analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 295,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import scipy.linalg as la\n",
    "from scipy import stats\n",
    "\n",
    "import math\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.mlab as mlab\n",
    "\n",
    "import seaborn as sns\n",
    "# sns.set(color_codes=False)\n",
    "from scipy.sparse import dok_matrix, csr_matrix, diags, linalg\n",
    "import scipy.linalg as la\n",
    "\n",
    "import dask.array as da\n",
    "\n",
    "from hyperwords.representations.matrix_serializer import load_matrix, load_vocabulary, load_count_vocabulary\n",
    "from hyperwords.representations.embedding import SVDEmbedding, EnsembleEmbedding, Embedding\n",
    "from hyperwords.representations.explicit import Explicit, PositiveExplicit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 296,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# svd100_one = SVDEmbedding('w2.dyn.dirty.100/svd', True, 1)\n",
    "svd100_none = SVDEmbedding('w2.dyn.dirty.100/svd', True, 0)\n",
    "svd = svd100_none\n",
    "# svd100 = SVDEmbedding('w2.dyn.dirty.100/svd', True, 0.5)\n",
    "# svd10 = SVDEmbedding('w2.dyn.dirty.10/svd', True, 0.5)\n",
    "# svd1 = SVDEmbedding('w2.dyn.dirty.1/svd', True, 0.5)\n",
    "# svd01 = SVDEmbedding('w2.dyn.dirty.0.1/svd', True, 0.5)\n",
    "# svd001 = SVDEmbedding('w2.dyn.dirty.0.01/svd', True, 0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 297,
   "metadata": {},
   "outputs": [],
   "source": [
    "sgns100 = Embedding('w2.dyn.dirty.100/sgns.words', True)\n",
    "sgns = sgns100\n",
    "# sgns10 = Embedding('w2.dyn.dirty.10/sgns.words', True)\n",
    "# sgns1 = Embedding('w2.dyn.dirty.1/sgns.words', True)\n",
    "# sgns01 = Embedding('w2.dyn.dirty.0.1/sgns.words', True)\n",
    "# sgns001 = Embedding('w2.dyn.dirty.0.01/sgns.words', True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 308,
   "metadata": {},
   "outputs": [],
   "source": [
    "sgns100_s = la.svd(sgns100.m, compute_uv=False, full_matrices=False, check_finite=False)\n",
    "# sgns10_s = la.svd(sgns10.m, compute_uv=False, full_matrices=False, check_finite=False)\n",
    "# sgns1_s = la.svd(sgns1.m, compute_uv=False, full_matrices=False, check_finite=False)\n",
    "# sgns01_s = la.svd(sgns01.m, compute_uv=False, full_matrices=False, check_finite=False)\n",
    "# sgns001_s = la.svd(sgns001.m, compute_uv=False, full_matrices=False, check_finite=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# pmi = Explicit('w2.dyn.dirty.0.01/pmi', normalize=False)\n",
    "# m = pmi.m.todense()\n",
    "# np.all(m == m.T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# cooc = Explicit('w2.dyn.dirty.0.01/pmi.cooc', normalize=False)\n",
    "# m = cooc.m.todense()\n",
    "# np.all(m == m.T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ppmi = PositiveExplicit('w2.dyn.dirty.0.01/pmi', normalize=False)\n",
    "# m = ppmi.m.todense()\n",
    "# np.all(m == m.T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# d = np.diag(m)\n",
    "# d"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # svd = Embedding('w2.dyn.dirty.0.01/sgns.words', False)\n",
    "# # svd.m = (svd100_none.m + svd100_none_c.m) / 2\n",
    "# # svd.wi, svd.iw = (svd100.wi, svd100.iw)\n",
    "# svd = svd100_none\n",
    "# # sgns = sgns100\n",
    "# # sgns_s = sgns100_s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# rnorm = np.sqrt(np.sum(svd.m * svd.m, axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 300,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.0"
      ]
     },
     "execution_count": 300,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cnorm = np.sqrt(np.sum(svd.m * svd.m, axis=0))\n",
    "svdcs = svd.m / cnorm[np.newaxis]\n",
    "np.sqrt(np.sum(svdcs[:, 0] * svdcs[:, 0]))\n",
    "# svd.m = svdcs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 301,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.0000471"
      ]
     },
