Cichocki_Synthetic_Low_Noise.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Details\n",
    "\n",
    "The notebook contains the synthetic data experiment with high noise (sigma^2 = 0.5) and low noise (sigma^2 = 00.5) for HNTF (Cichocki et al.). Due to the non-convexity of the HNCPD task, results will vary run to run, so the results from this notebook will not exactly match those in the paper."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using numpy backend.\n"
     ]
    }
   ],
   "source": [
    "# loading packages and functions\n",
    "import sys\n",
    "import itertools\n",
    "\n",
    "sys.path.append(\"./src\")\n",
    "import torch\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "from NNCPD import outer_product, outer_product_np\n",
    "from utils import *\n",
    "#\n",
    "import torch.nn as nn\n",
    "from torch.autograd import Variable\n",
    "from writer import Writer\n",
    "\n",
    "from sklearn.decomposition import NMF\n",
    "\n",
    "import tensorly as tl\n",
    "from tensorly import unfold as tl_unfold\n",
    "from tensorly.decomposition import parafac, non_negative_parafac"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def cichocki(X, r, to_transpose=(0,1,2), seed=None):\n",
    "    \"\"\"\n",
    "    Given a 3-mode tensor X and a list of rank r, computes to Chichocki \n",
    "    \"\"\"\n",
    "    np.random.seed(1)\n",
    "    #factors = non_negative_parafac(X, r[0], n_iter_max=2000, tol=0)[1]\n",
    "\n",
    "    factors = non_negative_parafac(X, r[0])\n",
    "    A, D, S = factors[to_transpose[0]], factors[to_transpose[1]], factors[to_transpose[2]]\n",
    "    \n",
    "    factorizations = [[A, D, S]]\n",
    "    \n",
    "    for i in range(1, len(r)):\n",
    "        \n",
    "        #factors = non_negative_parafac(outer_product_np(np.identity(r[i-1]), D, S), r[i], n_iter_max=2000, tol=0)[1]\n",
    "        if type(seed) == type(None):\n",
    "            factors = non_negative_parafac(outer_product_np(np.identity(r[i-1]), D, S), r[i])\n",
    "        else:\n",
    "            factors = non_negative_parafac(outer_product_np(np.identity(r[i-1]), D, S), r[i], init=\"random\", random_state=seed)\n",
    "        A, D, S = factors[0], factors[1], factors[2]\n",
    "        factorizations.append([A,D,S])\n",
    "        \n",
    "    return factorizations  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def run_cichocki_factorization(X, ranks, S_trues, to_transpose=(0,1,2), X_orig=None, seed=None, verbose=True, display=True):#, S_true=None, S_func=S_func_1):\n",
    "    \n",
    "    factorizations = cichocki(X, ranks, to_transpose=to_transpose, seed=seed)\n",
    "    \n",
    "    X = np.transpose(X, to_transpose)\n",
    "    \n",
    "    if X_orig is not None:\n",
    "        X_orig = np.transpose(X_orig, to_transpose)\n",
    "    \n",
    "    recon_losses = []\n",
    "    recon_losses_orig = []\n",
    "    topic_losses = []\n",
    "    \n",
    "    for i in range(len(factorizations)):\n",
    "\n",
    "        A = factorizations[0][0]\n",
    "        for j in range(1, i+1):\n",
    "            A = A @ factorizations[j][0]\n",
    "\n",
    "        _, D, S = factorizations[i]\n",
    "\n",
    "        approx = outer_product_np(A,D,S)\n",
    "\n",
    "        relative_loss = np.linalg.norm(np.ndarray.flatten(X-approx), 2)  / np.linalg.norm(np.ndarray.flatten(X), 2)\n",
    "        \n",
    "        if X_orig is not None:\n",
    "            relative_loss_orig = np.linalg.norm(np.ndarray.flatten(X_orig-approx), 2)  / np.linalg.norm(np.ndarray.flatten(X_orig), 2)\n",
    "            recon_losses_orig.append(relative_loss_orig)\n",
