Upload convert.ipynb

convert.ipynb

@@ -2,133 +2,479 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 108,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [],
    "source": [
+    "import spacy\n",
     "import pandas as pd\n",
     "import glob\n",
-    "import spacy\n",
-    "from spacy.tokens import Span, Doc\n",
-    "import os\n",
-    "from spacy.training import biluo_tags_to_offsets, biluo_tags_to_spans, iob_to_biluo\n",
-    "import srsly"
+    "import tqdm"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "17\n"
+     ]
+    }
+   ],
+   "source": [
+    "files = glob.glob(\"Database/*/CONLL/*.xlsx\")\n",
+    "print(len(files))"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 113,
+   "execution_count": 26,
    "metadata": {},
    "outputs": [],
    "source": [
-    "def create_spacy_training_data(file_path):\n",
-    "    # Load data from the Excel file\n",
-    "    data = pd.read_excel(file_path)\n",
-    "    # if \"Line_ID\" in data.columns:\n",
-    "    #     group_col = \"Line_ID\"\n",
-    "    if \"ACT\" in data.columns:\n",
+    "def group_acts(df):\n",
+    "    if \"ACT\" in df.columns:\n",
     "        group_col = \"ACT\"\n",
-    "    elif \"Original_Act_ID\" in data.columns:\n",
+    "    elif \"Original_Act_ID\" in df.columns:\n",
     "        group_col = \"Original_Act_ID\"\n",
     "    else:\n",
     "        \"unknown\"\n",
-    "    # data = data[~data['Word_x'].apply(lambda x: isinstance(x, int))]\n",
-    "    # data = data[~data['Word_x'].apply(lambda x: isinstance(x, float))]\n",
-    "    data['Word_x'] = data['Word_x'].astype(str).str.strip()\n",
+    "\n",
+    "    if \"LANG\" in df.columns:\n",
+    "        df = df[df[\"LANG\"] == \"LAT\"]\n",
+    "\n",
+    "    df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
     "    \n",
     "    # Combine words into sentences, assumed by unique 'Line_ID'\n",
-    "    grouped_data = data.groupby(group_col)\n",
+    "    grouped_data = df.groupby(group_col)\n",
+    "\n",
+    "    return grouped_data\n",
+    "\n",
+    "def create_spacy_doc(df, nlp):\n",
+    "    grouped_df = group_acts(df)\n",
+    "    docs = []\n",
     "    \n",
-    "    # Prepare training data in spaCy format\n",
-    "    training_data = []\n",
-    "    for _, item in grouped_data:\n",
-    "        bilo_loc = item[\"LOC_x\"].tolist()\n",
-    "        bilo_person = item[\"PERS_x\"].tolist()\n",
-    "        tokens = item[\"Word_x\"].tolist()\n",
-    "        doc = Doc(nlp.vocab, words=tokens, spaces=[True for i in range(len(tokens))])\n",
-    "        # doc = nlp(\" \".join(tokens))\n",
+    "    for key, group in grouped_df:\n",
+    "        tokens = []\n",
+    "        spaces = []\n",
+    "        entities = []\n",
+    "        current_entity = None\n",
     "\n",
-    "        spans = iob_to_biluo(bilo_person)\n",
-    "        spans = biluo_tags_to_spans(doc, spans)\n",
+    "        # Iterate over the rows in the group\n",
+    "        for i, row in enumerate(group.itertuples()):\n",
+    "            word = row.Word_x\n",
+    "            length_word = len(word)\n",
+    "            # Check if there is a next word and whether it should be followed by a space\n",
+    "            space_after = not (i < len(group) - 1 and group.iloc[i + 1].Word_x in [',', '.', ';', ':'])\n",
+    "            tokens.append(word)\n",
