simulated_annealing_clustering.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from random import choices, randrange\n",
    "import random\n",
    "import matplotlib.pyplot as plt\n",
    "from numba import njit\n",
    "from sklearn.datasets import make_blobs\n",
    "from utility import *\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Simulated Annealing\n",
    "\n",
    "Implementazione dell'algoritmo di simulated annealing per la soluzione del problema di clustering.\n",
    "\n",
    "L'algoritmo parte da una soluzione di bassa qualità realizzata tramite un semplice algoritmo che inizializza casualmente i centroidi e che associa ogni punto al cluster con il centroide più vicino.\n",
    "\n",
    "Data una soluzione la soluzione candidato, che viene valutata per ogni iteazione del SA è ottenuta selezionando un nodo casualmente e spostandolo da un cluster ad un altro.\n",
    "\n",
    "La funzione change_sol effettua la mossa sopra descritta per generare la soluzione candidato.\n",
    "\n",
    "La funzione simulated annealing richiede l'inizializzazione di quattro parametri:\n",
    "*   $T_i$ e $T_f$ : rispettivamente la temperatura iniziale e finale dell'algoritmo. La prima regola la probabilità con il quale soluzioni pegiorative sono accettate, l'altra è utilizzata per il criterio d'arresto.\n",
    "*   $\\alpha$ : Il parametro di raffreddamento, tra un'iterazione e la successiva la temperatura è ridotta secondo la seguente legge esponenziale : $T_{k+1} = \\alpha * T_k$.\n",
    "Un parametro di $\\alpha$ prossimo ad 1 garantisce un numero di iterazioni molto grande ma anche delle soluzioni migliori.\n",
    "*   Iter : Numero di iterazioni per una fissata temperatura. All'aumentare di questo parametro il valore di $\\alpha$ deve essere ridotto.\n",
    "\n",
    "L'euristica utilizza come metrica la funzione di squared_inner_distance definita in utiliy.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "@njit\n",
    "def change_sol(sol, k):\n",
    "    new_sol = sol.copy()\n",
    "    n = randrange(len(new_sol))#ritorna un numero casualmente nel range (0, len(new_sol))\n",
    "    old_value = new_sol[n]\n",
    "\n",
    "    while old_value == new_sol[n]:\n",
    "        new_sol[n] = randrange(K)\n",
    "\n",
    "    return new_sol"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "@njit\n",
    "def simulated_annealing(base_sol, points, K, iters, alpha, Ti = 10, Tf = 10**-8, verbose = True):\n",
    "    best_sol = base_sol\n",
    "    base_val = squared_inner_distance(best_sol, points, K)\n",
    "    best_value = base_val\n",
    "    T = Ti\n",
    "    curr_sol = base_sol\n",
    "    curr_val = best_value\n",
    "    \n",
    "    iter = 1\n",
    "    old_sol = base_sol\n",
    "    finito = False\n",
    "    no_update = 0\n",
    "\n",
    "    while(finito == False):\n",
    "        if(verbose and iter%10 == 1):\n",
    "            print(\"Iteration number:\", iter, \"Best value percentuale: \", curr_val/base_val*100, \"% T:\", T, no_update)\n",
    "        \n",
    "        iter = 1 + iter\n",
    "        old_sol = curr_sol\n",
    "        \n",
    "        for i in range(iters):\n",
    "            candidate = change_sol(curr_sol, K)\n",
    "            val_candidate = squared_inner_distance(candidate, points, K)\n",
    "\n",
    "            if(val_candidate < best_value):\n",
    "                best_value = val_candidate#viene memorizzato il best value, perchè potrei perderlo nelle iterazioni successive\n",
    "                best_sol = candidate\n",
    "            \n",
    "            if(val_candidate < curr_val):\n",
