platform

Official implementation of "Can You Spot the Chameleon? Adversarially Camouflaging Images from Co-Salient Object Detection" in CVPR 2022.

Winning solution of the Indoor Location & Navigation Kaggle competition

PyTorch implementation of deep GRAph Contrastive rEpresentation learning (GRACE).

OpenMMLab Pose Estimation Toolbox and Benchmark.

DeepLab resnet v2 model in pytorch

Object detection on multiple datasets with an automatically learned unified label space.