     "execution_count": 301,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cnorm = np.sqrt(np.sum(sgns.m * sgns.m, axis=0))\n",
    "sgnscs = sgns.m / cnorm[np.newaxis]\n",
    "np.sqrt(np.sum(sgnscs[:, 0] * sgnscs[:, 0]))\n",
    "# svd.m = svdcs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 302,
   "metadata": {},
   "outputs": [],
   "source": [
    "def ipr(v, p=4):\n",
    "    return np.sum(np.power(v, p))\n",
    "get_iprs = lambda m: [ipr(m[:, i]) for i in range(500)]\n",
    "iprs_svd = np.array(get_iprs(svd.m))\n",
    "iprs_svdcs = np.array(get_iprs(svdcs))\n",
    "iprs_sgnscs = np.array(get_iprs(sgnscs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 303,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.05151469236172347"
      ]
     },
     "execution_count": 303,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.std(np.array(list(map(lambda ipr: 1/ipr, iprs_svdcs)))) / 346000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 304,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.2759569103251422"
      ]
     },
     "execution_count": 304,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(np.array(list(map(lambda ipr: 1/ipr, iprs_svdcs)))) / 346000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 305,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.046946430906232274"
      ]
     },
     "execution_count": 305,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "1 / np.max(np.array(list(map(lambda ipr: ipr, iprs_svdcs)))) / 346000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 317,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.03907754300124037\n",
      "0.3601664422048553\n",
      "0.2706608988975468\n",
      "0.5598168843601704\n"
     ]
    }
   ],
   "source": [
    "print(np.std(np.array(list(map(lambda ipr: 1/ipr, iprs_sgnscs)))) / 346000)\n",
    "print(np.mean(np.array(list(map(lambda ipr: 1/ipr, iprs_sgnscs)))) / 346000)\n",
    "print(1 / np.max(np.array(list(map(lambda ipr: ipr, iprs_sgnscs)))) / 346000)\n",
    "print(1 / np.min(np.array(list(map(lambda ipr: ipr, iprs_sgnscs)))) / 346000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[  0  20   5   7  11  37   8  34  36  16 153  44  26 184  56  19  32 188\n",
      "  59  93   9 161  41  31 179  86  25 158  65 139  84 245 191  91 177 215\n",
      " 112   4 163 174  10 116 111  53  82 133 221  29 243 198 144 264 246 255\n",
      " 228 204 241 292 108 258 280 293 273 247 160 299 211 251 169 182 167 236\n",
      " 266 114 234 138 194 185  94 189 124 224 240  85 106 334 196 122 166  48\n",
      " 146 223 225 115  43 172 219 277 129 173 285 237 271 109 104 249 164 131\n",
      " 260 176 306 305 208 197 202 275  47 134 317 254 206 193 286 296  97 235\n",
      " 183 239  51 244 257 130 181 180 262  71  60 279  58 150 295 252 261  67\n",
      " 168 123  95 165 186  87  30 126 125  73 268 248 303 220  38  28 373 233\n",
      "  90  40 205 350 337  92 201 309 333 127 326 210 320 316 302 324 308 342\n",
      " 291 250  99 298 259 222 120 214 141 374  45 304 117 321 276 156  83 230\n",
      "  46 319  17 157 335 325 353 253 362 314 386 329 281 307 278 300 297 227\n",
      " 192 359 207 318 344 422 354 132 170 269  74 392 338 389 226 312 351 343\n",
      " 137 256 110 331 383 290 352 395 213 232 341  22 145  54 311 216 159 270\n",
      " 367 336  89 348 347 345 152 263 363 360 370 136 372 218 265 274 355 328\n",
      " 327 378 384 369 142   1 371 267 403 365 310 323 162 407 149 346  39  24\n",
      "  69 287 272 357 340 393 282 381 330 195 171 349 410 398 200 368 140 387\n",
      " 313 283 332 400  57 100 212 229 289 242 409 376 390 424 322 379 418 397\n",
      " 356 155 217 151 315 135 404 432 366 339 434 385 414 421 451 284 402 437\n",
      "  55 294 401 143 388 396 413 394 426  49 406 148 415 420 199 459 433 382\n",
      " 408 399 419 448 405 423 364 238 375 430   3 436 417 412 380 469  62 391\n",
      " 441  21 416 231 444 458 427 301 460 103 435 442 358 457 447 105 481  35\n",
