Update

BirkhoffG · Oct 20, 2023 · 5ec864f · 5ec864f
1 parent b120eb8
commit 5ec864f
Show file tree

Hide file tree

Showing 5 changed files with 122 additions and 82 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1,4 +1,5 @@
 _docs/
+_proc
 
 *.bak
 .gitattributes

diff --git a/README.md b/README.md
@@ -1 +1,59 @@
-This file will be overwritten by `index.ipynb`
+# RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model
+
+[![Arxiv](https://img.shields.io/badge/Arxiv-2206.00700-orange)](https://arxiv.org/pdf/2206.00700.pdf)
+
+This repo contains code to reproduce our paper published at [CIKM 2023](https://arxiv.org/pdf/2206.00700.pdf).
+
+To cite this paper:
+
+
+```bibtex
+@inproceedings{guo2021rocoursenet,
+    author={Guo, Hangzhi and Jia, Feiran and Chen, Jinghui and Squicciarini, Anna and Yadav, Amulya},
+    title = {RoCourseNet: Robust Training of a Prediction Aware Recourse Model},
+    year = {2023},
+    isbn = {9798400701030},
+    publisher = {Association for Computing Machinery},
+    address = {New York, NY, USA},
+    url = {https://doi.org/10.1145/3583780.3615040},
+    doi = {10.1145/3583780.3615040},
+    booktitle = {Proceedings of the 32nd ACM International Conference on Information and Knowledge Management},
+    numpages = {15},
+    location = {Birmingham, United Kingdom},
+    series = {CIKM ’23}
+}
+
+```
+
+## Install
+
+This project uses 
+[jax-relax](https://github.com/BirkhoffG/ReLax/tree/master) (a fast and scalable recourse explanation library).
+Ths library is highly scalable and extensible, which enables our experiments to be finished within 30 minutes.
+In contrast, a pytorch implementation of RoCourseNet takes around 12 hours to run.
+
+```sh
+pip install -e ".[dev]" --upgrade
+```
+
+## Run Experiments
+
+Running `scripts.experiment.py` with different arguments will reproduce results in our paper. For example,
+
+1. Train and Evaluate RoCourseNet on Loan Application Dataset:
+
+```sh
+python -m scripts.experiment.py -d loan
+```
+
+2. Train and Evaluate CounterNet on Loan Application Dataset:
+
+```sh
+python -m scripts.experiment.py -m CounterNet -d loan
+```
+
+3. Train and Evaluate ROAR on Loan Application Dataset:
+
+```sh
+python -m scripts.experiment.py -m ROAR -d loan
+```
diff --git a/nbs/index.ipynb b/nbs/index.ipynb
@@ -1,88 +1,69 @@
 {
  "cells": [
-  {
-   "cell_type": "raw",
-   "metadata": {},
-   "source": [
-    "---\n",
-    "description: Summary description here.\n",
-    "output-file: index.html\n",
-    "title: Project name here\n",
-    "\n",
-    "---\n",
-    "\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from .core import *"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "This file will become your README and also the index of your documentation."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Install"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "`pip install your_project_name`"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## How to use"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Fill me in please! Don't forget code examples:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "2"
-      ]
-     },
-     "execution_count": null,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "1+1"
+    "# RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model\n",
+    "\n",
+    "[![Arxiv](https://img.shields.io/badge/Arxiv-2206.00700-orange)](https://arxiv.org/pdf/2206.00700.pdf)\n",
+    "\n",
+    "This repo contains code to reproduce our paper published at [CIKM 2023](https://arxiv.org/pdf/2206.00700.pdf).\n",
+    "\n",
+    "To cite this paper:\n",
+    "\n",
+    "\n",
+    "```bibtex\n",
+    "@inproceedings{guo2021rocoursenet,\n",
+    "    author={Guo, Hangzhi and Jia, Feiran and Chen, Jinghui and Squicciarini, Anna and Yadav, Amulya},\n",
+    "    title = {RoCourseNet: Robust Training of a Prediction Aware Recourse Model},\n",
+    "    year = {2023},\n",
+    "    isbn = {9798400701030},\n",
+    "    publisher = {Association for Computing Machinery},\n",
+    "    address = {New York, NY, USA},\n",
+    "    url = {https://doi.org/10.1145/3583780.3615040},\n",
+    "    doi = {10.1145/3583780.3615040},\n",
+    "    booktitle = {Proceedings of the 32nd ACM International Conference on Information and Knowledge Management},\n",
+    "    numpages = {15},\n",
+    "    location = {Birmingham, United Kingdom},\n",
+    "    series = {CIKM ’23}\n",
+    "}\n",
+    "\n",
+    "```\n",
+    "\n",
+    "## Install\n",
+    "\n",
+    "This project uses \n",
+    "[jax-relax](https://github.com/BirkhoffG/ReLax/tree/master) (a fast and scalable recourse explanation library).\n",
+    "Ths library is highly scalable and extensible, which enables our experiments to be finished within 30 minutes.\n",
+    "In contrast, a pytorch implementation of RoCourseNet takes around 12 hours to run.\n",
+    "\n",
+    "```sh\n",
+    "pip install -e \".[dev]\" --upgrade\n",
+    "```\n",
+    "\n",
+    "## Run Experiments\n",
+    "\n",
+    "Running `scripts.experiment.py` with different arguments will reproduce results in our paper. For example,\n",
+    "\n",
+    "1. Train and Evaluate RoCourseNet on Loan Application Dataset:\n",
+    "\n",
+    "```sh\n",
+    "python -m scripts.experiment.py -d loan\n",
+    "```\n",
+    "\n",
+    "2. Train and Evaluate CounterNet on Loan Application Dataset:\n",
+    "\n",
+    "```sh\n",
+    "python -m scripts.experiment.py -m CounterNet -d loan\n",
+    "```\n",
+    "\n",
+    "3. Train and Evaluate ROAR on Loan Application Dataset:\n",
+    "\n",
+    "```sh\n",
+    "python -m scripts.experiment.py -m ROAR -d loan\n",
+    "```\n"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
   }
  ],
  "metadata": {

diff --git a/scripts/utils_configs.py b/scripts/utils_configs.py
@@ -118,7 +118,7 @@
         'n_epochs': 100,
         "batch_size": 256,
         # 'n_epochs': 10,
-        'monitor_metrics': 'val/val_loss'
+        # 'monitor_metrics': None#'val/val_loss'
     }, 
     'data_dir_list': {
         "assets/data/german_credit/org.csv", "assets/data/german_credit/upt.csv"
@@ -198,7 +198,7 @@
         'n_epochs': 50,
         "batch_size": 128,
         # 'n_epochs': 10,
-        'monitor_metrics': 'val/val_loss'
+        'monitor_metrics': None
     },
     "data_dir_list": [
         "assets/data/covarient/org.csv", "assets/data/covarient/upt.csv"

diff --git a/settings.ini b/settings.ini
@@ -9,7 +9,7 @@ version = 0.0.1
 
 ### OPTIONAL ###
 
-requirements = jax-relax<0.2
+requirements = jax-relax===0.1.2 lime folktables jax[cpu]<0.4.1
 dev_requirements = nbdev jupyter
 # console_scripts =