    "        \n",
    "        if(verbose):\n",
    "            print(\"Layer\", i+1, \"Relative reconstruction loss:\", relative_loss)\n",
    "\n",
    "        recon_losses.append(relative_loss)\n",
    "\n",
    "        if(display):\n",
    "            display_tensor(approx, figsize=(8,4))\n",
    "      \n",
    "    Ss = [factorizations[1][0] @ factorizations[2][0], factorizations[1][0], factorizations[2][0]]\n",
    "    for i, S in enumerate(Ss):\n",
    "\n",
    "        best_value = measure_modeling((S.T / np.sum(S, axis=1)).T, S_trues[i])\n",
    "        topic_losses.append(best_value)\n",
    "        \n",
    "        best_value, best_i, best_j = find_fit((S.T / np.sum(S, axis=1)).T, S_trues[i])\n",
    "            \n",
    "    if(display):\n",
    "        display_topic_matrices(Ss, S_trues)\n",
    "\n",
    "    if X_orig is not None:\n",
    "        return recon_losses, recon_losses_orig, topic_losses\n",
    "    else:\n",
    "        return recon_losses, topic_losses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def handle_losses_cichocki(recon, topic):\n",
    "    \n",
    "    #average_recons = np.mean(recon, axis=1)\n",
    "    #average_topics = np.mean(topic, axis=1)\n",
    "\n",
    "    average_recons = np.mean(np.mean(recon.reshape((3, 10, 3, 2)), axis=3), axis=1)\n",
    "    average_topics = np.mean(np.mean(topic.reshape((3, 10, 3, 2)), axis=3), axis=1)\n",
    "    \n",
    "    for i in range(3):\n",
    "    \n",
    "        print(\"Topic Modeling Loss 7 to 2 (average, best, worst):\", average_topics[0][i])\n",
    "        print(\"Topic Modeling Loss 7 to 4 (average, best, worst):\", average_topics[1][i])\n",
    "        print(\"Topic Modeling Loss 4 to 2 (average, best, worst):\", average_topics[2][i])\n",
    "        print(\"\")\n",
    "        print(\"Relative Recon Loss Rank 7 (average, best, worst):\", average_recons[0][i])\n",
    "        print(\"Relative Recon Loss Rank 4 (average, best, worst):\", average_recons[1][i])\n",
    "        print(\"Relative Recon Loss Rank 2 (average, best, worst):\", average_recons[2][i])\n",
    "        print(\"\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate Data Tensor (Low Noise)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = get_synthetic_tensor()\n",
    "X_original = get_synthetic_tensor(noise=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualize Original Data Tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA2gAAAEkCAYAAABaExIDAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAAA5WklEQVR4nO3dy+ul2X7X8e/at9+1qrtPPMFJ0JmKIDiRxKOCI8WDKI4U0WmOAxFFBf8NnehU8EomToIZKISYiJEIipcoiKhx4OV0d9Xvsu97OWhR1P15V+2nuuqs6n6/hufJc93P+j5r9S/1+bbee0mSJEmSfvRmP+oLkCRJkiR9xQWaJEmSJA3CBZokSZIkDcIFmiRJkiQNwgWaJEmSJA3CBZokSZIkDWJxyf/xr/vOvP/Gn1i+r2v5v/zLxx+L21rLrQGoaUCDbfPP5/mYtIyFE8Jl5sPBRU453vs6ZjJ/3MGFTD0ZPWT6VacdctK56N5msN/pRCfMm+bwUh6Plx+THmODc53gXGG/9eF17U7riT/ch61DH9qv/ufvfrBzUV1r8Fq+l3pCQ4t2CxvnD1vYaeo4pnEAD4zedHxeaSMcEGsN1VG6DkIvA+yWnhf9NpOv//zG9fHBOjSAX/01qHnwk5/ylK1m9AkEVNvwVQ+v8/wR6hDBcUC1ZuJklUwoQ++ljhK8bxKuc+pznFDP14dchy5aoP3Gn1jWL//cT1yyy2S/5Rf/RNy2XB7itsMhj9rFIo/a+7/1ST7mdX4T5/v8i7Rwutkx73Nc5nPRpKnBx+u4ouufdsweJgKf/uJ/yvsc8u+GYMHRFhe9wv/nWo7wMMO9tWX+GPdtXpi266u83/Nzvg64t3Z3m4/5+iEfMxSetoCvHV3/w2M+1dX5/X7pv/7tfK638CHr0If2O//sD+I2Wvw0qCl9fv593t/murB8zsejiRHVw1O4jqrCjz1OxEJJ+fQf/vt8qqtV3Ibj+O4m77fe5P1gsoV1qKdFTJ48YK3Z5Mlio//gQ5NFmmxBTelP5+se1tj1Gs4F34HwvP7x5z+T93kL3+Q69CH9zj+Xa97skOvJ5tP87l29pkk01CiYf9E8arE+f8zPfv4/5OsgVKMen+K2dn2dj0mLH3gmqVdyg7Hfd1BHb2HuQvMhqEN437RYTHNLqKN4vAP8l4Fwrl/6H3837uL/i6MkSZIkDcIFmiRJkiQNwgWaJEmSJA3CBZokSZIkDWJawsIHMJ/nf9C42+XLvrvJ/xB6d4DQBQjEWEwM4unh30jvV3ldPN9NS6qiAJFl+AesVfwPX6f8o/7+DP+QG7T7u7iN/nF7+gesVfwPzmsP1zk7/55QgACmp01Nr9xBgssshxLUpy/ztlch1IP+kTAFCMwhwSH+o+SvOTr0G4TqUArEqOIgo+Xz+d+hU9gEXAeGXsA4SPWwKgcqVVUttvCP+hepEFGiEvxj89v8j80pCKQgAKlTgMUeakoK9HkJ/8g+hG/Q8aq4jtK9tZcv8n57eGE/C6Fcj/n6KQiAAqhiAMrk/EZ9rSjDAUI7bn6YxziFfXDYUt5GwUkpzKTv8ze8rSAI5BlqzQwKKQSqTQ1pi0FsGBAENY/mSlMTdum+4TdI8xcsDRQEQsEpm/Cbwn35FzRJkiRJGoQLNEmSJEkahAs0SZIkSRqECzRJkiRJGoQLNEmSJEkahAs0SZIkSRrEsDH7ZDbLsZSPzzmKl6L7O0TKn+ApzSECOkVOT47SP0LcNMXK0n6Q2ErXMk/p0BCv2lY59r4/PuX9PslRzv3LV3Fb3eR3gSL4U+R0u8qxuEVR9BC92o4QtQuR2T1G2FfVF/mZxKhq+G3aJkeBd/htYpsAiPv+tqM6dKRXLMQ8V+U6RFH61H6Doqip1kDZ5vuG2jZL10LvGIydvoO4dojFLqon1J7j6iqfLxyT2o7ESOwqrM1ToqiriluBQPR1ewzPmc41sRUARnfrR47qENWFPrEeHq7yMWluNqlGUUuSXa4L7eYm7wdjFdsOQZsKHFvhHrBmkIlR+nj9VGugxvbt+Vran3Ibpti2A45XVdAeAb5v+WiSJEmSpA/JBZokSZIkDcIFmiRJkiQNwgWaJEmSJA3CBZokSZIkDcIFmiRJkiQNYtiYfUr9JBTBf6Sc6onnSxHWVRzBn1C8dcG5MHIW0lCXa3heEG+dzkcRpLWACGWIea8jRMpDhGrtIXp5SnQ0RDn3DcRzQwQ3vuh0bxBNnONcK8fRrnNMLUb3PkEscXqO1CLgW67Bo4GODDXfUZuQ87/RaTEtbvpE0dcYRR03cUsPqKOpXlJLjL7Z5MuAeGtEdQ80aMGRYqzbNcRGP+d7qwNsg7YEDeo2thCA71ULLVCw3Qq1EIB6HluSTPzm6+tFNQ9bEgGaD3X4PB5u8ju7gLnSpE4OE+PmJ6P5C0Tmtxf35/93ms/R3OsAfVoARdhTbehw3ymCvz/nmP1aXN4a5auDht+Upvx5kyRJkiTpQ3KBJkmSJEmDcIEmSZIkSYNwgSZJkiRJg3CBJkmSJEmDGDjFEZLCYBtZrXKay4lCBCHNkFKGUjoRJQVRshRdIyU1UrIaZQXNjpDQFxK6OiQuNkr8glTCvoH0ntvzaWBVVUXno0Sw/fn9euVkOErv6RNTi6pTvFVOUMJEo5Bg11YQEQgpSHV3m68jJdRButu3HZU2GuPkFOrXbD8tKWx2oLqQ95vvKWky73e4zgeNNYqSR2nsg/78nI95DXWI0tNojKd9nuA6IPk1pqq+CSRitmt4YSGttYdjNki2649031C/YnqadWgEDcrQ7jaPfZp70dyF0hgp2fYA7/pic/6YWGsgjbVozgPJrzivwdRFqA0hkRHTEWls0VilBGpKMf/kRd7v81d5W6hRmJRLKdRQY7FGBf4FTZIkSZIG4QJNkiRJkgbhAk2SJEmSBuECTZIkSZIG4QJNkiRJkgbhAk2SJEmSBjFszP5ikaPJT6e8rjwe87bTKcd0ziBJnGKladtxdf58KX6/qjD3fg6x2CeI0icU601HPKZU+akR6ieIoocI/qIIe4g1pej+GFUL19j38KO2/E62FfROoFjWEH1bVRinGyN6J8aSY+T3hHN921ErDRrjVBtSTaH2G1AqsX51SEnGliTHfMLlLp/wcHV5ven0/lFdIBRhPYfxPyHyn+KtaVt9cp+3fZGjqPvTOm7DOGpoS1CvH8/v82Of5nPd55Ye/eH88aoqtySxDg2P6hrF80+pC2865hGOSbUtgrkL1ah2Ay09IOa9P0J0P7UJSsek2HsaW8uJy45jfl4NWnBQbe7UXiDuBPcG9as/PIUN+XD+BU2SJEmSBuECTZIkSZIG4QJNkiRJkgbhAk2SJEmSBuECTZIkSZIG4QJNkiRJkgZxUd7lv37+Tv32f/pHz2573uRI8AbZpYvF+Qjlf/mTf+OSS3tn3/vbP503NogThSVuC/HwHNeaj0eR+BRvfVxQewFoEwAx3OknxbhWijSdQTT8Eh4yRK/2LURmU4ztixBHvc0/Dt4btRAoeF4plrWq2t1NPiREhfcdvGBTUAT/KcSjm24dLZ9hPEIE/+H68gj+BmO/Joz9qqqiCH6IZcZjTpHeveJoeGrp0dc5bh7bdqSY93pDLH6qUVQraTxCPaFjYusUun74brbrMF+AuOx+hN+U2hWk3zu1AdEHRZH4U+vCHD7HNGejOdbyKV/M/jYcFMYAjg+YR+H3neYh0O6nQwuLFtpsdGgT0KjFEbT0aCvYbwdzPahD1JYgtgKitinU2mkN34H0e8MrYoWSJEmSpEG4QJMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQbhAkyRJkqRBXBSzfzq1Wm/Px38ulzl68njM68DDYZA1IkXpY7w9RKWGyObFBiLeIab6BHH5DZI/l9t80NOSbi5vyvvAvVEEP0TYY0z1DmJlAcbYhmuJsfFV1a5ydDfF/U+9/oLI6Vrle4uxstSuAKJ7MbI8/abwyn3b7e5yPVxs8m++2MBBw/Pm2pUPR3H/VNv2cG+pVlZVzfbUJuT8/47tJCgmGcT2G1XVn3MEP0X3ozQmKRIfahSNcayHgNoLtPu7vGOqKRDPjfHiFPf/EM6F7U/0oeC8BuYT8y3UDPhpKWb/uLp8PleV6y/G3lMdopYk0FIC203Q2Nrm60xjnKL0qf1Gu8ktgrhNwMR2H1Q30vu1yR9UbAUwpc0BfB4GWR1JkiRJklygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iAuitmv+ipq/5yrZY6XXJ9yLCVF8H9IFKU/O+YcTIqITZGtFGE9O0yLZKbo2Klx2qeL347CmGeMIKUofYpsncONLyAGFuJ7Y5wrXX+Kr6/pEfzt7jafD2L2+wYixkPUNsbzwrb+DPnuKcaanv23HI3/49XlcfNVeRzP4TUhy+dpv9/yCYoN4Bju8L/f5rGD0fB3EAFNUfoUYU9x83AtdX0dLmTi+Ekxz2+A7UWozQbU7VTTMbobYJuA9LwotlsfzPI5v1/722mtOUiHCHuKPMd5VJgj4neVUEsiaiFC84mp7YpS2x44HsG2PTeh5lVVX0P0PbXJgrlSrF9U1+BZ4Zwt7Udrj7xJkiRJkvQhuUCTJEmSpEG4QJMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQVycATqbnY8Tfd7kWMr5PMdcLhbTope/dhB1STHPFMs6C8nLGKUdovmrqlpOcuaYfYr1hxYCswO0EAjJq/0AcdMUhUwRsOscb91nEPW6z/vRtcRYf4hrrQW9CBAVTDG88CypnQEKzxmjaO8hsvyLL/N+NyGy3HjraL6H2gBFKsU8V1XN4TWKx6P2G0uINIaflmKxqe4t1xP2mxpFT/ULopc7xcO/fsz7UU0J9aut8rmw/tL1Q3R3g/pFtQ3HeapftA/E9seWBHQuDYHmLofc9aKWzxPPhzVq2jHjHPEWbgBi44vaB1F7AWjpUwuYKz3kGtVevsj7TYCR+E8wZ6NnAsfE/VJNoRYC9G2hOduEb5J/QZMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQbhAkyRJkqRBuECTJEmSpEFcFLM/n53q5d35aNDnbY7+nc1ydunxeHHS/3tB0fcUy4qR02H52ykSdGKU/mKbr5/aBJzmcC0Q55oi+DHKGeLy6w7iaLc5AppglP4GIm5beNAU1wpx+f3xKZ+KYniP8DJArDeG2KdYb4iw7hDBS3qKt56YgP5tl9p2VHEdSjhSmupJ3muxmVZHyXFKjYWa0SgSH6KQ+2ab93vO9YRe97aC1iOL89HxdB1To/Rx/FMrk23OOqfa1m7CvWGtzO0+8PqtQ0PrM2ix8TStNQd9BOe7r3+uN0/zL4rSpzkD7NcgLr/v9/l81ILjO5/m/VK9obh8qpVU8+BdKGrTROiY4Zngc6R5GXw/qL1A4l/QJEmSJGkQLtAkSZIkaRAu0CRJkiRpEC7QJEmSJGkQLtAkSZIkaRAu0CRJkiRpEBdl3J96i3H6x2Ne61HMPm37kChetUMU/Syn+8bITYrLPlxDdCldI0SJUqwsnQ+zasMmjFdN8fVV1bY51rSfIB6WovQp8vTuLm5LEfb9IUdAY7zqPZyL4qFhW1EMLEmR2R3GIfxutYS4/1nYb2Lc+rcBtcQgsz1ER4fa0CBmnGoeRemfIJGd6h5to/YiqSY2aIlBLTaonpB2nyPg+3NuL4JjfJviremBwI8KbVOm1tEKcflv0j//8vx1fPZJ3oe+LRB9He/NOjQEaucz39Pca1odohpLx6QWSHEeRWMHvuENIvgLYva5xU6+AapD6Ts+NUqf4uap1vTn3NIjzmvqDW0JQgsOfP4Ul0/1kJ5X4F/QJEmSJGkQLtAkSZIkaRAu0CRJkiRpEC7QJEmSJGkQLtAkSZIkaRAXpTjOP5/Xp3/j/uy20xKScSBhLPmpn/lBPh4k7czhXJTe84t/+a+93YUJff93fD9vhGS12kGi0csXcVv/4Rf5mJAk1CHRKKanAUxBmpr6QylPL86Pw6qqvoaUunQtc3hWExKeqqp6un5KtnoL/+bXvls/+RfP14dGYZSYInj+mig59biaFgO32Ob7/6W//FcnHfPr9lv/8R+ftN/L2/zu/fAhJx1erfI7ttvnz1QLUZQ/9g8pgXZaCmK7vcm7PTxOOx9oV1fnz0WpkJR4e5Ovn9Jk213+3QpqTX+R02vjr0PfAao1lFC324UN71aHfvU/f7d+15/+6fOHhp/8eJXfzeXT+Wui+dUJZnGY4AzXgUmtcC0FjzTN237xr3xz517f/94fyhv3kOBKaYaPOU0a5wWUrDghoRrrAqFxBwm7lJ7YaK4H8zlMa5yAanOslSFJssq/oEmSJEnSMFygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iAuypjss6rd/fk13XIN0ZmQypqiXmcQiU2efnwet60e3y1WV2/WMSY1//eAdrXKx6QI61//3XzMCdGxCK6xtiHKuao6RXdDTPVU7fp8PHdVxevsEG9dEN2N7QpCTDgWhLcUI/MpATqXhqqQeLwP9a6qavWQ43EpSvs0f/f7f9/2+/yw5vP8Pr96ylHu83l+Xttd/hQdD/lalimef573aRRhTWMVYrFTJH7VG2riDF7KxfltWCuprkFMdT9Cf4rXD/mQyxxvXZ9/mc+Xfh96/hCzj5HfqbZR+5O30GdVuxdhPvQMvwNtCrfYTvS9gvhxajsC26h+0fVT/cVjflNtoGUP1ChsbUNjjmoUzFEKxtakcQLXjzUKYu8pEh/j7SmCP7UQgOh7nAfS752+EU/52fsXNEmSJEkahAs0SZIkSRqECzRJkiRJGoQLNEmSJEkahAs0SZIkSRqECzRJkiRJGsRFMfvtVLV6CvGTkJx5WkyIgYXjbT/J68qbH1KubN6krwlExxZFBVNcO8Tz1w+/yNuur/M2kq6FYmohirZRvPUe4u1Bf5zWQqCFZ0LXWDOI9aV43nRvX8M4TJH5U2Ol0zXNd/lZHq7zjSy2eb/javxCtFzm8UivyumUN06N7qdj9n5+W19D7DLENSOIjsaHQtHRtN/h/G+AEdx0b3T9M8pHz5s4nhti8cN9U9x/f5FbONSr3IqlxZjtdxyHvWoexnl4Lb86K9QhmivFy6BEfDrc1Nun88HrMIfP5zdVv5v2zqIriI3HKH344SbMQ7C1CNWo+7u8H7QdmhqlT+05+i48L5g79ud8jW0O55rw5zD/giZJkiRJg3CBJkmSJEmDcIEmSZIkSYNwgSZJkiRJg3CBJkmSJEmDcIEmSZIkSYO4KG+4txwtvdjkmOTZEaKqQ6zsfJ/3uf4ib1v/WF5zrh4h5l1fC4x5nhLJXhxdWrcQh7rZ5v1AugeMjqUD0jNJMa9V1ZYQb0/tDCBytlKMLbUC2ORY2Zrl63hv8dYEhjhFTh+XoQ5BzP5pTnnTedPHbreDdw8e8s1NftfX61Xcdnef37/l/HwGfFvl4+F4hDo0Od6eWohQK5B0Pmo7AmKkdFW1ly/yjjuI4D5BZjw8r54yp+neKEr//jbvF78D7zZIW1W1cAia8xyhbqT9qKXHjBLSqXsKdF0gJ3jVG3XLoTY731RfvIqbcMwRiKJvV7nuUQQ/1i9qBZLA/KQ/5bh8jKmneSC5zu0AUr2nOQ/Ny/AbkfajMZo3SZIkSZI+JBdokiRJkjQIF2iSJEmSNAgXaJIkSZI0CBdokiRJkjQIF2iSJEmSNIiLYvZbz3H6fQYx3RAjOQ8RsRhhDa5e5yjXxQYigfW1mBQzWlX9CL9Nz9smB7ZD5G8PsdINIv0LImz7GmJlKUqfou8p+hp+g3p5f/5Uz9NiZUl8F/q7xy238EocV/mZUSuQHlKBT6ENCF1DFbf7uPtvkEU9iP0+xyTfQVz+E8Tlb3cU5Zx/m6tFfl6P6xChDLHRFDdPdaFBJD5GWF9BzDON1Zvz52t72AfGFtbfx+d8zBNlp8P4uL3J50v3Td+P+7t8PLj+lmK2p8SHvyWav8y30LojtPugKH2K4F8+53PNd7mAHVcQdw7/WZ/ujWrzN1W7gTHw+iHvR7WGWhJRXD6hsRDGT3/O8xqqo7WH+nv3SdzUKtfR/pBbcGBbgtRSCZ4xtmmheWzaD569f0GTJEmSpEG4QJMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQbhAkyRJkqRBXJTJ2du0qNTZ/vJYWYqwxpjXHcS8hnPpa0Tx9RBFXdCmoa1yVC1GtlMc7RxihNN+h4nx6LMcWd7p+ily+sX5uPyqqtps8/lehzhaiuCmKPB5vrc2C8/4HYdhO1Ut1uefG0Xpn/KlxouiWkO3scjJvnW4zgf9zf/oT8Rts1m+N9p2OFz+3+H+3e/56xfvM5Lv739f3nicNo4xSh/qSS3gxUtjpDiGOx8PxiNcI14/boN7O01oa0O18vEpbmur3N4h1tF3bffRq9rx/DFojFNNSUVl+QStZuCAHeL