      " 455 377 473 463  88  77 453 128 468 492 467 361 446 475 445 428 203 482\n",
      " 449 119  33 494 479 478 470 462 456  72 443 411 438 454 484 429 483 425\n",
      " 452 488 489  76 465  12 113 480  23 477 431 496 486 464 466  79 471 487\n",
      " 497 493 498 495  80 107  68 461  15  98 485 472 450 154 491 476 490 178\n",
      "  18 288 209  61 439  66  63 474  27 499   2 175  78 118 190 147  52 102\n",
      "  81   6 440 187  13  64  70 121  96  75 101  14  50  42]\n"
     ]
    }
   ],
   "source": [
    "iprs_svd_argsort = np.argsort(iprs_svd)\n",
    "# iprs_sgns_argsort = np.argsort(iprs_sgns)\n",
    "# print(iprs_svd[iprs_svd_argsort])\n",
    "print(iprs_svd_argsort)\n",
    "# print(iprs_sgns[iprs_sgns_argsort])\n",
    "# print(iprs_sgns_argsort)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 260,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.04816955684007707"
      ]
     },
     "execution_count": 260,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(1 / 6e-05) / 346000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.32466557041844474"
      ]
     },
     "execution_count": 259,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "1 / 8.90200154786856e-06 / 346000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 312,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f69961a7828>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f6993f3fe48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn import preprocessing\n",
    "# from matplotlib import rcParams\n",
    "# rcParams['font.family'] = 'sans-serif'\n",
    "# rcParams['font.sans-serif'] = 'Helvetica'\n",
    "min_max_scaler = preprocessing.MinMaxScaler()\n",
    "color_map = plt.cm.Spectral_r\n",
    "y = iprs_svdcs\n",
    "x = svd.s\n",
    "x_norm = min_max_scaler.fit_transform([x])\n",
    "y_norm = min_max_scaler.fit_transform([y])\n",
    "# y_norm = y\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "\n",
    "normx=(max(x)-min(x))*0.05\n",
    "normy=(max(y)-min(y))*0.05\n",
    "\n",
    "xbnds = np.array([min(x)-normx,max(x)+normx])\n",
    "ybnds = np.array([min(y)-normy,max(y)+normy])\n",
    "extent = [xbnds[0],xbnds[1],ybnds[0],ybnds[1]]\n",
    "# Set gridsize just to make them visually large\n",
    "image = plt.hexbin(x,y,cmap=color_map,gridsize=20,extent=extent,mincnt=1,bins='log')\n",
    "# Note that mincnt=1 adds 1 to each count\n",
    "counts = image.get_array()\n",
    "ncnts = np.count_nonzero(np.power(10,counts))\n",
    "verts = image.get_offsets()\n",
    "for offc in range(verts.shape[0]):\n",
    "    binx,biny = verts[offc][0],verts[offc][1]\n",
    "    if counts[offc]:\n",
    "        plt.plot(binx,biny,'k.',zorder=100)\n",
    "ax.set_xlim(xbnds)\n",
    "ax.set_ylim(ybnds)\n",
    "# ax = plt.gca()\n",
    "y_labels = [' ','1e-5', '2e-5', '3e-5','4e-5','5e-5','6e-5']\n",
    "ax.set_yticklabels(y_labels)\n",
    "# x_labels = ['0','2e3', '4e3', '6e3','8e3','1e4', '1.2e4']\n",
    "# ax.set_xticklabels(x_labels)\n",
    "\n",
    "# cb = plt.colorbar(image,spacing='uniform',extend='max')\n",
    "\n",
    "# plt.plot(iprs_svd, svd.s, 'go')\n",
    "# plt.xscale(\"log\")\n",
    "# plt.yscale(\"log\")\n",
    "# plt.xlabel(\"IPR\")\n",
    "# plt.ylabel(\"Eigenvalue\")\n",
    "plt.colorbar()\n",
    "# plt.show()\n",
    "# plt.plot(iprs_sgns, sgns_s, 'go')\n",
    "# plt.xscale(\"log\")\n",
    "# plt.yscale(\"log\")\n",
    "plt.ylabel(\"IPR\")\n",
    "plt.xlabel(\"Eigenvalue\")\n",
    "plt.savefig('ipr.png', format='png',dpi=300)\n",
    "plt.show()\n",
    "\n",
    "plt.plot(iprs_svd, svd.s, 'go')\n",
    "# plt.xscale(\"log\")\n",
    "# plt.yscale(\"log\")\n",
    "plt.ylabel(\"IPR\")\n",
    "plt.xlabel(\"Eigenvalue\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 315,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f699577a518>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f699478a198>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn import preprocessing\n",
