{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sparse Index for RAG Wikipedia Corpus\n",
    "\n",
    "This creates a sparse Terrier index using PyTerrier for the Wikipedia corpus used by Natural Questions and TextbookQuestionAnswering.\n",
    "\n",
    "The corpus is downloaded from https://huggingface.co/datasets/RUC-NLPIR/FlashRAG_datasets/resolve/main/retrieval-corpus/wiki18_100w.zip by `\n",
    "pt.get_dataset('rag:nq_wiki').get_corpus_iter()`.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pyterrier as pt\n",
    "import pyterrier_rag"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook requires PyTerrier 0.13 or higher."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'0.13.0'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pt.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lets prepare the index. We're going to store the title and text of the documents in the Terrier index, so we can use them for reranking. A study of title and text length distributions found that very few were cutoff with for max lengths of 1750 and 125, respectively.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "13:45:49.361 [ForkJoinPool-2-worker-3] WARN org.terrier.structures.BaseCompressingMetaIndex -- Structure meta reading lookup file directly from disk (SLOW) - try index.meta.index-source=fileinmem in the index properties file. 137.3 MiB of memory would be required.\n",
      "13:45:49.366 [ForkJoinPool-2-worker-3] WARN org.terrier.structures.BaseCompressingMetaIndex -- Structure meta reading data file directly from disk (SLOW) - try index.meta.data-source=fileinmem in the index properties file. 7 GiB of memory would be required.\n",
      "13:56:25.302 [ForkJoinPool-2-worker-3] WARN org.terrier.structures.BaseCompressingMetaIndex -- Structure meta reading data file directly from disk (SLOW) - try index.meta.data-source=fileinmem in the index properties file. 1.2 GiB of memory would be required.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<org.terrier.querying.IndexRef at 0x7fa3d024d5b0 jclass=org/terrier/querying/IndexRef jself=<LocalRef obj=0xc526808 at 0x7fa274037470>>"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "index_dir = \"./nq_index_new\"\n",
    "ref = pt.IterDictIndexer(\n",
    "        index_dir, \n",
    "        text_attrs=['title', 'text'], \n",
    "        meta={'docno' : 20, 'text' : 1750, 'title' : 125}\n",
    "    ).index(pt.get_dataset('rag:nq_wiki').get_corpus_iter())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We then upload the index to Huggingface..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "adding data.direct.bf [1.9 GB]\n",
      "adding data.document.fsarrayfile [340.7 MB]\n",
      "adding data.inverted.bf [1.5 GB]\n",
      "adding data.lexicon.fsomapfile [330.0 MB]\n",
      "adding data.lexicon.fsomaphash [1017 B]\n",
      "adding data.lexicon.fsomapid [15.3 MB]\n",
      "adding data.meta-0.fsomapfile [1.3 GB]\n",
      "adding data.meta.idx [160.3 MB]\n",
      "adding data.meta.zdata [8.2 GB]\n",
      "adding data.properties [4.1 KB]\n",
      "adding pt_meta.json [79 B]\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d807844944c94c4cb5b76e1472d062f8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "artifact.tar.lz4.json:   0%|          | 0.00/913 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8477f74a10114db0ab4c62be17d21385",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "artifact.tar.lz4:   0%|          | 0.00/12.9G [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b7082bc99c9a439dbb6ed8ab9fc484a1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Upload 2 LFS files:   0%|          | 0/2 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "Artifact uploaded to https://huggingface.co/datasets/pyterrier/ragwiki-terrier/tree/main/\n",
      "Consider editing the README.md to help explain this artifact to others.\n"
     ]
    }
   ],
   "source": [
    "index = pt.terrier.TerrierIndex(ref)\n",
    "index.to_hf('pyterrier/ragwiki-terrier')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:rag]",
   "language": "python",
   "name": "conda-env-rag-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}