Bert/03-subword-tokenizer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 如何处理海量单词"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "s1 = 'albums sold 124443286539 copies'\n",
    "s2 = 'technically perfect, melodically correct'\n",
    "s3 = 'featuring a previously unheard track'\n",
    "s4 = 'bestselling music artist'\n",
    "s5 = 's1 d1 o1 and o2'\n",
    "s6 = 'asbofwheohwbeif'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 实例化 tokenizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/zoupeng/miniconda3/envs/lightning/lib/python3.9/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n"
     ]
    }
   ],
   "source": [
    "from transformers import BertTokenizer\n",
    "\n",
    "model_name = 'bert-base-uncased'\n",
    "tokenizer = BertTokenizer.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## vocab\n",
    "\n",
    "- tokenizer.vocab, tokenizer.id_to_tokens\n",
    "- len(tokenizer.vocab) == 30522"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "30522"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(tokenizer.vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "100"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.vocab['[UNK]']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['best', '##sell', '##ing']"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.tokenize('bestselling')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5828\n"
     ]
    }
   ],
   "source": [
    "cnt_sharp = 0\n",
    "for token, id in tokenizer.vocab.items():\n",
    "    if token.startswith('##'):\n",
    "        # print(token)\n",
    "        cnt_sharp += 1\n",
    "print(cnt_sharp)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## token 测试"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'input_ids': [101, 2004, 5092, 2546, 2860, 5369, 11631, 2860, 19205, 2546, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "inputs = tokenizer(s6)\n",
    "inputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['[CLS]', 'as', '##bo', '##f', '##w', '##he', '##oh', '##w', '##bei', '##f', '[SEP]']\n"
     ]
    }
   ],
   "source": [
    "print(tokenizer.convert_ids_to_tokens(inputs['input_ids']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['as', '##bo', '##f', '##w', '##he', '##oh', '##w', '##bei', '##f']"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokens = tokenizer.tokenize(s6)\n",
    "tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[2004, 5092, 2546, 2860, 5369, 11631, 2860, 19205, 2546]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ids = tokenizer.convert_tokens_to_ids(tokens)\n",
    "ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'asbofwheohwbeif'"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "word = tokenizer.decode(ids)\n",
    "word"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 总结\n",
    "\n",
    "- tokenizer 轻易不会将一个词处理为 [UNK] (100)\n",
    "- 基于词汇表，tokenize, encode, decode 一体\n",
    "  - tokenize：word => token(s)，将word尽可能地映射为 vocab 中的 keys\n",
    "  - encode: token => id\n",
    "  - decode: id => token => word\n",
    "    - encode 完了之后也不是终点（word），decode 还要能很好地将 id 还原，尽可能与输入的 word 对齐"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "lightning",
   "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": "3.9.17"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}