    "# from matplotlib import rcParams\n",
    "# rcParams['font.family'] = 'sans-serif'\n",
    "# rcParams['font.sans-serif'] = 'Helvetica'\n",
    "min_max_scaler = preprocessing.MinMaxScaler()\n",
    "color_map = plt.cm.Spectral_r\n",
    "y = iprs_sgnscs\n",
    "x = sgns100_s\n",
    "x_norm = min_max_scaler.fit_transform([x])\n",
    "y_norm = min_max_scaler.fit_transform([y])\n",
    "# y_norm = y\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "\n",
    "normx=(max(x)-min(x))*0.05\n",
    "normy=(max(y)-min(y))*0.05\n",
    "\n",
    "xbnds = np.array([min(x)-normx,max(x)+normx])\n",
    "ybnds = np.array([min(y)-normy,max(y)+normy])\n",
    "extent = [xbnds[0],xbnds[1],ybnds[0],ybnds[1]]\n",
    "# Set gridsize just to make them visually large\n",
    "image = plt.hexbin(x,y,cmap=color_map,gridsize=20,extent=extent,mincnt=1,bins='log')\n",
    "# Note that mincnt=1 adds 1 to each count\n",
    "counts = image.get_array()\n",
    "ncnts = np.count_nonzero(np.power(10,counts))\n",
    "verts = image.get_offsets()\n",
    "for offc in range(verts.shape[0]):\n",
    "    binx,biny = verts[offc][0],verts[offc][1]\n",
    "    if counts[offc]:\n",
    "        plt.plot(binx,biny,'k.',zorder=100)\n",
    "ax.set_xlim(xbnds)\n",
    "ax.set_ylim(ybnds)\n",
    "# ax = plt.gca()\n",
    "y_labels = [' ','5e-6', '6e-6', '7e-6','8e-6','9e-6','1e-5']\n",
    "ax.set_yticklabels(y_labels)\n",
    "# x_labels = ['0','2e3', '4e3', '6e3','8e3','1e4', '1.2e4']\n",
    "# ax.set_xticklabels(x_labels)\n",
    "\n",
    "# cb = plt.colorbar(image,spacing='uniform',extend='max')\n",
    "\n",
    "# plt.plot(iprs_svd, svd.s, 'go')\n",
    "# plt.xscale(\"log\")\n",
    "# plt.yscale(\"log\")\n",
    "# plt.xlabel(\"IPR\")\n",
    "# plt.ylabel(\"Eigenvalue\")\n",
    "plt.colorbar()\n",
    "# plt.show()\n",
    "# plt.plot(iprs_sgns, sgns_s, 'go')\n",
    "# plt.xscale(\"log\")\n",
    "# plt.yscale(\"log\")\n",
    "plt.ylabel(\"IPR\")\n",
    "plt.xlabel(\"Eigenvalue\")\n",
    "plt.savefig('ipr.png', format='png',dpi=300)\n",
    "plt.show()\n",
    "\n",
    "plt.plot(iprs_svd, svd.s, 'go')\n",
    "# plt.xscale(\"log\")\n",
    "# plt.yscale(\"log\")\n",
    "plt.ylabel(\"IPR\")\n",
    "plt.xlabel(\"Eigenvalue\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "65\n",
      "\n",
      "1218.9035865313533\n",
      "1233.2178526165962\n",
      "1238.5373810670262\n",
      "1249.7044041650738\n",
      "1252.8879179714388\n",
      "1260.395724916006\n",
      "1266.3971175036208\n",
      "1278.9464706847941\n",
      "1290.713799650007\n",
      "1291.9531265273206\n",
      "1302.8943002065394\n",
      "\n",
      "2.7969232903477767e-05\n",
      "1.2279016098327443e-05\n",
      "1.6483320956214176e-05\n",
      "1.1651438743343272e-05\n",
      "1.76261188412675e-05\n",
      "8.90200154786856e-06\n",
      "2.8855831098217828e-05\n",
      "1.765450370868061e-05\n",
      "1.5252081588920636e-05\n",
      "1.5763115846686307e-05\n",
      "1.2417974792174688e-05\n"
     ]
    }
   ],
   "source": [
    "i = iprs_svd_argsort[10]\n",
    "print(i)\n",
    "print()\n",
    "print(svd.s[i+5])\n",
    "print(svd.s[i+4])\n",
    "print(svd.s[i+3])\n",
    "print(svd.s[i+2])\n",
    "print(svd.s[i+1])\n",
    "print(svd.s[i])\n",
    "print(svd.s[i-1])\n",
    "print(svd.s[i-2])\n",
    "print(svd.s[i-3])\n",
    "print(svd.s[i-4])\n",
    "print(svd.s[i-5])\n",
    "print()\n",
    "print(iprs_svd[i+5])\n",
    "print(iprs_svd[i+4])\n",
    "print(iprs_svd[i+3])\n",
    "print(iprs_svd[i+2])\n",
    "print(iprs_svd[i+1])\n",
    "print(iprs_svd[i])\n",
    "print(iprs_svd[i-1])\n",
    "print(iprs_svd[i-2])\n",
    "print(iprs_svd[i-3])\n",
    "print(iprs_svd[i-4])\n",
    "print(iprs_svd[i-5])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 294,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "499\n",
      "7.386612132547977\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f694953f438>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "i = iprs_svd_argsort[499]\n",