+    "            spaces.append(space_after)\n",
     "\n",
+    "            # Handle entity recognition\n",
+    "            if row.PERS_x != 'O':\n",
+    "                entity_type = 'PERSON'\n",
+    "            elif row.LOC_x != 'O':\n",
+    "                entity_type = 'LOC'\n",
+    "            else:\n",
+    "                entity_type = None\n",
     "\n",
-    "        loc_spans = iob_to_biluo(bilo_loc)\n",
-    "        loc_spans = biluo_tags_to_spans(doc, loc_spans)\n",
+    "            if current_entity is None and entity_type is not None:\n",
+    "                # Start new entity\n",
+    "                current_entity = [i, i, entity_type]\n",
+    "            elif current_entity is not None:\n",
+    "                if entity_type == current_entity[2]:\n",
+    "                    # Extend current entity\n",
+    "                    current_entity[1] = i\n",
+    "                else:\n",
+    "                    # Finish current entity and add to entities list\n",
+    "                    entities.append(current_entity)\n",
+    "                    current_entity = [i, i, entity_type] if entity_type else None\n",
     "\n",
-    "        spans = loc_spans + spans\n",
+    "        # Check if an entity is still open at the end of the group\n",
+    "        if current_entity is not None:\n",
+    "            entities.append(current_entity)\n",
+    "\n",
+    "        # Create a spacy Doc object\n",
+    "        doc = spacy.tokens.Doc(nlp.vocab, words=tokens, spaces=spaces)\n",
+    "        # Get the sentencizer component from the pipeline\n",
+    "        sentencizer = nlp.get_pipe(\"sentencizer\")\n",
+    "\n",
+    "        # Apply the sentencizer to the Doc\n",
+    "        sentencizer(doc)\n",
+    "        # Create Span objects for the entities\n",
+    "        spans = [doc.char_span(doc[ent[0]].idx, doc[ent[1]].idx + len(doc[ent[1]].text), label=ent[2])\n",
+    "                 for ent in entities if doc.char_span(doc[ent[0]].idx, doc[ent[1]].idx + len(doc[ent[1]].text), label=ent[2])]\n",
     "        doc.spans[\"sc\"] = spans\n",
-    "        span_ents = []\n",
-    "        for span in doc.spans[\"sc\"]:\n",
-    "            span_ents.append({\"text\": span.text, \"label\": span.label_, \"start\": span.start, \"end\": span.end})\n",
-    "        training_data.append({\"text\": doc.text, \"spans\": span_ents})\n",
-    "    return training_data"
+    "        \n",
+    "\n",
+    "        docs.append(doc)\n",
+    "\n",
+    "    return docs\n"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 98,
+   "execution_count": 27,
    "metadata": {},
    "outputs": [
     {
-     "name": "stdout",
+     "name": "stderr",
      "output_type": "stream",
      "text": [
-      "17\n"
+      "  0%|          | 0/17 [00:00<?, ?it/s]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      "  6%|▌         | 1/17 [00:01<00:23,  1.45s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 12%|█▏        | 2/17 [00:03<00:26,  1.77s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 24%|██▎       | 4/17 [00:18<01:08,  5.30s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 29%|██▉       | 5/17 [00:26<01:15,  6.29s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 35%|███▌      | 6/17 [00:29<00:57,  5.25s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 41%|████      | 7/17 [00:30<00:38,  3.85s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 47%|████▋     | 8/17 [00:32<00:30,  3.43s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 53%|█████▎    | 9/17 [00:35<00:26,  3.30s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      " 76%|███████▋  | 13/17 [00:50<00:14,  3.51s/it]/var/folders/4f/ddlj81h90_n0_h5wwvjbd2b40000gn/T/ipykernel_8164/298127518.py:12: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  df['Word_x'] = df['Word_x'].astype(str).str.strip()\n",
+      "100%|██████████| 17/17 [01:06<00:00,  3.89s/it]\n"
      ]
     }
    ],
    "source": [
-    "files = glob.glob(\"Database/*/CONLL/*.xlsx\")\n",
-    "print(len(files))"
+    "\n",
+    "docs = []\n",
+    "nlp = spacy.blank(\"en\")\n",
+    "nlp.add_pipe(\"sentencizer\")\n",
+    "\n",
+    "for file in tqdm.tqdm(files):\n",
+    "    df = pd.read_excel(file)\n",
+    "    docs = docs+create_spacy_doc(df, nlp)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 117,
+   "execution_count": 40,
    "metadata": {},
    "outputs": [
     {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "0 Notre_Dame_Roche_Paris_BnF_10996\n",
-      "1 Pontoise_Paris_BnF_5657\n",
-      "2 Saint_Denis_Paris_AN_LL_1157\n",
-      "3 Pontigny_Paris_BnF_lat_9887_inner_unmerged.xlsx\n",
-      "4 Navarre_Pau_AD_E513\n",
-      "5 Clairmarais_Troyes_AD_3H3700_inner_unmerged.xlsx\n",
-      "6 Port_Royal_2_Paris_BnF_10998\n",
-      "7 Nesle_Chantilly_GB_Reg12_14F22\n",
-      "8 Fervaques_Paris_BnF_lat_11071\n",
-      "9 Molesme_2_Dijon_ADCO_Cart_143_7H7\n",
-      "10 Saint_Nicaise_Reims_BM_1843_inner_unmerged.xlsx\n",
-      "11 Sommereux_Paris_Bnf_nal_1934\n",
-      "12 Chartres_2_Paris_BnF_lat_10095\n",
-      "13 Chartres_1_Paris_BnF_lat_10094\n",
-      "14 Vauluisant_Paris_BnF_lat_9901\n",
-      "15 Port_Royal_1_Paris_BnF_10997\n",
-      "16 Molesme_1_Dijon_ADCO_Cart_142_7H6\n"
-     ]
+     "data": {
+      "text/plain": [
+       "{'tokenized_text': ['In',\n",
+       "  'nomine',\n",
+       "  'Domini',\n",
+       "  ',',\n",
+       "  'amen',\n",
+       "  '.',\n",
+       "  'Ego',\n",
+       "  'Mauricius',\n",
+       "  ',',\n",
+       "  'Dei',\n",
+       "  'gracia',\n",
+       "  'Parisiensis',\n",
+       "  'episcopus',\n",
+       "  ',',\n",
+       "  'universitati',\n",
+       "  'presencium',\n",
+       "  'ac',\n",
+       "  'futurorum',\n",
+       "  'hujus',\n",
+       "  'pagine',\n",
+       "  'attestatione',\n",
+       "  'notificare',\n",
+       "  'curamus',\n",
+       "  'quod',\n",
+       "  'dominus',\n",
+       "  'Guido',\n",
+       "  'de',\n",
+       "  'Levies',\n",
+       "  ',',\n",
+       "  'pia',\n",
+       "  'et',\n",
+       "  'honesta',\n",
+       "  'consideratione',\n",
+       "  'ductus',\n",
+       "  ',',\n",
+       "  'ad',\n",
+       "  'edificandam',\n",
+       "  'quandam',\n",
+       "  'novellam',\n",
+       "  'plantationem',\n",
+       "  ',',\n",
+       "  'amore',\n",
+       "  'Dei',\n",
+       "  'et',\n",
+       "  'remedio',\n",
+       "  'anime',\n",
+       "  'sue',\n",
+       "  'et',\n",
+       "  'animarum',\n",
+       "  'parentum',\n",
+       "  'predecessorum',\n",
+       "  'suorum',\n",
+       "  ',',\n",
+       "  'fratribus',\n",
+       "  'ibi',\n",
+       "  'Deo',\n",
+       "  'servituris',\n",