+tzd5P5pHzeEzDt0ysDZ/U2EkPn1zn/K7jjHvFJdPrbCg9Ujaq9G5FvBOdrh+alc08VnS3CbWAIjLx+8A1fp0/VCH/AuaJEmSJA3CBZokSZIkDcIFmiRJkiQNwgWaJEmSJA3CBZokSZIkDcIFmiRJkiQN4qKY/aqqComQGIsPUa9pv9kB4toh3XP7Mq85l0/fvpjXD43jpifGw0K8Ku3Xrq/yfjuIbL05H+uPMa8Q89xq2vX3T1/mbZ+/yod8cZfPF6K7O+zTIAa2w3OsZYqcfbec/T7P45zioWeQrh7joaGuHa7g3Qvx21Vcv5bLfJG7HZXryyPNO13IRw5bekAdald5HNMxMXo5X0l1iHNun5wf/xS/326v87kgpp7i7avDNea9qm5ze5QWnle/gZr9wy/zNqpR7ylmv7eq0+L8E1hsod0PdARI86v1d/Lvs4J5DUb6w+1TSxL6VDdqlzOh68LHjuYF/fk573edxzHOGeCY1AKJ6l4cJzC+sUZdwRin1hwwH6p9bgXUn+A5UwR/PCBc4wxaJ2xD+yNj9iVJkiRpfC7QJEmSJGkQLtAkSZIkaRAu0CRJkiRpEC7QJEmSJGkQLtAkSZIkaRAXxey3XjXfn4+EPC5z9Gfap6rqFCL4pyZALzaQHfvNTZUeB2XwTow1pijqOuVY8jbL//2hQ5xrm5/PGG50bxAvPPn6n9Z5P2pnsN7kbZ99cv5///zLvA9F0cL1x3hbitJ9C+2Ux3mMy683RE5P2GexyfdBrUWoFQBF6c/n8M7CjS8W57edTt/ggkjjA95Z3G8P0f1LiKmm9iId2lSk6H6oJxSlj5HSoeZVcSuNGGFfFVt6fHUx559zO8Jvcw/tQ+A6Yqx3uIaLhCE0pdZU5VYgDWL7p56L6hAdk1ogYQT8t/DPATgeIaYev5Ewr8F4fhhb2CYkXGd/mFhraIzDfu0RWggQmn+9PP+8qA5RbH9/znO2lurXq/x7fguHjCRJkiSNyQWaJEmSJA3CBZokSZIkDcIFmiRJkiQNwgWaJEmSJA3iohTH+dOuPvkn/+Xstv6QE5vaTU6r6buQHpPS5t5gatLLH/itvzdvXEGyzN1t3i+kX/V9TsXCpMCJCYmU0EMoZSgle/3sP//7k841kp/68z84vwHCq1L61pvMIOG052C1Oq7yu7BY52PG/dqPx33oGk8Lel/DNfwcJE29hfnTrj77pV+7fEdKnEtjnMYOJN/hWIVtL34e6tcJXkCqUeF8lMD3/eXvn3QuusZ+nZMO2zMkj1KiWah7P/svfi7v85H43p/56bP/OyWENgpIhW8E7be/zf8dl5JMDzd5v5QgSNdBKYC4X6xDkLz5FhaPu/rOL4Q6BN/4ug6pklXVQwJvg33qAB8fSjJOc6+qapA8ismdL+/zMdfbs//793/qD+bjEUgXxeRU+G3o3miul5L9fvZf/YN8HR+J3/bLf+zs/94hap22HQ75nVwu87s8o6Tied5v+Xe+k/cLcyWqeZQGv1hfnuxMdci/oEmSJEnSIFygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iAuitmvU8/RrFcQAwuRpy3EwPYvXk06XqcoakKxrMcc4dlfQ3uB9EwoHpaiXDc5ippicSmit1OENcSBt6t3iyge2Swl7VIrA4i+nhK9+iaQYosR/Ckith2nRelD8m21Q9g4cYi+jf4MMfX0Poffth9hfFAk84scN10hSruKW5LQ9VNkdjwc1GyMm4ZY7749H6VdlZ9xVU2K0q8qrtsfuT4Lvzc8Doqij3WtCqOjG3xTqTZQ3Usx1svnCXn5vKlmqQ69q175vaU5yuNT3Ja+4/31Y97n05dxG7ZWwAh+aAVE3/6HfG9YS9O55hP/hkBR+nQd9I3f5NpWE+7tY5Ei8/f73GbmeMjbGkwadhjdHzfVbJbH248/5t9md3f+/ZpDayGqJ7FmV9V8G/aDV8e/oEmSJEnSIFygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iAuzyieEGOPEdAhzhWjXOc5whPjoSdGMmN0NLUXCPH87SVEcFNca4NY3IkxrxhjS+0FIMb2Y5eiqlt+HOi0zNGrKea1qqpR/CpEvVL0/eHq/LUs1nmn+S5vO1xDvnVoIUCR2G+lV87cpXYTFH39vL74eDgeqf3G9XU+JqA6irVttTx/vMfckqCFfargWVXx81rC84cWIqRB64GP3Sn8BDMovTQeab8Z1LZFioeuqhO0CZnSnoNiqqnm7e7zWFzC6/puevxG0hihFhxxHEyNm9/C3Ava7xTUBsw7h7kZ7pfQ+IY2AbitU0sSuH4y9ff5COx25+v2/W2u2Q9P+Rt3c50L0eGYn+NqkYvUepu/V6mlRxXPsSJ4jY/5MvK56BV/qwuSJEmSJL13LtAkSZIkaRAu0CRJkiRpEC7QJEmSJGkQLtAkSZIkaRAu0CRJkiRpEJfnzkMM7iQT4kl7iOavqmoLiOCH/arnvE2M56Z46BARi0+QYsIp+npiewGMA57Y6uBjNyVmn+JaKYp69ZR3pP0atLvAeOuwX6e4bHhh6Vwpnp/aALydHG/dlpBzS9IYmVoX4FRT6xeilhiP4XxQy6mepNj+N9rkdiVF9QvaEvTDxN4XH4FUU2j8LKGeUGQ5tb6g6PsTzCDm9LmdhTpEXV9Wl9caPObX0O6jp+h4aClB29ri7vwGistfwxwEovSpFQjpB/ogQPuXNJ+AsT+5nQ+0QEGnXE/aIs+Herjv3/QLfzIfDwbyapWfyRYi5emYZD7PdePf/u6/PumYo/jez/wgbku1geoh1aEZfDan8C9okiRJkjQIF2iSJEmSNAgXaJIkSZI0CBdokiRJkjQIF2iSJEmSNAgXaJIkSZI0iMtj9kOcKMbbnyD6dwqIQq3K14ER1pDv248QtQ0R0DFOlyJzIQq5UqRvceQ3aXe3+XTUQuAbHLOf4u1ne4i2h8cxO0AE9MSoZ4q+psj/FC3djhDPC+/k8hnaBCzTyfKp3krvOX6Zag1EQKea0m5u8mU8PefjURQ9xUrvIOZ5BW0vZpRPfv6Z4PHwOeZtfZdrW4qirqqqqxwHTq1Hap/bhHzs5tvzzyuOq+K2F1QzFusJ47g43p7iqGONos/fxFYAi0042Tu3+6g8H6J5AXz/41yJ4ushSr/209pXUD2hMU61LUX+U5Q+1iiYD2FLCaptM2idQvcdzwVjZ5mvY7eDVgwQpT9x+viNRnObNNc73EyreTjXm9CizL+gSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIC7MZu9VPUSDLnLUa4MY1RivTLHRFFNLka1LiL6+yZGzbQfXT5GtId66H6BNwGabt1F090SNclnfw34fsx4iWauKo+Mp8hdec4qVpij9RnHm4VroeAdIY17AqxyP+R7jfhvVjSW0AgktJTq1xCAQN43x1oTi7bdQN1JLDIj97Ruoo3N4xlQXaPwc84vU6VlCG4SP3f72/HNePcJ7AM9/aisNqg1UEymOOh0TI/2pzQl8GiGV/N20N4yFhGpUmBu0F3dxl/7qdT4etKjA+RCNOfqewfn6c2hLQvvAXKnBfIjq4dSWRPgahW/ubA7tKyCCf+