    "i = 499\n",
    "print(i)\n",
    "print(iprs_svd[i])\n",
    "# print(iprs_sgns[i])\n",
    "# kurtosis = stats.kurtosistest(svd.m[:, i])\n",
    "# print(kurtosis)\n",
    "# print(stats.kurtosis(svd.m[:, i]))\n",
    "# skew = stats.skewtest(svd.m[:, i])\n",
    "# print(skew)\n",
    "# print(stats.skew(svd.m[:, i]))\n",
    "\n",
    "mu = np.mean(svd.m[:, i])\n",
    "sigma = np.std(svd.m[:, i])\n",
    "x = np.linspace(mu - 3*sigma, mu + 3*sigma, 100)\n",
    "\n",
    "# plt.xlim((-0.2, 0.2))\n",
    "sns.distplot(svd.m[:, i], kde=False, norm_hist=True)\n",
    "plt.plot(x,mlab.normpdf(x, mu, sigma))\n",
    "# plt.savefig('svd_{}.png'.format(i+1), format='png',dpi=300)\n",
    "plt.show()\n",
    "\n",
    "# mu = np.mean(sgns.m[:, i])\n",
    "# sigma = np.std(sgns.m[:, i])\n",
    "# x = np.linspace(mu - 3*sigma, mu + 3*sigma, 100)\n",
    "\n",
    "# # plt.xlim((-0.2, 0.2))\n",
    "# sns.distplot(sgns.m[:, i], kde=False, norm_hist=True)\n",
    "# plt.plot(x,mlab.normpdf(x, mu, sigma))\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 254,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n"
     ]
    }
   ],
   "source": [
    "yey = []\n",
    "for i in range(500):\n",
    "    l = len(svd.m[:,i][svd.m[:,i] >= 0])\n",
    "#     r = len(svd.m[:,i][svd.m[:,i] <= 0])\n",
    "    if not l or not r:\n",
    "        print(i)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8.771826355536499\n",
      "173733\n",
      "173010\n",
      "['vello' 'kohtla' 'urmas' 'selver' 'piret' 'mihkel' 'siim' 'raivo' 'luik'\n",
      " 'juhan' 'virumaa' 'viiding' 'kadarik' 'eesti' 'hikool' 'viru' 'kuusk'\n",
      " 'olev' 'vana' 'kalju']\n",
      "[1.         0.83274051 0.87709534 0.78673334 0.8320892  0.86488594\n",
      " 0.82198131 0.86273587 0.78310403 0.86003788 0.79830142 0.86360648\n",
      " 0.85218797 0.85455743 0.86086675 0.82874892 0.82852215 0.86188243\n",
      " 0.72805513 0.82534474]\n",
      "0.8411738420547291\n",
      "['vello' 'priit' 'indrek' 'urmas' 'tiiu' 'tiit' 'madis' 'toomas' 'virve'\n",
      " 'kangro' 'mihkel' 'nurme' 'viiding' 'raivo' 'estonian' 'olev' 'hikool'\n",
      " 'juhan' 'ilmar' 'mnaasium']\n",
      "[1.         0.8871178  0.8806167  0.87709534 0.8770746  0.8741316\n",
      " 0.87219614 0.86911887 0.86901796 0.8661384  0.8648859  0.8647226\n",
      " 0.86360645 0.86273587 0.8625992  0.86188245 0.8608667  0.86003786\n",
      " 0.85936373 0.8591162 ]\n",
      "0.87461627\n"
     ]
    }
   ],
   "source": [
    "i = iprs_svd_argsort[499]\n",
    "i = 440\n",
    "print(iprs_svd[i])\n",
    "\n",
    "print(len(svd.m[:,i][svd.m[:,i] >= 0]))\n",
    "print(len(svd.m[:,i][svd.m[:,i] <= 0]))\n",
    "x = np.argsort(np.absolute(svd.m[:,i]))[::-1]\n",
    "# x = np.argsort(svd.m[:,i])[::-1]\n",
    "# print(svd.m[:,i][x])\n",
    "\n",
    "locs = np.array(list(map(lambda s: svd.iw[s], x[:20])))\n",
    "print(locs)\n",
    "sims = np.array(list(map(lambda sim: svd.similarity(locs[0], sim), locs))) #between locs and locs[0]\n",
    "print(sims)\n",
    "print(np.mean(sims))\n",
    "\n",
    "# svd nns of locs[0]\n",
    "nns = np.array(svd.closest(locs[0], 20))[:, 1]\n",
    "print(nns)\n",
    "nn_sims = np.array(svd.closest(locs[0], 20))[:, 0].astype(np.float32)\n",
    "print(nn_sims)\n",
    "print(np.mean(nn_sims))\n",
    "\n",
    "# sgns nns of locs[0]\n",
    "# nns = np.array(sgns100.closest(locs[0], 20))[:, 1]\n",
    "# print(nns)\n",
    "# nn_sims = np.array(sgns100.closest(locs[0], 20))[:, 0].astype(np.float32)\n",
    "# print(nn_sims)\n",
    "# print(np.mean(nn_sims))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [],
   "source": [
    "# sns.heatmap(svd.m[:, 42][np.newaxis], cmap=\"RdBu_r\", center=0, xticklabels=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-67.58080105754351"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.dot(svd.m[:, 0],svd.m[:, 1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 289,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['airport' 'airports' 'iata' 'destinations' 'flights' 'dafif' 'airline'\n",