+       "  'in',\n",
+       "  'perpetuam',\n",
+       "  'elemosinam',\n",
+       "  'donavit',\n",
+       "  'unam',\n",
+       "  'carrucam',\n",
+       "  'de',\n",
+       "  'terra',\n",
+       "  'quam',\n",
+       "  'emit',\n",
+       "  'des',\n",
+       "  'Fers',\n",
+       "  'Dasnois',\n",
+       "  ',',\n",
+       "  'et',\n",
+       "  'de',\n",
+       "  'decima',\n",
+       "  'duas',\n",
+       "  'partes',\n",
+       "  'quas',\n",
+       "  'ab',\n",
+       "  'hiisdem',\n",
+       "  'emit',\n",
+       "  ',',\n",
+       "  'et',\n",
+       "  'unam',\n",
+       "  'partem',\n",
+       "  'nemoris',\n",
+       "  'quantum',\n",
+       "  'semita',\n",
+       "  'dividit',\n",
+       "  'versus',\n",
+       "  'terram',\n",
+       "  'datam',\n",
+       "  ';',\n",
+       "  'hanc',\n",
+       "  'elemosinam',\n",
+       "  'in',\n",
+       "  'manu',\n",
+       "  'nostra',\n",
+       "  'resignatam',\n",
+       "  'benigne',\n",
+       "  'tribuit',\n",
+       "  '.',\n",
+       "  'Sciendum',\n",
+       "  'autem',\n",
+       "  'quod',\n",
+       "  'de',\n",
+       "  'hac',\n",
+       "  'elemosina',\n",
+       "  'investivimus',\n",
+       "  'Guidonem',\n",
+       "  ',',\n",
+       "  'quondam',\n",
+       "  'presbiterum',\n",
+       "  'de',\n",
+       "  'Meencort',\n",
+       "  ',',\n",
+       "  'pro',\n",
+       "  'se',\n",
+       "  'et',\n",
+       "  'pro',\n",
+       "  'aliis',\n",
+       "  'ibi',\n",
+       "  'Deo',\n",
+       "  'se',\n",
+       "  'reddituris',\n",
+       "  '.',\n",
+       "  'Actum',\n",
+       "  'apud',\n",
+       "  'Sanctum',\n",
+       "  'Victorem',\n",
+       "  ',',\n",
+       "  'astantibus',\n",
+       "  'Petro',\n",
+       "  ',',\n",
+       "  'precentore',\n",
+       "  'Parisiensi',\n",
+       "  ';',\n",
+       "  'Nicholao',\n",
+       "  ',',\n",
+       "  'presbitero',\n",
+       "  ';',\n",
+       "  'Philippo',\n",
+       "  ',',\n",
+       "  'canonico',\n",
+       "  ';',\n",
+       "  'Haimerico',\n",
+       "  ',',\n",
+       "  'capellano',\n",
+       "  'nostro',\n",
+       "  ';',\n",
+       "  'Enardo',\n",
+       "  ',',\n",
+       "  'presbitero',\n",
+       "  'de',\n",
+       "  'Balneolis',\n",
+       "  ';',\n",
+       "  'fratre',\n",
+       "  'Stephano',\n",
+       "  'de',\n",
+       "  'Monte-Fermeolo',\n",
+       "  ';',\n",
+       "  'incarnationis',\n",
+       "  'dominice',\n",
+       "  'anno',\n",
+       "  'millesimo',\n",
+       "  'centesimo',\n",
+       "  'XCº',\n",
+       "  ',',\n",
+       "  'episcopatus',\n",
+       "  'nostri',\n",
+       "  'tricesimo',\n",
+       "  'sexto',\n",
+       "  '.'],\n",
+       " 'spans': [{'text': 'Mauricius', 'label': 'PERSON', 'start': 7, 'end': 8},\n",
+       "  {'text': 'Parisiensis', 'label': 'LOC', 'start': 11, 'end': 12},\n",
+       "  {'text': 'Guido de Levies', 'label': 'PERSON', 'start': 25, 'end': 28},\n",
+       "  {'text': 'Fers Dasnois', 'label': 'LOC', 'start': 68, 'end': 70},\n",
+       "  {'text': 'Guidonem', 'label': 'PERSON', 'start': 108, 'end': 109},\n",
+       "  {'text': 'Meencort', 'label': 'LOC', 'start': 113, 'end': 114},\n",
+       "  {'text': 'Sanctum Victorem', 'label': 'LOC', 'start': 127, 'end': 129},\n",
