{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "IIgKlE2DbZaP"
   },
   "source": [
    "#  Préparer  des données (TensorFlow)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Nc00b8BWbZaQ"
   },
   "source": [
    "Installez les bibliothèques 🤗 *Transformers* et 🤗 *Datasets* pour exécuter ce *notebook*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "me9NX9X4bZaQ"
   },
   "outputs": [],
   "source": [
    "!pip install datasets transformers[sentencepiece]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "FFc_IGipbZaS"
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "from transformers import AutoTokenizer, TFAutoModelForSequenceClassification\n",
    "\n",
    "# Same as before\n",
    "checkpoint = \"camembert-base\"\n",
    "tokenizer = AutoTokenizer.from_pretrained(checkpoint)\n",
    "model = TFAutoModelForSequenceClassification.from_pretrained(checkpoint)\n",
    "sequences = [\n",
    "    \"J'ai attendu un cours d'HuggingFace toute ma vie.\", \n",
    "    \"Je déteste tellement ça !\"]\n",
    "batch = dict(tokenizer(sequences, padding=True, truncation=True, return_tensors=\"tf\"))\n",
    "\n",
    "# This is new\n",
    "model.compile(optimizer=\"adam\", loss=\"sparse_categorical_crossentropy\")\n",
    "labels = tf.convert_to_tensor([1, 1])\n",
    "model.train_on_batch(batch, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "mnRaEEZibZaT"
   },
   "outputs": [],
   "source": [
    "from datasets import load_dataset\n",
    "\n",
    "raw_datasets = load_dataset(\"paws-x\", \"fr\")\n",
    "raw_datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "HO2D7pJYbZaU"
   },
   "outputs": [],
   "source": [
    "raw_train_dataset = raw_datasets[\"train\"]\n",
    "raw_train_dataset[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "iJxImFEQbZaU"
   },
   "outputs": [],
   "source": [
    "raw_train_dataset.features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "mcfjYv9hbZaV"
   },
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer\n",
    "\n",
    "checkpoint = \"camembert-base\"\n",
    "tokenizer = AutoTokenizer.from_pretrained(checkpoint)\n",
    "tokenized_sentences_1 = tokenizer(raw_datasets[\"train\"][\"sentence1\"])\n",
    "tokenized_sentences_2 = tokenizer(raw_datasets[\"train\"][\"sentence2\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "Jd5FXWQBbZaW"
   },
   "outputs": [],
   "source": [
    "inputs = tokenizer(\"C'est la première phrase.\", \"C'est la deuxième.\")\n",
    "inputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "ly68WIIHbZaX"
   },
   "outputs": [],
   "source": [
    "tokenizer.convert_ids_to_tokens(inputs[\"input_ids\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "Avts72_WbZaY"
   },
   "outputs": [],
   "source": [
    "tokenized_dataset = tokenizer(\n",
    "    raw_datasets[\"train\"][\"sentence1\"],\n",
    "    raw_datasets[\"train\"][\"sentence2\"],\n",
    "    padding=True,\n",
    "    truncation=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "TMvS1GB-bZaZ"
   },
   "outputs": [],
   "source": [
    "def tokenize_function(example):\n",
    "    return tokenizer(example[\"sentence1\"], example[\"sentence2\"], truncation=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "AH7OVmjAbZaa"
   },
   "outputs": [],
   "source": [
    "tokenized_datasets = raw_datasets.map(tokenize_function, batched=True)\n",
    "tokenized_datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "B-mS1PgvbZaa"
   },
   "outputs": [],
   "source": [
    "from transformers import DataCollatorWithPadding\n",
    "\n",
    "data_collator = DataCollatorWithPadding(tokenizer=tokenizer, return_tensors=\"tf\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "huGLcqWdbZab"
   },
   "outputs": [],
   "source": [
    "samples = tokenized_datasets[\"train\"][:8]\n",
    "samples = {k: v for k, v in samples.items() if k not in [\"idx\", \"sentence1\", \"sentence2\"]}\n",
    "[len(x) for x in samples[\"input_ids\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "d6g3J9rbbZab"
   },
   "outputs": [],
   "source": [
    "batch = data_collator(samples)\n",
    "{k: v.shape for k, v in batch.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "th3DW4MhbZab"
   },
   "outputs": [],
   "source": [
    "tf_train_dataset = tokenized_datasets[\"train\"].to_tf_dataset(\n",
    "    columns=[\"attention_mask\", \"input_ids\", \"token_type_ids\"],\n",
    "    label_cols=[\"labels\"],\n",
    "    shuffle=True,\n",
    "    collate_fn=data_collator,\n",
    "    batch_size=8,\n",
    ")\n",
    "\n",
    "tf_validation_dataset = tokenized_datasets[\"validation\"].to_tf_dataset(\n",
    "    columns=[\"attention_mask\", \"input_ids\", \"token_type_ids\"],\n",
    "    label_cols=[\"labels\"],\n",
    "    shuffle=False,\n",
    "    collate_fn=data_collator,\n",
    "    batch_size=8,\n",
    ")"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "provenance": []
  },
  "gpuClass": "standard",
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}