      " 'stopover' 'airlines' 'runway' 'icao' 'sheremetyevo' 'airways'\n",
      " 'fiumicino' 'malpensa' 'rwy' 'hub' 'airstrip' 'hubs' 'vueling']\n",
      "['airport' 'airports' 'airstrip' 'iata' 'sheremetyevo' 'malpensa'\n",
      " 'flights' 'dafif' 'heliport' 'runway' 'stopover' 'airline' 'runways'\n",
      " 'taxiway' 'linate' 'aerodrome' 'faa' 'ourairports' 'stansted' 'airpark']\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f69962435f8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f6994b81ba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "SVD_NEG\n",
      "[ 53 163]\n",
      "['located' 'near' 'connecting' 'situated' 'connects' 'town' 'towns'\n",
      " 'branch' 'junction' 'geography']\n",
      "['stakecenter' 'mescal' 'centralists' 'cananea' 'quemado' 'potosi'\n",
      " 'garita' 'estancia' 'tecolote' 'manganese']\n",
      "\n",
      "SVD_POS\n",
      "[337 123]\n",
      "['vueling' 'tuifly' 'eurowings' 'tunisair' 'fiumicino' 'interjet'\n",
      " 'transavia' 'wizz' 'easyjet' 'volotea']\n",
      "['tolmachevo' 'syktyvkar' 'yemelyanovo' 'tsentralny' 'bolshoye'\n",
      " 'domodedovo' 'salekhard' 'cccp' 'yuzhny' 'yakutsk']\n",
      "\n",
      "SVD_ALL\n",
      "[ 53 337 163 178 123 369  60  67]\n",
      "['located' 'near' 'connecting' 'situated' 'connects' 'bandeira' 'trindade'\n",
      " 'penha' 'concei' 'velho']\n",
      "['vueling' 'tuifly' 'eurowings' 'tunisair' 'fiumicino' 'interjet'\n",
      " 'transavia' 'wizz' 'easyjet' 'volotea']\n",
      "['stakecenter' 'ever' 'mescal' 'centralists' 'cananea' 'quemado' 'potosi'\n",
      " 'garita' 'estancia' 'tecolote']\n",
      "['emburgh' 'kempers' 'kronemann' 'oosting' 'salumaa' 'tramacchi' 'nijssen'\n",
      " 'noteboom' 'orsanic' 'kratzmann']\n",
      "['yamanashi' 'gun' 'tolmachevo' 'syktyvkar' 'yemelyanovo' 'hyogo'\n",
      " 'ibaraki' 'iwate' 'tottori' 'tsentralny']\n",
      "['formatpropertyvalues' 'nicandro' 'preddy' 'hoggatt' 'ceed'\n",
      " 'andersonville' 'evh' 'brades' 'kepley' 'karon']\n",
      "['eretz' 'shalom' 'eliezer' 'avraham' 'moshe' 'yehoshua' 'akiva' 'yisrael'\n",
      " 'meir' 'eliyahu']\n",
      "['brightening' 'brightness' 'intensities' 'speckle' 'blueshift'\n",
      " 'redshifted' 'jsource' 'flicker' 'fluctuation' 'modular']\n",
      "\n",
      "SGNS_ALL\n",
      "[447 229 279  84 188 347  27  70 472 451]\n",
      "['garabed' 'hagopian' 'activist' 'abrahamyan' 'marash' 'voices' 'papazian'\n",
      " 'vardapet' 'erden' 'documentaries']\n",
      "['itch' 'negros' 'sulu' 'sulawesi' 'fukuyama' 'ozu' 'dimetrodon'\n",
      " 'occidental' 'paralysis' 'paths']\n",
      "['bridgewater' 'vieux' 'kiawah' 'cruise' 'waterboro' 'marianna' 'flacq'\n",
      " 'almina' 'aroostook' 'medora']\n",
      "['airlifted' 'railhead' 'huts' 'oroya' 'grows' 'helipad' 'onchocerciasis'\n",
      " 'airstrips' 'berbera' 'remoteness']\n",
      "['akcija' 'protiv' 'kroz' 'galija' 'prvi' 'kolektiv' 'enje' 'elektri'\n",
      " 'ritam' 'beogradski']\n",
      "['tigua' 'overidentification' 'wallenpaupack' 'rshs' 'districthas'\n",
      " 'highspire' 'kwantlen' 'northeastern' 'pateros' 'zyberk']\n",
      "['tanzania' 'regencies' 'list' 'rebun' 'murayama' 'tombo' 'asiatica'\n",
      " 'bitou' 'dive' 'adelidae']\n",
      "['rundown' 'faithfully' 'sanctification' 'apostolate' 'minden'\n",
      " 'philadelphians' 'endeavor' 'bonhoeffer' 'choristers' 'instruction']\n",
      "['donetsk' 'pacemaker' 'samp' 'alife' 'amsl' 'prati' 'penance' 'flyover'\n",
      " 'dodder' 'luhansk']\n",
      "['overviewbody' 'intermountain' 'afile' 'hqlocmap' 'jr' 'amex' 'murphey'\n",
      " 'peopleevents' 'shriners' 'freeman']\n"
     ]
    }
   ],
   "source": [
    "w = 'airport'\n",
    "\n",
    "nns = np.array(svd.closest(w, 20))\n",
    "print(nns[:,1])\n",