    "                curr_val = val_candidate\n",
    "                curr_sol = candidate\n",
    "                no_update = 0\n",
    "            else:\n",
    "                r = random.uniform(0, 1)\n",
    "                delta = abs(curr_val - val_candidate)\n",
    "                tresh = np.exp(-delta/T)\n",
    "                if(r < tresh):\n",
    "                    curr_val = val_candidate\n",
    "                    curr_sol = candidate\n",
    "                    no_update = 0\n",
    "                else:\n",
    "                    no_update += 1\n",
    "\n",
    "        if(no_update >= 10000 or T < Tf):\n",
    "            finito = True\n",
    "            break\n",
    "        \n",
    "        T = alpha*T\n",
    "    return best_sol"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Esempio di utilizzo del Simulated annealing\n",
    "\n",
    "Testiamo l'euristica dell'euristica con un istanza di 1000 punti di $R^2$ e un numero di cluster pari K pari a 10.\n",
    "\n",
    "L'insieme di punti su cui è testata l'euristica è generato tramite la funzione make_blobs che genera un insieme di punti di $R^n$ con specifici parametri regolabili.\n",
    "\n",
    "La scelta dei parametri di $T_i$, $T_f$, $\\alpha$, ed Iter è stata fatta sperimentalmente. \n",
    "\n",
    "Abbiamo osservato come all'aumentare del numero di iterazioni la soluzione prodotta dall'euristica migliori drasticamente ma peggiori allo stesso tempo il tempo di esecuzione.\n",
    "Aumentare il numero di iterazioni significa però anche aumentare la durata della fase peggiorativa dell'algoritmo, può quindi essere necessario, all'aumentare di questo paramentro, ridurre la temperatura iniziale."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iteration number: 1 Best value percentuale:  100.0 % T: 100.0 0\n",
      "Iteration number: 11 Best value percentuale:  496.6713844485589 % T: 90.43820750088045 0\n",
      "Iteration number: 21 Best value percentuale:  709.9674833623337 % T: 81.79069375972308 0\n",
      "Iteration number: 31 Best value percentuale:  852.4271145633533 % T: 73.97003733882804 1\n",
      "Iteration number: 41 Best value percentuale:  927.4492874168778 % T: 66.89717585696805 1\n",
      "Iteration number: 51 Best value percentuale:  926.5306395489139 % T: 60.500606713753655 0\n",
      "Iteration number: 61 Best value percentuale:  931.4764335843203 % T: 54.71566423907614 0\n",
      "Iteration number: 71 Best value percentuale:  899.1218638893331 % T: 49.483865960020715 0\n",
      "Iteration number: 81 Best value percentuale:  859.2622059152619 % T: 44.752321376381076 0\n",
      "Iteration number: 91 Best value percentuale:  797.9434039366872 % T: 40.473197267832404 0\n",
      "Iteration number: 101 Best value percentuale:  768.1227832361036 % T: 36.60323412732294 1\n",
      "Iteration number: 111 Best value percentuale:  718.768306083758 % T: 33.10330883210141 0\n",
      "Iteration number: 121 Best value percentuale:  695.414212774229 % T: 29.93803913123316 0\n",
      "Iteration number: 131 Best value percentuale:  625.9962883194265 % T: 27.075425951199424 2\n",
      "Iteration number: 141 Best value percentuale:  570.9250913210703 % T: 24.48652990349296 1\n",
      "Iteration number: 151 Best value percentuale:  511.6870036155248 % T: 22.145178723886104 7\n",
      "Iteration number: 161 Best value percentuale:  469.52440387541685 % T: 20.027702685748938 2\n",
      "Iteration number: 171 Best value percentuale:  424.17043567266444 % T: 18.11269531259703 3\n",
      "Iteration number: 181 Best value percentuale:  385.02351000081785 % T: 16.380796970808742 0\n",
      "Iteration number: 191 Best value percentuale:  348.2723652002432 % T: 14.814499154757945 1\n",
      "Iteration number: 201 Best value percentuale:  332.6807730785935 % T: 13.397967485796167 0\n",
      "Iteration number: 211 Best value percentuale:  309.74204074354355 % T: 12.116881635704834 0\n",
      "Iteration number: 221 Best value percentuale:  294.1132767655137 % T: 10.958290556334813 1\n",
      "Iteration number: 231 Best value percentuale:  274.4304006657562 % T: 9.910481551887466 0\n",
      "Iteration number: 241 Best value percentuale:  255.69152434663667 % T: 8.962861870232462 4\n",
      "Iteration number: 251 Best value percentuale:  243.94892468527308 % T: 8.105851616218128 3\n",
      "Iteration number: 261 Best value percentuale:  229.713408655929 % T: 7.330786904388821 7\n",
      "Iteration number: 271 Best value percentuale:  226.00749844532828 % T: 6.6298322720385325 4\n",
      "Iteration number: 281 Best value percentuale:  213.16959784838173 % T: 5.9959014671465445 1\n",
      "Iteration number: 291 Best value percentuale:  197.1534414602695 % T: 5.422585810406325 1\n",
      "Iteration number: 301 Best value percentuale:  183.90629049424206 % T: 4.9040894071285726 3\n",
      "Iteration number: 311 Best value percentuale:  174.54765400584742 % T: 4.4351705540476365 9\n",
      "Iteration number: 321 Best value percentuale:  161.5838186222856 % T: 4.011088748687548 19\n",
      "Iteration number: 331 Best value percentuale:  149.5902050848471 % T: 3.6275567655825145 8\n",
      "Iteration number: 341 Best value percentuale:  138.0731459798303 % T: 3.2806973148697414 16\n",
      "Iteration number: 351 Best value percentuale:  130.57127795928707 % T: 2.967003845097709 15\n",
      "Iteration number: 361 Best value percentuale:  126.16446180319602 % T: 2.6833050939885674 29\n",
      "Iteration number: 371 Best value percentuale:  117.99155271501252 % T: 2.426733028783076 3\n",
      "Iteration number: 381 Best value percentuale:  113.64532302380736 % T: 2.194693852063239 16\n",
      "Iteration number: 391 Best value percentuale:  106.67298480825997 % T: 1.9848417799380185 5\n",
      "Iteration number: 401 Best value percentuale:  101.81189792462806 % T: 1.7950553275045138 6\n",
      "Iteration number: 411 Best value percentuale:  95.37571235578945 % T: 1.623415861844141 12\n",
      "Iteration number: 421 Best value percentuale:  90.35476537807352 % T: 1.468188205736811 43\n",
      "Iteration number: 431 Best value percentuale:  86.34748397605279 % T: 1.3278030960077105 5\n",
      "Iteration number: 441 Best value percentuale:  83.14305449106438 % T: 1.2008413191705682 61\n",
      "Iteration number: 451 Best value percentuale:  77.75688431743677 % T: 1.0860193639877882 18\n",
      "Iteration number: 461 Best value percentuale:  75.02233743595485 % T: 0.982176445903018 21\n",
      "Iteration number: 471 Best value percentuale:  71.9534763345418 % T: 0.8882627721705442 16\n",
      "Iteration number: 481 Best value percentuale:  69.89593574027782 % T: 0.8033289290486697 15\n",
      "Iteration number: 491 Best value percentuale:  66.54636494218158 % T: 0.7265162837676364 91\n",
      "Iteration number: 501 Best value percentuale:  64.19557520607778 % T: 0.6570483042414603 39\n",
      "Iteration number: 511 Best value percentuale:  61.93670638587364 % T: 0.594222708770908 69\n",
      "Iteration number: 521 Best value percentuale:  61.121081035419735 % T: 0.5374043663755863 44\n",
      "Iteration number: 531 Best value percentuale:  59.23619983609275 % T: 0.4860188759815446 81\n",