+       "  {'text': 'Petro', 'label': 'PERSON', 'start': 131, 'end': 132},\n",
+       "  {'text': 'Parisiensi', 'label': 'LOC', 'start': 134, 'end': 135},\n",
+       "  {'text': 'Nicholao', 'label': 'PERSON', 'start': 136, 'end': 137},\n",
+       "  {'text': 'Philippo', 'label': 'PERSON', 'start': 140, 'end': 141},\n",
+       "  {'text': 'Haimerico', 'label': 'PERSON', 'start': 144, 'end': 145},\n",
+       "  {'text': 'Enardo', 'label': 'PERSON', 'start': 149, 'end': 150},\n",
+       "  {'text': 'Balneolis', 'label': 'LOC', 'start': 153, 'end': 154},\n",
+       "  {'text': 'Stephano de Monte-Fermeolo',\n",
+       "   'label': 'PERSON',\n",
+       "   'start': 156,\n",
+       "   'end': 159}]}"
+      ]
+     },
+     "execution_count": 40,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "hf_docs = []\n",
+    "hf_docs_sents = []\n",
+    "for doc in docs:\n",
+    "    tokenized_text = [token.text for token in doc]\n",
+    "    spans = [{\"text\": span.text, \"label\": span.label_, \"start\": span.start, \"end\": span.end} for span in doc.spans[\"sc\"]]\n",
+    "    hf_docs.append({\"tokenized_text\": tokenized_text, \"spans\": spans})\n",
+    "    for sent in doc.sents:\n",
+    "        sent_doc = sent.as_doc()  # Create a new Doc from the sentence\n",
+    "        \n",
+    "        # Create a list of spans that are within the boundaries of this sentence\n",
+    "        sent_spans = [\n",
+    "            {\"start\": span.start - sent.start, \"end\": span.end - sent.start, \"label\": span.label_, \"text\": span.text}\n",
+    "            for span in doc.spans[\"sc\"] if span.start >= sent.start and span.end <= sent.end\n",
+    "        ]\n",
+    "        \n",
+    "        # Check if there are entities in the sentence, if yes, add to hf_docs_sents\n",
+    "        if sent_spans:\n",
+    "            hf_docs_sents.append({\n",
+    "                \"tokenized_text\": [token.text for token in sent],\n",
+    "                \"spans\": sent_spans,\n",
+    "                \"ms\": {}  # Assuming 'ms' should be some metadata, define or update accordingly\n",
+    "            })\n",
+    "hf_docs[0]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 33,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "15141"
+      ]
+     },
+     "execution_count": 33,
+     "metadata": {},
+     "output_type": "execute_result"
     }
    ],
    "source": [
-    "training_data = []\n",
-    "for i, filename in enumerate(files):\n",
-    "    manuscript = os.path.basename(filename).split('_final_version_inner')[0]\n",
-    "    print(i, manuscript)\n",
-    "    res = create_spacy_training_data(filename)\n",
-    "    for r in res:\n",
-    "        r[\"ms\"] = manuscript\n",
-    "        training_data.append(r)"
+    "len(hf_docs_sents)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 34,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import srsly"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 35,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "srsly.write_jsonl(\"home-alcar-ner.jsonl\", hf_docs)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 120,
+   "execution_count": 39,
    "metadata": {},
    "outputs": [],
    "source": [
-    "srsly.write_jsonl(\"home-alcar-ner.jsonl\", training_data)"
+    "srsly.write_jsonl(\"home-alcar-ner-sents.jsonl\", hf_docs_sents)"
    ]
   },
   {