    "nns = np.array(sgns.closest(w, 20))\n",
    "print(nns[:,1])\n",
    "\n",
    "# words = np.array(svd.closest(w, 4))[:,1]\n",
    "# presidents = ['president','commissioner','minister','superintendent','chairman']\n",
    "words = ['baseball','pitchers','yankees','sox']\n",
    "words_svd = np.array(list(map(lambda w: svd.m[svd.wi[w]], words)))\n",
    "words_sgns = np.array(list(map(lambda w: sgns.m[sgns.wi[w]], words)))\n",
    "\n",
    "ax = sns.heatmap(words_svd, cmap='RdBu_r', center=0, xticklabels=0)\n",
    "# ax = sns.heatmap(words_svd, cmap=sns.diverging_palette(0, 255, sep=32, n=256), center=0, xticklabels=0)\n",
    "plt.subplots_adjust(left=0, right=1, top=0.86, bottom=0.23)\n",
    "ax.set_yticklabels(words,rotation='horizontal',fontsize=12)\n",
    "plt.savefig('heatmap_svd.png', format='png',dpi=300)\n",
    "plt.show()\n",
    "ax = sns.heatmap(words_sgns, cmap='RdBu_r', center=0, xticklabels=0)\n",
    "# ax = sns.heatmap(words_sgns, cmap=sns.diverging_palette(0, 255, sep=32, n=256), center=0, xticklabels=0)\n",
    "plt.subplots_adjust(left=0, right=1, top=0.86, bottom=0.23)\n",
    "ax.set_yticklabels(words,rotation='horizontal',fontsize=12)\n",
    "plt.savefig('heatmap_sgns.png', format='png',dpi=300)\n",
    "plt.show()\n",
    "\n",
    "word = svd.m[svd.wi[w]]\n",
    "neg_col_idx = np.argsort(word)[:2]\n",
    "pos_col_idx = np.argsort(word)[::-1][:2]\n",
    "all_col_idx = np.argsort(np.absolute(word))[::-1][:8]\n",
    "\n",
    "\n",
    "print()\n",
    "print(\"SVD_NEG\")\n",
    "print(neg_col_idx)\n",
    "for col in neg_col_idx:\n",
    "    x = np.argsort(svd.m[:,col])\n",
    "    locs = np.array(list(map(lambda s: svd.iw[s], x[:10])))\n",
    "    print(locs)\n",
    "print()\n",
    "print(\"SVD_POS\")\n",
    "print(pos_col_idx)\n",
    "for col in pos_col_idx:\n",
    "    x = np.argsort(svd.m[:,col])[::-1]\n",
    "    locs = np.array(list(map(lambda s: svd.iw[s], x[:10])))\n",
    "    print(locs)\n",
    "print()\n",
    "print(\"SVD_ALL\")\n",
    "print(all_col_idx)\n",
    "for col in all_col_idx:\n",
    "    x = np.argsort(np.absolute(svd.m[:,col]))[::-1]\n",
    "    locs = np.array(list(map(lambda s: svd.iw[s], x[:10])))\n",
    "    print(locs)\n",
    "    \n",
    "word = sgns.m[sgns.wi[w]]\n",
    "neg_col_idx = np.argsort(word)[:2]\n",
    "pos_col_idx = np.argsort(word)[::-1][:2]\n",
    "all_col_idx = np.argsort(np.absolute(word))[::-1][:10]\n",
    "\n",
    "print()\n",
    "# print(\"SGNS_NEG\")\n",
    "# for col in neg_col_idx:\n",
    "#     x = np.argsort(sgns.m[:,col])\n",
    "#     locs = np.array(list(map(lambda s: sgns.iw[s], x[:10])))\n",
    "#     print(locs)\n",
    "# print()\n",
    "# print(\"SGNS_POS\")\n",
    "# for col in pos_col_idx:\n",
    "#     x = np.argsort(sgns.m[:,col])[::-1]\n",
    "#     locs = np.array(list(map(lambda s: sgns.iw[s], x[:10])))\n",
    "#     print(locs)\n",
    "# print()\n",
    "print(\"SGNS_ALL\")\n",
    "print(all_col_idx)\n",
    "for col in all_col_idx:\n",
    "    x = np.argsort(np.absolute(sgns.m[:,col]))[::-1]\n",
    "    locs = np.array(list(map(lambda s: sgns.iw[s], x[:10])))\n",
    "    print(locs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "import csv\n",
    "\n",
    "with open('analysis.csv', 'w', newline='') as csvfile:\n",
    "    csvwriter = csv.writer(csvfile, delimiter='\\t')\n",
    "    csvwriter.writerow(['i', 'eig', 'ipr', 'cols', 'sim', 'rows', 'nn'])\n",
    "#     spamwriter.writerow(['Spam'] * 5 + ['Baked Beans'])\n",
    "#     spamwriter.writerow(['Spam', 'Lovely Spam', 'Wonderful Spam'])\n",
    "    \n",
    "    for i in range(len(svd.s)):\n",
    "        eig = svd.s[i]\n",
    "        ipr = iprs_svd[i]\n",
    "        \n",
    "        x = np.argsort(np.absolute(svd.m[:,i]))[::-1]\n",
    "\n",
    "        locs = np.array(list(map(lambda s: svd.iw[s], x[:20])))\n",
    "        sims = np.array(list(map(lambda sim: svd.similarity(locs[0], sim), locs))) #between locs and locs[0]\n",
