Comparison RNN simple LSTM.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def diffusive(scale,ndim,time,epsilon=1e-7):\n",
    "        \n",
    "    mus1 = np.zeros((ndim))\n",
    "    rho1 = np.random.random(1+(ndim-2)*2) + epsilon\n",
    "    sigmas1 = scale*np.random.random(1) + epsilon\n",
    "\n",
    "\n",
    "    cov = sigmas1 * np.eye(ndim) + epsilon\n",
    "    \n",
    "    return np.cumsum(numpy.random.multivariate_normal(mus1,cov,time),axis=0)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from Tools import generate_traj,random_rot,traj_to_dist\n",
    "from scitool.propertie import Propertie\n",
    "\n",
    "def add_miss_tracking(traj,N,f=10):\n",
    "    \n",
    "    step = traj[1:]-traj[:-1]\n",
    "    \n",
    "    std = np.average(np.sum(step**2,axis=1)**0.5)\n",
    "    \n",
    "    for i in range(N):\n",
    "        w = np.random.randint(0,len(traj))\n",
    "        traj[w] = np.random.normal(traj[w],f*std)\n",
    "    \n",
    "    return traj\n",
    "\n",
    "\n",
    "def generate_N_nstep(N,nstep,ndim=2,sub=False):\n",
    "    add = 0\n",
    "    if ndim == 3:\n",
    "        add = 1\n",
    "    size = nstep\n",
    "    \n",
    "    X_train = np.zeros((N,size,(5+add)))\n",
    "    Y_trains = np.zeros((N,size,10))\n",
    "    Y_train_cat = np.zeros((N,1,27))\n",
    "    Y_train_traj = []\n",
    "\n",
    "    #12\n",
    "    for i in range(N):\n",
    "    #for i in range(1000):\n",
    "\n",
    "        #if i % 1000 == 0:\n",
    "        #    print i\n",
    "        sigma = max(np.random.normal(0.5,1),0.05)\n",
    "        step = max(np.random.normal(1,1),0.2)\n",
    "        tryagain = True\n",
    "        while tryagain:\n",
    "            try:\n",
    "\n",
    "         \n",
    "                clean = 4\n",
    "                if size >=50:\n",
    "                    clean = 8\n",
    "                    \n",
    "                clean = False\n",
    "                \"\"\"\n",
    "                ModelN,Model_num,s,sc,real_traj,norm,Z = generate_traj(size,sub=True,\n",
    "                                                                       clean=clean,diff_sigma=2.0,\n",
    "                                                                       delta_sigma_directed=1.,ndim=ndim,\n",
    "                                                                      anisentropy=0.1,deltav=0.2,rho_fixed=False)\n",
    "                \"\"\"\n",
    "                clean=10\n",
    "                \n",
    "                \n",
    "                \n",
    "                real_traj = diffusive(1,2,size+1)\n",
    "                \n",
    "                real_traj[:10] *= 2\n",
    "                real_traj[-10:] *= 2\n",
    "                ModelN = 2\n",
    "                Model_num = 2\n",
    "                sc = [0] * size\n",
    "                s = sc\n",
    "                Z = []\n",
    "                \n",
    "                \n",
    "                mu=2\n",
    "                Ra0 = [0,1.]\n",
    "             \n",
    "                \n",
    "                alpharot = 2*3.14*np.random.random()\n",
    "                \n",
    "                dt = real_traj[1:]-real_traj[:-1]\n",
    "                std = np.mean(np.sum(dt**2,axis=1)/3)**0.5\n",
    "                \n",
    "                \n",
    "                noise_level = .25*np.random.rand()\n",
    "                real_traj += np.random.normal(0,noise_level*std,real_traj.shape)\n",
    "                \n",
