{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "From this blog: [How to Implement a Beam Search Decoder for Natural Language Processing](https://machinelearningmastery.com/beam-search-decoder-natural-language-processing/)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math\n",
    "import numpy as np\n",
    "from pprint import pprint"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = [[0.1, 0.2, 0.3, 0.4, 0.5],\n",
    "        [0.5, 0.4, 0.3, 0.2, 0.1],\n",
    "        [0.1, 0.2, 0.3, 0.4, 0.5],\n",
    "        [0.5, 0.4, 0.3, 0.2, 0.1],\n",
    "        [0.1, 0.2, 0.3, 0.4, 0.5],\n",
    "        [0.5, 0.4, 0.3, 0.2, 0.1],\n",
    "        [0.1, 0.2, 0.3, 0.4, 0.5],\n",
    "        [0.5, 0.4, 0.3, 0.2, 0.1],\n",
    "        [0.1, 0.2, 0.3, 0.4, 0.5],\n",
    "        [0.5, 0.4, 0.3, 0.2, 0.1]]\n",
    "data = np.array(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def greedy_decoder(data):\n",
    "    return [np.argmax(s) for s in data]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# beam search\n",
    "def beam_search_decoder(data, k):\n",
    "    sequences = [[list(), 1.0]]\n",
    "    for row in data:\n",
    "        all_candidates = list()\n",
    "        for i in range(len(sequences)):\n",
    "            seq, score = sequences[i]\n",
    "            for j in range(len(row)):\n",
    "                candidate = [seq + [j], score * -math.log(row[j]+1e-100)]\n",
    "                all_candidates.append(candidate)\n",
    "#         pprint((\"all candidates: \", all_candidates))\n",
    "        ordered = sorted(all_candidates, key=lambda tup:tup[1])\n",
    "#         pprint((\"ordered: \", ordered))\n",
    "        print(\"=\"*50)\n",
    "        sequences = ordered[:k]\n",
    "        pprint((\"sequence: \", sequences))\n",
    "    return sequences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[4, 0, 4, 0, 4, 0, 4, 0, 4, 0]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "greedy_decoder(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4], 0.6931471805599453],\n",
      "  [[3], 0.916290731874155],\n",
      "  [[2], 1.2039728043259361]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0], 0.4804530139182014],\n",
      "  [[4, 1], 0.6351243373717793],\n",
      "  [[3, 0], 0.6351243373717793]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4], 0.33302465198892944],\n",
      "  [[4, 0, 3], 0.4402346437542523],\n",
      "  [[4, 1, 4], 0.4402346437542523]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0], 0.23083509858308343],\n",
      "  [[4, 0, 3, 0], 0.3051474021030719],\n",
      "  [[4, 1, 4, 0], 0.3051474021030719]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0, 4], 0.1600026977571413],\n",
      "  [[4, 0, 3, 0, 4], 0.21151206142293622],\n",
      "  [[4, 1, 4, 0, 4], 0.21151206142293622]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0, 4, 0], 0.11090541883234757],\n",
      "  [[4, 0, 4, 0, 4, 1], 0.1466089890297302],\n",
      "  [[4, 0, 3, 0, 4, 0], 0.1466089890297302]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0, 4, 0, 4], 0.07687377837246158],\n",
      "  [[4, 0, 4, 0, 4, 0, 3], 0.10162160739070145],\n",
      "  [[4, 0, 4, 0, 4, 1, 4], 0.10162160739070145]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0, 4, 0, 4, 0], 0.05328484273786184],\n",
      "  [[4, 0, 4, 0, 4, 0, 4, 1], 0.07043873064683441],\n",
      "  [[4, 0, 4, 0, 4, 0, 3, 0], 0.07043873064683441]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0, 4, 0, 4, 0, 4], 0.03693423851032901],\n",
      "  [[4, 0, 4, 0, 4, 0, 4, 0, 3], 0.04882440755007468],\n",
      "  [[4, 0, 4, 0, 4, 0, 4, 1, 4], 0.04882440755007468]])\n",
      "==================================================\n",
      "('sequence: ',\n",
      " [[[4, 0, 4, 0, 4, 0, 4, 0, 4, 0], 0.025600863289563108],\n",
      "  [[4, 0, 4, 0, 4, 0, 4, 0, 4, 1], 0.03384250043584397],\n",
      "  [[4, 0, 4, 0, 4, 0, 4, 0, 3, 0], 0.03384250043584397]])\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[[[4, 0, 4, 0, 4, 0, 4, 0, 4, 0], 0.025600863289563108],\n",
       " [[4, 0, 4, 0, 4, 0, 4, 0, 4, 1], 0.03384250043584397],\n",
       " [[4, 0, 4, 0, 4, 0, 4, 0, 3, 0], 0.03384250043584397]]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "beam_search_decoder(data, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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": "3.7.4"
  },
  "latex_envs": {
   "LaTeX_envs_menu_present": true,
   "autoclose": false,
   "autocomplete": true,
   "bibliofile": "biblio.bib",
   "cite_by": "apalike",
   "current_citInitial": 1,
   "eqLabelWithNumbers": true,
   "eqNumInitial": 1,
   "hotkeys": {
    "equation": "Ctrl-E",
    "itemize": "Ctrl-I"
   },
   "labels_anchors": false,
   "latex_user_defs": false,
   "report_style_numbering": false,
   "user_envs_cfg": false
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}