    "\n",
    "        # svd nns of locs[0]\n",
    "        nns = np.array(svd.closest(locs[0], 20))[:, 1]\n",
    "        nn_sims = np.array(svd.closest(locs[0], 20))[:, 0].astype(np.float32)\n",
    "        csvwriter.writerow([i, eig, ipr, locs, np.mean(sims), nns, np.mean(nn_sims)])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preliminary Analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# s = la.svd(W, compute_uv=False, full_matrices=True, check_finite=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# x = da.from_array(W, chunks=(1000, 1000))\n",
    "# xt = da.from_array(W.T, chunks=(1000, 1000))\n",
    "# X = da.matmul(x, xt)\n",
    "# # x = da.from_array(X, chunks=(1000, 1000))\n",
    "# u,s,v = da.linalg.svd(X, name=None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "plt.hist(s, 50, normed=True)\n",
    "# plt.xlim((0, 0.3))\n",
    "0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def marcenkopasturpdf(x, c):\n",
    "    # Marchenko Pastur Density Function for c > 1\n",
    "#     ub = (1 + np.sqrt(c))**2\n",
    "#     lb = (1 - np.sqrt(c))**2 \n",
    "    ub = 1 + 1/c + 2 * np.sqrt(1/c)\n",
    "    lb = 1 + 1/c - 2 * np.sqrt(1/c)\n",
    "    mp = np.zeros(len(x))\n",
    "    \n",
    "    # Figure out indices where mp is to be calculated\n",
    "    lbidx = np.where(x > lb)\n",
    "    ubidx = np.where(x < ub)\n",
    "    a = lbidx[0][0]\n",
    "    b = ubidx[-1][-1]\n",
    "    xh = x[a:b+1]\n",
    "    # MP distribution\n",
    "    mp[a:b+1] = c* np.sqrt((xh - lb)*(ub - xh))/(2*math.pi*xh)              \n",
    "#     mp[a:b+1] = np.sqrt((xh - lb)*(ub - xh))/(2*math.pi*c*xh)              \n",
    "    return lb, ub, mp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "l = np.arange(0, 5, step=0.01)\n",
    "# print(l)\n",
    "c = 461408206 / 344332\n",
    "print(c)\n",
    "lb, ub, mp = marcenkopasturpdf(l, c)\n",
    "print(lb,ub)\n",
    "print(np.mean(mp))\n",
    "print(np.min(mp), np.max(mp))\n",
    "# print(mp)\n",
    "# sns.distplot(l, kde=True, norm_hist=True)\n",
    "plt.plot(l, mp, linewidth=1, color='red')\n",
    "# plt.hist(mp, linewidth=1, color='red',normed=True)\n",
    "# sns.distplot(sgns001_s)\n",
    "# plt.hist(svd.s/1000, 100, normed=True)\n",
    "# plt.hist(s, 50, normed=True)\n",
    "# plt.xlim((0, 0.3))\n",
    "# plt.hist(mp, 'auto', normed=True)\n",
    "# plt.hist(mp, 'auto', normed=True, range=(0,ub))\n",
    "sns.distplot(svd.s/c, kde=False, norm_hist=True)\n",
    "plt.ylabel = 'MP'\n",
    "plt.xlabel = 'Eigenvalue'\n",
    "plt.show()\n",
    "# sns.distplot(mp)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "s = [0, 1, 2, 3, 4]\n",
    "# s = [0.01,0.1, 1, 10, 100] 10^i\n",
    "wss = [0,0,0,2,34]#[::-1]\n",
    "wsr = [0,0,0,1,37]#[::-1]\n",
    "men = [0,0,0,22,968]#[::-1]\n",
    "mt = [1,1,1,4,73]#[::-1]\n",
    "rw = [410,410,524,1117,1744]#[::-1]\n",
    "sl = [1,1,1,6,269]#[::-1]\n",
    "glr = [0,0,58,2792,11488]#[::-1]\n",
    "msr = [1386,1386,1620,2598,4598]#[::-1]\n",
    "\n",
    "sns.set_style(\"whitegrid\")\n",
    "\n",
    "plt.plot(s, rw, color='red', label='rw')\n",
    "plt.plot(s, men, color='blue',label='men')\n",
    "plt.plot(s, sl, label='sl')\n",
    "plt.plot(s, mt, label='mt')\n",
    "plt.plot(s, wss, label='wss')\n",
    "plt.plot(s, wsr, label='wsr')\n",
    "plt.xticks(s)\n",
    "plt.xlabel('Negative Exponents of 10')\n",
    "plt.ylabel('UNK tokens')\n",
    "leg = plt.legend(loc='best', ncol=1, mode=\"expand\", shadow=True, fancybox=True)\n",
    "\n",
    "plt.show()\n",
    "\n",
    "plt.plot(s, glr, label='glr')\n",
    "plt.plot(s, msr, label='msr')\n",
    "plt.xticks(s)\n",
    "plt.xlabel('Negative Exponents of 10')\n",
    "plt.ylabel('UNK tokens')\n",
    "leg = plt.legend(loc='best', ncol=1, mode=\"expand\", shadow=True, fancybox=True)\n",
    "\n",
    "plt.show()\n",
    "\n"
   ]
  },
  {
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}