r06rDP37g5fKz3EOv/scO2Q8HqEX5tandF87K0H5zKv6BJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgLsvWbI2jWYMUYf3VFYRLoCjL6xzp3+9y7HJ/eMrH3OdYWYx5psjWEPlP8bAYpb+AmPBHuDf4zabeG+33sVtsz797FHtPkdKEYrF7/rlRiueuyhHdp8XXH7MdWwi8a4eGXrmFRUFLjPvbuCmNyf4UoqGL6xDBCGiI4O8Qs491OcR699ePeR+KsIbaXC2/exiL/ZTrV/34r8v7vXrI+33kZodQh6AuHK7z4Lp6oIGcN51gjKdrrHpDW5IJvTYogp9qM217Ny3PX6htxHqdDxnaDuHchVpN0FiFa4ytOeoN8zk6X0JzkNUq7wfX31L7pnpD/YX9cD4U2rHMIWb/eMy1ckGR+BClf32T28IsZhD5//4GyY/c8ery+kW1i+ohtfSIc0R49P4FTZIkSZIG4QJNkiRJkgbhAk2SJEmSBuECTZIkSZIG4QJNkiRJkgbhAk2SJEmSBnFZzD6gmHeKbE3x0H2X40J7iK+vqmpwGQWR5hj1SlHoFDkb9sNofoq+3cK2GcTi0r1RBP+Lu3y+z7/M2z5yKR6e4ua3LyDa/glaRpzyttmBou/zflcPeVuMiKWoV4jSxnj+aZ0H3mw2q3YboqVp/Dznsdp3oSXGHP4bFkRKU42isYp16C63Cag13FuqUXBvU2tUP8B9Q5sQjOdf51jsHsbBT/6FH8R9ZtQhBMbBcZU3Lp9zceizCa0oquqHf/h8i4fTKe90e5uf1TPEeh8O8I2GgTybTRvkyxAjvt3l944+33SNyeFXoFa8jX7KkfmffQInhnj4cJOd2hJMbHmTouGritv9AIrF75vz7+bUdiVFrUVojghR+lzb4DmH+nX/917GXZZPEMEPtYbmISco21j3wPf+5vlaSucidP1T743q6PqPvorbDqEmUnsEqpWkh1YGh1/J5/IvaJIkSZI0CBdokiRJkjQIF2iSJEmSNAgXaJIkSZI0CBdokiRJkjSIy3JY2qzq6nxKT4OkRkxWS0lokHTWIA2sU0ISpYjdXOdjvnqdjwmpRSm9ElMoKVXx9cPF56qqnHhXxdFYX8B9X+fn9dELr2QI4amqqtUjpP7c5P8OsoSExxMFacHvdoLrnO/Pn49SoyjabjYlCO1d0x1bxURATCylY6b97l7kfSAFjcYxjjnQH3JELda2BGpGSrV8I0h/jIl3VZjiSIm+6b7ncPmU+EVjfLHOYxzTTI/5hT9cX/7fSFer/I7vIAXxsxfnUyGrqh43Odlus87fuBlcywkK0W5//jppaCxC8mMVX+OnL8/f9/xdY2Zby99/mIdQemp8ACdIF4SU0DaDdNQwl6uqOM+rqiqoDTT/SrUB96F7g1rfj3msUvWl/Si1t47nr2W+pe/7tO8AJTjP99OSn6l+VdoPiiUNLU5xhMRYqOk0f0lJjVU5EZfSaT+5y9+xV095rn1/cz7FdD4zxVGSJEmShucCTZIkSZIG4QJNkiRJkgbhAk2SJEmSBuECTZIkSZIG4QJNkiRJkgZxWcx+71X7EG0K+bgUK5+iS9tNjqvsj49xG8W/U1xznSCW9fY2H/M5RxenVgFtBVH6EN2NUfp3cI0UBw7XgtHdEHH7sUvx9hQPS3HzFMFPkfOneR5Tc4junsG2lIw7312+TxVHPL83p159cz6ytlG7jKccj9s+++T8hlTv6g3x7xSzD3HNGPNMz5rGMURtR9QaBSK4+/Mm7wdR+lS/6FrSb0BR+rMDxN5fwTOGbxxF6VOcNo271er8fd9e5ffucZ3j8rch2r6qqlNkNsRAHyHCmlLJr67Oj6vtJr/H8znEUcO2h+fzz+QYIrbfGrQdqlCfqorHz9P5+QS2taHxQfMa6mmAdQie2yd53hYj1OEaqWbXEuZD9PLRfVOs/zXMscLvvdhCaw74vmMEP7XYgRn9HKaW+5t8zEUo6TTnoTkDXWOnGguv5ALaGdxd5xt//XR5uyiqo9RaJJ2L6pB/QZMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQbhAkyRJkqRBuECTJEmSpEFcFrNfxRGlaReKUN5BBH883l3c1iH+vcO52gIiiCEyd0r8LV0jxtdDBGzf5Hjrdg/Paw37wTOp5eWvzsdiFn6CI0TfNnockPhLlmuIH6akYIi/3b44/99klmuIfof/jEPx4lNqxVsLkc391cO046VWFBABndpoVBXfO0RAN4rgvskR6vXwlLel9hwUpU9tAg55P2zNAduoFQvGgYffgMYHRelPjY7GeG54hXBbO/+evHrMUebLZb6B3SEXqdUif3coSv+wz7/pi/sckZ6iqikunyKs6RrTXpTE/lao7RDNJ0KUfhV8c2msQmuRRvMriuen1kgpLr/ys66COQpFq1PbJLrGl/dwJeDzL/P5Pn2Z9wvtlo4reC8ntMOpKvyzSpq7VFWdFvmg9P0/pSEOr2RsqVBc88j+Nt843dsBasNsdv46Dwf4VsGPg61AwrlSna/yL2iSJEmSNAwXaJIkSZI0CBdokiRJkjQIF2iSJEmSNAgXaJIkSZI0CBdokiRJkjSIy7LSe68e4qjbfY7S7485VjbGE0OUc/UcZUnR8BQ3T5m7FDNOcfN9G+L53znf9//XXr7IG6m9wApivam9AMWPf+TS7718zve8C/H1VVXz3cQIe4rghldof5NjYK8ezh8UWwhQHPA875fieTFC+F2tIB6eoqpTS4wUv19V7TbHncexX1V1hPoFMfuN6miK0i9o60F1gY5H0d17eF7UkoTAtaTnRbHLJ3hFZk/0ruf9EIzVGYyt7eb8hd7f5dYo621+f5bLnMH9+JR/m/kC2n3ELVVP69wWIrUDWNC54GTX1/mdTPH8LcRev7XeeSwktE+Yv3RqzQHfYqxDDSLgqV0GfXygxsbxPzHuv83y9ff/8UXchi0xbqCmf/4q7xfmltTSY/mcn+N8T7H30G6JOpIc4HejoRAOSnOXDs+Y5kM0D5k6/6L2Iin6frXI7x0d73SiSH+Yf6R9Lt5DkiRJkvReuECTJEmSpEG4QJMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQVwWs181KRa/QfR1vzkfxUuR0p0S+CmOFuKaKS6/1jnWuD+t8/mW5++707O6zdHEpD9BBPdEFPndN/mZfOxmIY36RBGwTzkCdn8LccYQb3uCV3JqBH+Mvqf/VNPyfc8gDrjDfu8s1KF2nccPtY3oIfqeovQLYp4pGr5S7H3xOE715KuNEB2dWo/Q9UPMdqNY7BnUr4nPBNsjhIhxiqlebCDCGuL5SaoZX22bFueeYuCfN7kuk8MhP/9FiL2vqprPp7VUWSzyMWP0PTz+/T5ff4rtr6rqKXv8Xft9NIijp3cdWhLFCH6Ky6eWRNSigmLqqY3OCeZYUH/j3BHORb8QXWN7eZ93hDYn2MsBouPTM8FI+RV8V+EnbfD8aY4C3am49UA4HQ2f+XbaNVLdpuh+qun4k4Yau9nlb23ap2p6rYzn+lqPJkmSJEmazAWaJEmSJA3CBZokSZIkDcIFmiRJkiQNwgWaJEmSJA3CBZokSZIkDeKymP2Wo1n7+vK4+aqq+uLV+eNRlCtFX1MkM2X4QvRqu4NY3Od83323O3+86+uL96mqau8ztvwcir5u39y1fYqPbUeIzL2aFh1LMfUUR0v5w6c5XAvE/k6xu8/vwSq0HqA2AG9lNqt2c34MYdsLaBuR6kZqA1JVVQ9P+VwUYX93d/F1VL0hVpqyhKn+JnT9BOo2xWnjfovLr4Wi7SlKn9pXcAR03o/OR/v1EEW/vMrvyHKet233+XN/POYLORzytvk8P5PdLp+vhSKQ4verOEqfWgik5/jOWqtKLSxgPHZo6dNCzH5slVFVPUXzV3E0PKHxCMfsz9AmJMx7sO0QtSR4eMzbtnkehS0QAM1j0/yXvnXcoibv1+H73qgjCcXbw7yA4u0Tai1EsN0PzRsmzikWs/MFP7bmqKrdFsbvhAh+uvRv7ixbkiRJkj4yLtAkSZIkaRAu0CRJkiRpEC7QJEmSJGkQLtAkSZIkaRAu0CRJkiRpEJfF7FfLEaUQoUzR8Sn6vl3nSOy+3ebDURwtRK+2BUSvQtx8h205Vhbi6yneFuJVKZ6bnkmFuPKqqlpv8jFX5yNnf8df+lNxH4pyxThaiMw+riiPFjbBf5r44o+cjwr+5D5HuL9+ys9xBtGr16sckUyx2BQDS8fcnM7f+GabI4QXC8juBSmE/vjPph3vfzudqsO7mfR9fi6phUV7zLHRGF9PsfcUNw+tAFIEd1VxdHSoN5NbelDN+OLLvO3+07yNnjM9y3DfFA19gkc1g9J8hJhqql8UmX24hmOGurHd5PduN8s3d3WVb45i9ql+USx+ipWuyvXrdIQfB2L2r6+g5kFteyenXn0T5iI0R5nnZ91D2x5qEdKucyuQ/phbgRS0D0rf96qa3IKjh1j8dgXX//ohbuN2RVDr45bC+ht/66pqL+7P/u8nmHo1GDs454FjUkuP1WMej1NqG7UkWT7D2IfXh+ZzM5g2zKAFEllPqA3zRb43mivNwoOk99G/oEmSJEnSIFygSZIkSdIgXKBJkiRJ0iBcoEmSJEnSIFygSZIkSdIgXKBJkiRJ0iAujNnvGOeetBVElx7OH48iTYsiYClu+gi5oBQdC9saRUCnfSDetihKm1ArAGovAHHaHZ5X2mu+nxalT5H4J2gv0OB1bNCyoMMxUzz0q8ebuM/tdX7Gu0N+Jx+fc8QwWUDU6x6iqlMs9hIirOezfK7tLpcQOuY7aS23jqDaANH8PY1jaokBY3/yuIJa0yGeH6O2U3sBiGTv69xSolUeB0UtPai2dajNDa5zc/43pWh7aveB4Del6H6qiUvoLpCsIFJ+B7HRFKX/8i6PjS3Ur806f9uvrvPY2ezOX+dimfc5wHVQpP9qdf6YDT9Ibym1c6Do+GdoXxNi3mN9qqqisUp1IYydqhwbX1VVMDfDlkpUG9J1QJQ+jUdsEwBzWJx3Ynuk8+8fte2YiuY8C6g1NMeioZDqJUX60zaKxMc2J1fwe6+hLQHUvTRHuYY6RPVwDg/yeZPGNjyruEWSJEmS9EG5QJMkSZKkQbhAkyRJkqRBuECTJEmSpEG4QJMkSZKkQVwWq9Nz6iKl5lBaUEz2mUNyIiUuAkz2WeZH0SH9DZPJwvna4vJUsq/2y9eIaUeUvEnpUCmhqiqmZWKqIqY45o2n5bQUx8N1fk9O8CqkVJ27G0ivgiQeShi7hUS2x3VO4JrNpqUWpWvZbvO7dXUFqUtwHdvN+YdMz+rt9JiaVXtIM6UU11Ooa0t4USiVDMZcg9oW7+sN+sNjPl9KlKO6cAdJjZAKWw1SKJ+e8m73kBqH6Wnn37/DdX7HFhDURhoFTcK2DnWUrjONVUz9glpzBclkm31+l3eQ1Hp9k9+FJ0qonZCgSKmwdI0p/TEl2r612azaTfjuhnnS/7qgvC0lE1LtmpAkXcVzhv76Ie8HqdxpPFZBsi3UPEyupaTclFxbb6jpdP20X/h9MEl64utHtWZ2gOun+gXJiimRsVPwOaRJHlfwu0Ea7mJNN5CPSXUjzZVebfL3r8GcZ7HI50pJ3zSH8i9okiRJkjQIF2iSJEmSNAgXaJIkSZI0CBdokiRJkjQIF2iSJEmSNAgXaJIkSZI0iMti9lvLUfUUD03x8CkynyKsKZb1eZ33g5j9RrHY1zkuuFNUdYpYpZjqEF//1YVALitcR2qNUFXV7m/zNoosp3sIKAL2cDUtSv8Er8kcol7n+BOcf7+Op/zfM2jbi5vcOuGHX+Z48eUqP//9Pr8nDSKse8ixXa3yQ04x1VVVS4juTvGxdH3vCt/LEKVflSOUO0ViQ5RzzeD3oXYfNMapxkLd6NvzufLtBqL0N5BFD3HTMXa8qvorGMhUR+GZpNpMY79g0xHqEMU8z7BGQQT0cz5mioC+gtYcFMG/2cHvBmOSoqNPUPcojjqhxHg6F0VVX1+ff160z1tL7ybUDWz3kVoydIibp3FM9Qu2YZQ+oXlbAvOrRi2OqC5QJD7VrwbPC6T2SPtbaHmzhe80nIs+n1RrampXiXA+ms/1ObU5gf3gT0ZHaLdE10LtNFK9pPYhVIfoXNvQygQ6MfgXNEmSJEkahQs0SZIkSRqECzRJkiRJGoQLNEmSJEkahAs0SZIkSRqECzRJkiRJGsSFMftVtTgf59w3EE/6MkeJ1zrEOVPeLmxrcK4O0dEdIrMbRFi3qxwRG6OjKYqWYsJhOU1R+hTRS20JYguEqmrz89tCintVVR0hFpdiqsl8R5Gt03JlU7w1RahSZPPjJr8j80X+bVLc/5vOt5jl/XYQmZ9MjfRP3l/IflUL9amqqhYwVkPkdIOx2qkOUVw+obpHrUCopoQY7r6GliRU8yhC/NXrvN9nn+TzQa0paPeRfjeKsD6uJsY1Q8wztQKh6OgTZF/3UG82mxwTTu0ydjsYG3SRgGL9Z1C/8gHz8VbL/I1ery+Phadrf7sDnHI7CmotVJdHufeWf58U8V5V1e5yGx2ca0AdwtYj1O4jtBdoO2hXAhH8dP14jXTf9LvBtyXNA2fQ7oNqDaEIe5p/zeCRnKi704TLpKFFLZUogh/bC0Bt3u/yb3oT4vQ321xjae5Fc7ZYf6mGxi2SJEmSpA/KBZokSZIkDcIFmiRJkiQNwgWaJEmSJA3CBZokSZIkDcIFmiRJkiQN4rKY/WpVIT6aItlrA7Gmq/Nxlv2LV3kfipR+zpGztYfrWOaY3r7N8fwYR52itikuO8TXV1X1E8QWQ5R+W+bIUIzT3cHzCjD2HqJX5xBHe1xB3D9FwMK2wzVEvS4uz/w/QHz99SrHCB8pPhkeGMXbU5R+bCEQ/veqqtUq5/Mu4Vmtwrb5lNze/xfF0addoM1Gnc5fK42dqeOKUK3Blh4EWohEUKP6Ce7t5jpvgzhtahPSrqA2h3ujKOrlGtqOQHsESqLHlHo4Jg2FFuKcaewv51C7IIl+S9H9V/l3o1jpVGuqcuuOdM9VVY+P+d1aLPN9p3o+pUXI/3OE2B6iQ5Q7vc807uLxoA51mA+1MPd603Vg6xGIt0/XiTWPngfNvV5Qa6f8TKjdUmp7UVVVYa5EsfEUbb+/zWOHjkmx/qcFzHk21K4o/O9QY2keSJH4NC9rMP2lNidLmL+kOP3b6zx+aX5FNeX6+vy7Rfv4FzRJkiRJGoQLNEmSJEkahAs0SZIkSRqECzRJkiRJGoQLNEmSJEkahAs0SZIkSRpE6xfEQbfW/ntV/cf3dzmSvgV+Q+/9u1N3tg5J+hpYhyT9qMU6dNECTZIkSZL0/vj/4ihJkiRJg3CBJkmSJEmDcIEmSZIkSYNwgSZJkiRJg3CBJkmSJEmDcIEmSZIkSYNwgSZJkiRJg3CBJkmSJEmDcIEmSZIkSYP4n/MNdXWGa/KqAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 864x360 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "display_tensor(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "S_trues = get_synthetic_gt_topics()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "num_trials = 10\n",
