{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Classification HuggingFace"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-info\">\n",
    "\n",
    "This tutorial is available as an IPython notebook at [Malaya/example/zeroshot-classification-huggingface](https://github.com/huseinzol05/Malaya/tree/master/example/zeroshot-classification-huggingface).\n",
    "    \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-info\">\n",
    "\n",
    "This module trained on both standard and local (included social media) language structures, so it is save to use for both.\n",
    "    \n",
    "</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = ''\n",
    "os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import logging\n",
    "\n",
    "logging.basicConfig(level=logging.INFO)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 3.1 s, sys: 3.47 s, total: 6.57 s\n",
      "Wall time: 2.28 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "import malaya"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### what is zero-shot classification\n",
    "\n",
    "Commonly we supervised a machine learning on specific labels, negative / positive for sentiment, anger / happy / sadness for emotion and etc. The model cannot give an output if we want to know how much percentage of 'jealous' in emotion analysis model because supported labels are only {anger, happy, sadness}. Imagine, for example, trying to identify a text without ever having seen one 'jealous' label before, impossible. **So, zero-shot trying to solve this problem.**\n",
    "\n",
    "zero-shot learning refers to the process by which a machine learns how to recognize objects (image, text, any features) without any labeled training data to help in the classification.\n",
    "\n",
    "[Yin et al. (2019)](https://arxiv.org/abs/1909.00161) stated in his paper, any pretrained language model finetuned on text similarity actually can acted as an out-of-the-box zero-shot text classifier."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So, we are going to use transformer models from `malaya.similarity.semantic.huggingface` with a little tweaks."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### List available HuggingFace models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:malaya.similarity.semantic:tested on matched dev set translated MNLI, https://huggingface.co/datasets/mesolitica/translated-MNLI\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Size (MB)</th>\n",
       "      <th>macro precision</th>\n",
       "      <th>macro recall</th>\n",
       "      <th>macro f1-score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>mesolitica/finetune-mnli-t5-super-tiny-standard-bahasa-cased</th>\n",
       "      <td>50.7</td>\n",
       "      <td>0.74562</td>\n",
       "      <td>0.74574</td>\n",
       "      <td>0.74501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mesolitica/finetune-mnli-t5-tiny-standard-bahasa-cased</th>\n",
       "      <td>139.0</td>\n",
       "      <td>0.76584</td>\n",
       "      <td>0.76565</td>\n",
       "      <td>0.76542</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mesolitica/finetune-mnli-t5-small-standard-bahasa-cased</th>\n",
       "      <td>242.0</td>\n",
       "      <td>0.78067</td>\n",
       "      <td>0.78063</td>\n",
       "      <td>0.78010</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mesolitica/finetune-mnli-t5-base-standard-bahasa-cased</th>\n",
       "      <td>892.0</td>\n",
       "      <td>0.78903</td>\n",
       "      <td>0.79064</td>\n",
       "      <td>0.78918</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                    Size (MB)  \\\n",
       "mesolitica/finetune-mnli-t5-super-tiny-standard...       50.7   \n",
       "mesolitica/finetune-mnli-t5-tiny-standard-bahas...      139.0   \n",
       "mesolitica/finetune-mnli-t5-small-standard-baha...      242.0   \n",
       "mesolitica/finetune-mnli-t5-base-standard-bahas...      892.0   \n",
       "\n",
       "                                                    macro precision  \\\n",
       "mesolitica/finetune-mnli-t5-super-tiny-standard...          0.74562   \n",
       "mesolitica/finetune-mnli-t5-tiny-standard-bahas...          0.76584   \n",
       "mesolitica/finetune-mnli-t5-small-standard-baha...          0.78067   \n",
       "mesolitica/finetune-mnli-t5-base-standard-bahas...          0.78903   \n",
       "\n",
       "                                                    macro recall  \\\n",
       "mesolitica/finetune-mnli-t5-super-tiny-standard...       0.74574   \n",
       "mesolitica/finetune-mnli-t5-tiny-standard-bahas...       0.76565   \n",
       "mesolitica/finetune-mnli-t5-small-standard-baha...       0.78063   \n",
