datagen_ValuePrism.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/allenai/ValuePrism"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd, numpy as np, os, json, tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "full = pd.read_csv('ValuePrism/full/full.csv')\n",
    "full.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "example = full.sample(1).iloc[0]\n",
    "print(example['situation'])\n",
    "print(example['explanation'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## How to generate a question?\n",
    "\n",
    "We need to generate a question for each situation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "full['question'] = \"Situation: \"+full['situation'] +\"\\nWhat values, rights, and duties are relevant for this situation?\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### How to generate a perspective?\n",
    "\n",
    "We need to generate a perspective for each question / answer pair. To keep it simple, we'll just use the question and explanation text with some labels."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "full['perspective'] = \"Value: \"+full['text']+\"\\nExplanation: \"+full['explanation']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "value_perspectives = full.groupby('question')['perspective'].apply(lambda x: x.to_list())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "full['vrd'].unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "value_perspectives.to_frame().reset_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "value_perspectives.to_frame().reset_index().to_csv('data/ValuePrism_perspectives.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}