    "transposes = [(0,1,2), (0,2,1), (1,0,2), (1,2,0), (2,0,1), (2,1,0)]\n",
    "ranks = [7,4,2]\n",
    "\n",
    "all_recon_losses = np.empty((len(ranks), num_trials, len(transposes)))\n",
    "all_recon_losses_orig = np.empty((len(ranks), num_trials, len(transposes)))\n",
    "all_topic_losses = np.empty((len(ranks), num_trials, len(transposes)))\n",
    "for i in range(num_trials):\n",
    "    for j, t in enumerate(transposes):\n",
    "        recon_losses, recon_losses_orig, topic_losses = run_cichocki_factorization(X, ranks, S_trues, to_transpose = t, X_orig = X_original, seed=i, verbose=False, display=False)\n",
    "        all_recon_losses[:,i,j] = np.asarray(recon_losses)\n",
    "        all_recon_losses_orig[:,i,j] = np.asarray(recon_losses_orig)\n",
    "        all_topic_losses[:,i,j] = np.asarray(topic_losses)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Topic Modeling Loss 7 to 2 (average, best, worst): 0.16321765484092482\n",
      "Topic Modeling Loss 7 to 4 (average, best, worst): 0.23602288098285878\n",
      "Topic Modeling Loss 4 to 2 (average, best, worst): 0.18202732336448119\n",
      "\n",
      "Relative Recon Loss Rank 7 (average, best, worst): 0.11878977133610587\n",
      "Relative Recon Loss Rank 4 (average, best, worst): 0.5017667231424502\n",
      "Relative Recon Loss Rank 2 (average, best, worst): 0.7945462173333894\n",
      "\n",
      "Topic Modeling Loss 7 to 2 (average, best, worst): 0.08670270370955138\n",
      "Topic Modeling Loss 7 to 4 (average, best, worst): 0.04011086039053309\n",
      "Topic Modeling Loss 4 to 2 (average, best, worst): 0.10137922485135571\n",
      "\n",
      "Relative Recon Loss Rank 7 (average, best, worst): 0.11878977133610494\n",
      "Relative Recon Loss Rank 4 (average, best, worst): 0.3091643238063878\n",
      "Relative Recon Loss Rank 2 (average, best, worst): 0.665057914875574\n",
      "\n",
      "Topic Modeling Loss 7 to 2 (average, best, worst): 0.07723971056139248\n",
      "Topic Modeling Loss 7 to 4 (average, best, worst): 0.1217103865945099\n",
      "Topic Modeling Loss 4 to 2 (average, best, worst): 0.10608507253394639\n",
      "\n",
      "Relative Recon Loss Rank 7 (average, best, worst): 0.11878977133610524\n",
      "Relative Recon Loss Rank 4 (average, best, worst): 0.4173701515408023\n",
      "Relative Recon Loss Rank 2 (average, best, worst): 0.7126782056893392\n",
      "\n"
     ]
    }
   ],
   "source": [
    "handle_losses_cichocki(all_recon_losses_orig, all_topic_losses)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer 1 Relative reconstruction loss: 0.09631970490841536\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer 2 Relative reconstruction loss: 0.4513456711370244\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer 3 Relative reconstruction loss: 0.776147641586295\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x360 with 6 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "ranks = [7,4,2]\n",
    "a, b = run_cichocki_factorization(X, ranks, S_trues, seed=0)"
   ]
  }
 ],
 "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"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}