       "mesolitica/finetune-mnli-t5-base-standard-bahas...       0.79064   \n",
       "\n",
       "                                                    macro f1-score  \n",
       "mesolitica/finetune-mnli-t5-super-tiny-standard...         0.74501  \n",
       "mesolitica/finetune-mnli-t5-tiny-standard-bahas...         0.76542  \n",
       "mesolitica/finetune-mnli-t5-small-standard-baha...         0.78010  \n",
       "mesolitica/finetune-mnli-t5-base-standard-bahas...         0.78918  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "malaya.zero_shot.classification.available_huggingface()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load HuggingFace model\n",
    "\n",
    "```python\n",
    "def huggingface(model: str = 'mesolitica/finetune-mnli-t5-small-standard-bahasa-cased', **kwargs):\n",
    "    \"\"\"\n",
    "    Load HuggingFace model to zeroshot text classification.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    model: str, optional (default='mesolitica/finetune-mnli-t5-small-standard-bahasa-cased')\n",
    "        Check available models at `malaya.zero_shot.classification.available_huggingface()`.\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    result: malaya.torch_model.huggingface.ZeroShotClassification\n",
    "    \"\"\"\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = malaya.zero_shot.classification.huggingface()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### predict batch\n",
    "\n",
    "```python\n",
    "def predict_proba(\n",
    "    self,\n",
    "    strings: List[str],\n",
    "    labels: List[str],\n",
    "    prefix: str = 'ayat ini berkaitan tentang',\n",
    "    multilabel: bool = True,\n",
    "):\n",
    "    \"\"\"\n",
    "    classify list of strings and return probability.\n",
    "\n",
    "    Parameters\n",
    "    ----------\n",
    "    strings: List[str]\n",
    "    labels: List[str]\n",
    "    prefix: str, optional (default='ayat ini berkaitan tentang')\n",
    "        prefix of labels to zero shot. Playing around with prefix can get better results.\n",
    "    multilabel: bool, optional (default=True)\n",
    "        probability of labels can be more than 1.0\n",
    "```\n",
    "\n",
    "Because it is a zero-shot, we need to give labels for the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# copy from twitter\n",
    "\n",
    "string = 'gov macam bengong, kami nk pilihan raya, gov backdoor, sakai'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "You're using a T5TokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[{'najib razak': 0.6651765,\n",
       "  'mahathir': 0.987833,\n",
       "  'kerajaan': 0.9912515,\n",
       "  'PRU': 0.9841426,\n",
       "  'anarki': 0.45587578}]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict_proba([string], labels = ['najib razak', 'mahathir', 'kerajaan', 'PRU', 'anarki'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "string = 'tolong order foodpanda jab, lapar'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'makan': 0.9698464,\n",
       "  'makanan': 0.9735605,\n",
       "  'novel': 0.19823082,\n",
       "  'buku': 0.00313239,\n",
       "  'kerajaan': 0.12976034,\n",
       "  'food delivery': 0.99331254}]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict_proba([string], labels = ['makan', 'makanan', 'novel', 'buku', 'kerajaan', 'food delivery'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "the model understood `order foodpanda` got close relationship with `makan`, `makanan` and `food delivery`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "string = 'kerajaan sebenarnya sangat prihatin dengan rakyat, bagi duit bantuan'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'makan': 0.0004689095,\n",
       "  'makanan': 0.0026079589,\n",
       "  'novel': 0.29850212,\n",
       "  'buku': 0.025044106,\n",
       "  'kerajaan': 0.76523817,\n",
       "  'food delivery': 0.0044676424,\n",
       "  'kerajaan jahat': 0.0023713536,\n",
       "  'kerajaan prihatin': 0.9468328,\n",
       "  'bantuan rakyat': 0.9923975}]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict_proba([string], labels = ['makan', 'makanan', 'novel', 'buku', 'kerajaan', 'food delivery',\n",
    "                                       'kerajaan jahat', 'kerajaan prihatin', 'bantuan rakyat'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### able to infer for mixed MS and EN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "string = 'Hi guys! I noticed semalam & harini dah ramai yang dapat cookies ni kan. So harini i nak share some post mortem of our first batch:'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'makan': 0.17769064,\n",
       "  'makanan': 0.94145703,\n",
