{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/nathanluskey/opt/anaconda3/envs/ml_env/lib/python3.10/site-packages/tqdm/auto.py:22: 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 DistilBertTokenizer, DistilBertModel, \\\n",
    "                         BertTokenizer, BertModel, \\\n",
    "                         RobertaTokenizer, RobertaModel, \\\n",
    "                         AutoTokenizer, AutoModelForMaskedLM\n",
    "import gradio as gr\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import torch\n",
    "from typing import List, Tuple\n",
    "from sklearn.cluster import KMeans"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# global variables\n",
    "encoder_options = [\n",
    "    'distilbert-base-uncased',\n",
    "    'bert-base-uncased',\n",
    "    'bert-base-cased'\n",
    "    'roberta-base',\n",
    "    'xlm-roberta-base',\n",
    "    ]\n",
    "\n",
    "current_encoder = encoder_options[0]\n",
    "tokenizer = None\n",
    "model = None\n",
    "\n",
    "genres = pd.read_csv(\"./all_genres.csv\")\n",
    "genres = genres[\"genre\"].to_list()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of the model checkpoint at distilbert-base-uncased were not used when initializing DistilBertModel: ['vocab_projector.bias', 'vocab_layer_norm.weight', 'vocab_transform.weight', 'vocab_layer_norm.bias', 'vocab_transform.bias', 'vocab_projector.weight']\n",
      "- This IS expected if you are initializing DistilBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
      "- This IS NOT expected if you are initializing DistilBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
     ]
    }
   ],
   "source": [
    "if current_encoder == 'distilbert-base-uncased':\n",
    "    tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')\n",
    "    model = DistilBertModel.from_pretrained('distilbert-base-uncased')\n",
    "elif current_encoder == 'bert-base-uncased':\n",
    "    tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')\n",
    "    model = BertModel.from_pretrained('bert-base-uncased')\n",
    "elif current_encoder == 'bert-base-cased':\n",
    "    tokenizer = BertTokenizer.from_pretrained('bert-base-cased')\n",
    "    model = BertModel.from_pretrained('bert-base-cased')\n",
    "elif current_encoder == 'roberta-base':\n",
    "    tokenizer = RobertaTokenizer.from_pretrained('roberta-base')\n",
    "    model = RobertaModel.from_pretrained('roberta-base')\n",
    "elif current_encoder == 'xlm-roberta-base':\n",
    "    tokenizer = AutoTokenizer.from_pretrained('xlm-roberta-base')\n",
    "    model = AutoModelForMaskedLM.from_pretrained('xlm-roberta-base')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def embed_string() -> np.ndarray:\n",
    "    output = []\n",
    "    for text in genres:\n",
    "        encoded_input = tokenizer(text, return_tensors='pt')\n",
    "        # forward pass\n",
    "        new_output = model(**encoded_input)\n",
    "        to_append = new_output.last_hidden_state\n",
    "        to_append = to_append[:, -1, :] #Take the last element\n",
    "        to_append = to_append.flatten().detach().cpu().numpy()\n",
    "        output.append(to_append)\n",
    "    np_output = np.zeros((len(output), output[0].shape[0]))\n",
    "    for i, vector in enumerate(output):\n",
    "        np_output[i, :] = vector\n",
    "    return np_output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gen_clusters(input_strs:np.ndarray, num_clusters:int) -> Tuple[KMeans, np.ndarray, float]:\n",
    "    clustering_algo = KMeans(n_clusters=num_clusters)\n",
    "    predicted_labels = clustering_algo.fit_predict(input_strs)\n",
    "\n",
    "    cluster_error = 0.0\n",
    "    for i, predicted_label in enumerate(predicted_labels):\n",
    "        predicted_center = clustering_algo.cluster_centers_[predicted_label, :]\n",
    "        new_error = np.sqrt(np.sum(np.square(predicted_center, input_strs[i])))\n",
    "        cluster_error += new_error\n",
    "\n",
    "    return clustering_algo, predicted_labels, cluster_error\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "def view_clusters(predicted_clusters:np.ndarray) -> pd.DataFrame:\n",
    "    mappings = dict()\n",
    "    for predicted_cluster, movie in zip(predicted_clusters, genres):\n",
    "        curr_mapping = mappings.get(predicted_cluster, [])\n",
    "        curr_mapping.append(movie)\n",
    "        mappings[predicted_cluster] = curr_mapping\n",
    "\n",
    "    output_df = pd.DataFrame()\n",
    "    max_len = max([len(x) for x in mappings.values()])\n",
    "    max_cluster = max(predicted_clusters)\n",
    "\n",
    "    for i in range(max_cluster + 1):\n",
    "        new_column_name = f\"cluster_{i}\"\n",
    "        new_column_data = mappings[i]\n",
    "        new_column_data.extend([''] * (max_len - len(new_column_data)))\n",
    "        output_df[new_column_name] = new_column_data\n",
    "\n",
    "    return output_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def add_new_genre(clustering_algo:KMeans, new_genre:str, recompute:bool = False) -> pd.DataFrame:\n",
    "    global genres\n",
    "    genres.append(new_genre)\n",
    "    embedded_genres = embed_string()\n",
    "    if recompute:\n",
    "        cluster_algo, cluster_centers, error = gen_clusters(embedded_genres, 5)\n",
    "    else:\n",
    "       cluster_centers = cluster_algo.predict(embedded_genres)\n",
    "    \n",
    "    ouput_df = view_clusters(cluster_centers)\n",
    "    return ouput_df\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedded_genres = embed_string()\n",
    "clustering_algo, predicted_labels, cluster_error = gen_clusters(embedded_genres, 5)\n",
    "output_df = view_clusters(predicted_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.10.6 ('ml_env')",
   "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": "3.10.6"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "2434bee09bcd67f653a1f2d2df1f4f18cabf9d6c39b42950acaa6ef605d590bc"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}