{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_dataset, DatasetDict, concatenate_datasets\n",
    "\n",
    "columns = set(['premise', 'hypothesis', 'subset', 'label_name'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SNLI\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "snli = load_dataset('snli')\n",
    "snli = DatasetDict({'train': snli['train'], 'test': snli['validation']})\n",
    "snli"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = {0: 'entailment', 1: 'non-entailment', 2: 'non-entailment'}\n",
    "\n",
    "def label_name(sample):\n",
    "    sample['label_name'] = mapping[sample['label']]\n",
    "    sample['subset'] = 'snli'\n",
    "    return sample\n",
    "\n",
    "snli = snli.filter(lambda x: x['label'] != -1, batched=False)\n",
    "snli = snli.map(label_name, batched=False, remove_columns=list(set(snli['train'].column_names) - columns))\n",
    "snli"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MNLI"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mnli = load_dataset('multi_nli')\n",
    "mnli = DatasetDict({'train': mnli['train'], 'test': mnli['validation_matched']})\n",
    "mnli"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = {0: 'entailment', 1: 'non-entailment', 2: 'non-entailment'}\n",
    "\n",
    "def label_name(sample):\n",
    "    sample['label_name'] = mapping[sample['label']]\n",
    "    sample['subset'] = 'mnli'\n",
    "    return sample\n",
    "\n",
    "mnli = mnli.map(label_name, batched=False, remove_columns=list(set(mnli['train'].column_names) - columns))\n",
    "mnli"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# FEVER"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fever = load_dataset('pietrolesci/nli_fever')\n",
    "fever = DatasetDict({'train': fever['train'], 'test': fever['dev']})\n",
    "fever"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = {'SUPPORTS': 'entailment', 'REFUTES': 'non-entailment', 'NOT ENOUGH INFO': 'non-entailment'}\n",
    "\n",
    "def label_name(sample):\n",
    "    sample['label_name'] = mapping[sample['fever_gold_label']]\n",
    "    sample['subset'] = 'fever'\n",
    "    return sample\n",
    "\n",
    "fever = fever.map(label_name, batched=False, remove_columns=list(set(fever['train'].column_names) - columns))\n",
    "fever"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SciTail"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "scitail = load_dataset('scitail', 'snli_format')\n",
    "scitail = DatasetDict({'train': scitail['train'], 'test': scitail['validation']})\n",
    "scitail"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = {'entailment': 'entailment', 'neutral': 'non-entailment'}\n",
    "\n",
    "def label_name(sample):\n",
    "    sample['label_name'] = mapping[sample['gold_label']]\n",
    "    sample['premise'] = sample['sentence1']\n",
    "    sample['hypothesis'] = sample['sentence2']\n",
    "    sample['subset'] = 'scitail'\n",
    "    return sample\n",
    "\n",
    "scitail = scitail.map(label_name, batched=False, remove_columns=list(set(scitail['train'].column_names) - columns))\n",
    "scitail"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# PAWS"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "paws = load_dataset('paws', 'labeled_final')\n",
    "paws = DatasetDict({'train': paws['train'], 'test': paws['validation']})\n",
    "paws"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = {1: 'entailment', 0: 'non-entailment'}\n",
    "\n",
    "def label_name(sample):\n",
    "    sample['label_name'] = mapping[sample['label']]\n",
    "    sample['premise'] = sample['sentence1']\n",
    "    sample['hypothesis'] = sample['sentence2']\n",
    "    sample['subset'] = 'paws'\n",
    "    return sample\n",
    "\n",
    "paws = paws.map(label_name, batched=False, remove_columns=list(set(paws['train'].column_names) - columns))\n",
    "paws"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# VitaminC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "vitaminc = load_dataset('tals/vitaminc')\n",
    "vitaminc = DatasetDict({'train': vitaminc['train'], 'test': vitaminc['validation']})\n",
    "vitaminc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "mapping = {'SUPPORTS': 'entailment', 'REFUTES': 'non-entailment', 'NOT ENOUGH INFO': 'non-entailment'}\n",
    "\n",
    "def label_name(sample):\n",
    "    sample['label_name'] = mapping[sample['label']]\n",
    "    sample['premise'] = sample['evidence']\n",
    "    sample['hypothesis'] = sample['claim']\n",
    "    sample['subset'] = 'vitaminc'\n",
    "    return sample\n",
    "\n",
    "vitaminc = vitaminc.map(label_name, batched=False, remove_columns=list(set(vitaminc['train'].column_names) - columns))\n",
    "vitaminc"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# NLI Mixture"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "nli_train = concatenate_datasets([snli['train'], mnli['train'], fever['train'], scitail['train'], paws['train'], vitaminc['train']])\n",
    "nli_test = concatenate_datasets([snli['test'], mnli['test'], fever['test'], scitail['test'], paws['test'], vitaminc['test']])\n",
    "\n",
    "nli_dataset = DatasetDict({'train': nli_train, 'test': nli_test})\n",
    "nli_dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "nli_dataset.push_to_hub('AntoineBlanot/nli_mixture')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "nlp-train",
   "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": "3.10.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}