      "Iteration number: 541 Best value percentuale:  57.548615035650656 % T: 0.4395467595536361 87\n",
      "Iteration number: 551 Best value percentuale:  55.479734080945676 % T: 0.39751821046851343 100\n",
      "Iteration number: 561 Best value percentuale:  54.61035871293126 % T: 0.3595083440373008 24\n",
      "Iteration number: 571 Best value percentuale:  52.047546416048775 % T: 0.32513290216343316 68\n",
      "Iteration number: 581 Best value percentuale:  49.71218945716999 % T: 0.29404436871220024 12\n",
      "Iteration number: 591 Best value percentuale:  49.15481366891169 % T: 0.2659284563205936 2\n",
      "Iteration number: 601 Best value percentuale:  48.8349221323062 % T: 0.24050092913110663 33\n",
      "Iteration number: 611 Best value percentuale:  45.942350567637035 % T: 0.21750472932913562 184\n",
      "Iteration number: 621 Best value percentuale:  45.38296164333266 % T: 0.19670737843491204 141\n",
      "Iteration number: 631 Best value percentuale:  44.774968526854934 % T: 0.17789862707850793 218\n",
      "Iteration number: 641 Best value percentuale:  44.49925780497048 % T: 0.16088832949847845 140\n",
      "Iteration number: 651 Best value percentuale:  44.367031181735456 % T: 0.14550452127653415 229\n",
      "Iteration number: 661 Best value percentuale:  44.10340906647097 % T: 0.1315916808752347 166\n",
      "Iteration number: 671 Best value percentuale:  43.86377915281651 % T: 0.11900915740384119 10\n",
      "Iteration number: 681 Best value percentuale:  43.188727210219504 % T: 0.1076297487179353 240\n",
      "Iteration number: 691 Best value percentuale:  43.04261768978289 % T: 0.09733841547820254 23\n",
      "Iteration number: 701 Best value percentuale:  42.71555208837361 % T: 0.08803111816824594 252\n",
      "Iteration number: 711 Best value percentuale:  42.65809598312068 % T: 0.07961376531434353 48\n",
      "Iteration number: 721 Best value percentuale:  42.57865368363448 % T: 0.07200126227425 326\n",
      "Iteration number: 731 Best value percentuale:  42.51961181182007 % T: 0.06511665097883937 278\n",
      "Iteration number: 741 Best value percentuale:  42.51863983291469 % T: 0.05889033192986684 739\n",
      "Iteration number: 751 Best value percentuale:  42.25976288250312 % T: 0.05325936058869023 317\n",
      "Iteration number: 761 Best value percentuale:  42.20651950594544 % T: 0.04816681104284182 29\n",
      "Iteration number: 771 Best value percentuale:  41.851301794118626 % T: 0.04356120051748228 662\n",
      "Iteration number: 781 Best value percentuale:  41.84697020451631 % T: 0.03939596891387523 841\n",
      "Iteration number: 791 Best value percentuale:  41.838800782969585 % T: 0.035629008113312835 505\n",
      "Iteration number: 801 Best value percentuale:  41.66048639156635 % T: 0.03222223628802339 66\n",
      "Iteration number: 811 Best value percentuale:  41.527043401473 % T: 0.029141212915586592 986\n",
      "Iteration number: 821 Best value percentuale:  41.51110408979093 % T: 0.026354790604871576 130\n",
      "Iteration number: 831 Best value percentuale:  41.51110408979093 % T: 0.023834800213656308 1130\n",
      "Iteration number: 841 Best value percentuale:  41.44354665838264 % T: 0.021555766074646786 793\n",
      "Iteration number: 851 Best value percentuale:  41.44354665838264 % T: 0.019494648450993447 1793\n",
      "Iteration number: 861 Best value percentuale:  41.4425551603651 % T: 0.017630610617676627 913\n",
      "Iteration number: 871 Best value percentuale:  41.42412250196691 % T: 0.01594480821408665 461\n",