       "  'novel': 0.51651853,\n",
       "  'buku': 0.21957111,\n",
       "  'kerajaan': 0.11726684,\n",
       "  'food delivery': 0.903062,\n",
       "  'kerajaan jahat': 0.33357194,\n",
       "  'kerajaan prihatin': 0.14763993,\n",
       "  'bantuan rakyat': 0.5784646,\n",
       "  'biskut': 0.8355128,\n",
       "  'very helpful': 0.39513826,\n",
       "  'sharing experiences': 0.64116335,\n",
       "  'sharing session': 0.675511}]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict_proba([string], labels = ['makan', 'makanan', 'novel', 'buku', 'kerajaan', 'food delivery',\n",
    "                                       'kerajaan jahat', 'kerajaan prihatin', 'bantuan rakyat',\n",
    "                                       'biskut', 'very helpful', 'sharing experiences',\n",
    "                                       'sharing session'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'makan': 0.23804268,\n",
       "  'makanan': 0.94474393,\n",
       "  'novel': 0.8238379,\n",
       "  'buku': 0.3343829,\n",
       "  'kerajaan': 0.092507444,\n",
       "  'food delivery': 0.94236046,\n",
       "  'kerajaan jahat': 0.15810412,\n",
       "  'kerajaan prihatin': 0.13604635,\n",
       "  'bantuan rakyat': 0.55307525,\n",
       "  'biskut': 0.92333925,\n",
       "  'very helpful': 0.39841577,\n",
       "  'sharing experiences': 0.7563246,\n",
       "  'sharing session': 0.86674726}]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict_proba([string], labels = ['makan', 'makanan', 'novel', 'buku', 'kerajaan', 'food delivery',\n",
    "                                       'kerajaan jahat', 'kerajaan prihatin', 'bantuan rakyat',\n",
    "                                       'biskut', 'very helpful', 'sharing experiences',\n",
    "                                       'sharing session'],\n",
    "                   prefix = 'teks ini berkaitan tentang')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Multiclasses but not multilabel\n",
    "\n",
    "Sum of probability equal to 1.0, so to do that, set `multilabel=False`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'makan': 0.0013036507,\n",
       "  'makanan': 0.0012489067,\n",
       "  'novel': 0.007235752,\n",
       "  'buku': 0.0022450346,\n",
       "  'kerajaan': 0.070251726,\n",
       "  'food delivery': 0.0042558503,\n",
       "  'kerajaan jahat': 0.0022728115,\n",
       "  'kerajaan prihatin': 0.20736308,\n",
       "  'bantuan rakyat': 0.57145786,\n",
       "  'biskut': 0.0020565772,\n",
       "  'very helpful': 0.11333891,\n",
       "  'sharing experiences': 0.007458821,\n",
       "  'sharing session': 0.00951114}]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "string = 'kerajaan sebenarnya sangat prihatin dengan rakyat, bagi duit bantuan'\n",
    "\n",
    "model.predict_proba([string], labels = ['makan', 'makanan', 'novel', 'buku', 'kerajaan', 'food delivery',\n",
    "                                       'kerajaan jahat', 'kerajaan prihatin', 'bantuan rakyat',\n",
    "                                       'biskut', 'very helpful', 'sharing experiences',\n",
    "                                       'sharing session'], multilabel = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Stacking models\n",
    "\n",
    "More information, you can read at https://malaya.readthedocs.io/en/latest/Stack.html\n",
    "\n",
    "If you want to stack zero-shot classification models, you need to pass labels using keyword parameter,\n",
    "\n",
    "```python\n",
    "malaya.stack.predict_stack([model1, model2], List[str], labels = List[str])\n",
    "```\n",
    "\n",
    "We will passed `labels` as `**kwargs`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'makan': 0.00046890916,\n",
       "  'makanan': 0.0026079628,\n",
       "  'novel': 0.29850233,\n",
       "  'buku': 0.02504399,\n",
       "  'kerajaan': 0.7652382,\n",
       "  'food delivery': 0.004467653,\n",
       "  'kerajaan jahat': 0.0023713524,\n",
       "  'kerajaan prihatin': 0.9468329,\n",
       "  'bantuan rakyat': 0.99239755,\n",
       "  'comel': 0.00077307917,\n",
       "  'kerajaan syg sgt kepada rakyat': 0.9818335}]"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "string = 'kerajaan sebenarnya sangat prihatin dengan rakyat, bagi duit bantuan'\n",
    "labels = ['makan', 'makanan', 'novel', 'buku', 'kerajaan', 'food delivery', \n",
    " 'kerajaan jahat', 'kerajaan prihatin', 'bantuan rakyat', 'comel', 'kerajaan syg sgt kepada rakyat']\n",
    "malaya.stack.predict_stack([model, model, model], [string], \n",
    "                           labels = labels)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.10"
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}