      "Iteration number: 881 Best value percentuale:  41.42412250196691 % T: 0.014420198738273111 1461\n",
      "Iteration number: 891 Best value percentuale:  41.3771766520438 % T: 0.01304136925695878 50\n",
      "Iteration number: 901 Best value percentuale:  41.3771766520438 % T: 0.011794380589564411 1050\n",
      "Iteration number: 911 Best value percentuale:  41.37444947670285 % T: 0.010666626391033827 264\n",
      "Iteration number: 921 Best value percentuale:  41.368333253490285 % T: 0.009646705708866846 805\n",
      "Iteration number: 931 Best value percentuale:  41.368333253490285 % T: 0.008724307725984277 1805\n",
      "Iteration number: 941 Best value percentuale:  41.35660952580537 % T: 0.007890107524241003 180\n",
      "Iteration number: 951 Best value percentuale:  41.34779303288026 % T: 0.00713567181481566 210\n",
      "Iteration number: 961 Best value percentuale:  41.34779303288026 % T: 0.006453373682464827 1210\n",
      "Iteration number: 971 Best value percentuale:  41.34779303288026 % T: 0.005836315481754749 2210\n",
      "Iteration number: 981 Best value percentuale:  41.34779303288026 % T: 0.005278259105795369 3210\n",
      "Iteration number: 991 Best value percentuale:  41.32060758369726 % T: 0.004773562922533333 971\n",
      "Iteration number: 1001 Best value percentuale:  41.32060758369726 % T: 0.004317124741065788 1971\n",
      "Iteration number: 1011 Best value percentuale:  41.32060758369726 % T: 0.003904330231396925 2971\n",
      "Iteration number: 1021 Best value percentuale:  41.15419268886863 % T: 0.0035310062761903568 504\n",
      "Iteration number: 1031 Best value percentuale:  41.15280226381148 % T: 0.0031933787829301467 74\n",
      "Iteration number: 1041 Best value percentuale:  41.15280226381148 % T: 0.0028880345299954566 1074\n",
      "Iteration number: 1051 Best value percentuale:  41.15280226381148 % T: 0.0026118866609343683 2074\n",
      "Iteration number: 1061 Best value percentuale:  41.15087667983992 % T: 0.0023621434781036415 2\n",
      "Iteration number: 1071 Best value percentuale:  41.15087667983992 % T: 0.002136280220195885 1002\n",
      "Iteration number: 1081 Best value percentuale:  41.15087667983992 % T: 0.0019320135383410198 2002\n",
      "Iteration number: 1091 Best value percentuale:  41.15087667983992 % T: 0.0017472784127499541 3002\n",
      "Iteration number: 1101 Best value percentuale:  41.150314144477804 % T: 0.0015802072765408938 864\n",
      "Iteration number: 1111 Best value percentuale:  41.150314144477804 % T: 0.0014291111357020649 1864\n",
      "Iteration number: 1121 Best value percentuale:  41.150314144477804 % T: 0.0012924624943244226 2864\n",
      "Iteration number: 1131 Best value percentuale:  41.150314144477804 % T: 0.0011688799124881767 3864\n",
      "Iteration number: 1141 Best value percentuale:  41.150314144477804 % T: 0.001057114040692167 4864\n",
      "1.81603E+03\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "N = 1000\n",
    "K = 10\n",
    "\n",
    "points, centroids = make_blobs(n_samples=N, centers=K, n_features=2, random_state=np.random.randint(100))\n",
    "\n",
    "sol = create_initial_sol(points, K)\n",
    "sol = simulated_annealing(sol, points, K, 100, 0.99, 100, 0.001)\n",
    "\n",
    "print(\"{:.5E}\".format(squared_inner_distance(sol, points, K)))\n",
    "\n",
    "printR2sol(points, sol, K)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8.5 ('base')",
   "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"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "e03b612d84ba21ce95ed447e81b3062e1eb99b56c6d885cdab4aaa12f1b8e240"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}