    "                real_traj  = random_rot(real_traj,alpharot,ndim=ndim)\n",
    "                \n",
    "\n",
    "                \n",
    "                #print real_traj.shape\n",
    "                alligned_traj,normed,alpha,_ = traj_to_dist(real_traj,ndim=ndim)\n",
    "                simple=True\n",
    "                if not simple:\n",
    "                    real_traj1 = np.array([Propertie(real_traj[::,0]).smooth(2),\n",
    "                                           Propertie(real_traj[::,1]).smooth(2)])\n",
    "                    alligned_traj1,normed1,alpha1,_ = traj_to_dist(real_traj1.T,ndim=ndim)\n",
    "                    real_traj2 = np.array([Propertie(real_traj[::,0]).smooth(5),\n",
    "                                           Propertie(real_traj[::,1]).smooth(5)])\n",
    "                    alligned_traj2,normed2,alpha2,_ = traj_to_dist(real_traj2.T,ndim=ndim)\n",
    "\n",
    "                    normed = np.concatenate((normed[::,:4],normed1[::,:4],normed2),axis=1)\n",
    "\n",
    "                for zero in Z:\n",
    "                    normed[zero,::] = 0\n",
    "            \n",
    "                tryagain=False\n",
    "                \n",
    "                \n",
    "            except IndexError:\n",
    "                tryagain=True\n",
    "                \n",
    "        Y_train_traj.append(real_traj)\n",
    "        X_train[i] = normed\n",
    "      \n",
    "        Y_trains[i][range(size),np.array(sc,dtype=np.int)] = 1\n",
    "     \n",
    "        Y_train_cat[i,0,Model_num] = 1\n",
    "        \n",
    "    \n",
    "    return X_train,Y_trains,Y_train_cat,Y_train_traj\n",
    "\n",
    "Normed = generate_N_nstep(1000,100)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "Normed = generate_N_nstep(1000,200)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from Specialist_layer import return_layer_paper\n",
    "\n",
    "graph = return_layer_paper(n_layers=4,simple=False,inputsize=5)\n",
    "graph_simplo = return_layer_paper(n_layers=4,simple=True,inputsize=5)\n",
    "\n",
    "graph_simplo.load_weights(\"weight-folder/4layer-LSTM-simple_7_140\")\n",
    "graph.load_weights(\"/home/jarbona/cluster_theano/weight-folder/4layer-LSTM_7_140\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "p = graph.predict({\"input1\":np.array([t for t in Normed])},batch_size=50)\n",
    "ps = graph_simplo.predict({\"input1\":np.array([t for t in Normed])},batch_size=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAFvCAYAAAAfX1/MAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\no6o3ZgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3cf9a184d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "title(\"LSTM\")\n",
    "hist(np.argmax(p[\"category\"],axis=-1),range=[-0.5,11.5],bins=12)\n",
    "xlim(-0.5,11.5)\n",
    "t = xticks(np.arange(0,12,1),M0,rotation=90)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAaEAAAFvCAYAAADjSnQGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xu4ZFV95vHvC42IFxANcJTmDoOoEMTQ6IMZDyYKKAGC\nEVCTGIi3IUQEQwRH7SNeIkwwSYcwgyLYMgi0otxUREKKTIyCCh2gG6GDNjftBkENBjVgv/PH3o3V\nRZ1bn9p71enzfp7nPFTtXVW/tYvq+tVee63fkm0iIiJK2Kh0AyIiYu5KEoqIiGKShCIiopgkoYiI\nKCZJKCIiikkSioiIYpKEIiKimCShiBmStFLSo5L+Q9IPJV0g6Zn1vk9LWiPpt7oev4ukNV33O5J+\nLmnbrm2/I+n77R5JRPuShCJmzsBrbW8O/CawJ/C+rn0PAR/u85zu2z8D3j/BYyI2SElCEYMhANsP\nAF8FXti1bzGwl6TfnuD5i4A3SNqpuSZGDJ8koYgBkjQfOBi4oWvzo8BH67/x3A98EjitudZFDJ8k\noYjBuEzSfwD3AHcBH+nZ/wlge0kHTvAaHwMOkbRHQ22MGDpJQhGDcVh9TWgUOAB4SfdO2/8FfKj+\n68v2j4CzJnpMxIYmSShiMNZeE/pnqkRyep/HnA88Czhigtf5a/oksYgNVZJQxOD9LbBA0n7dG23/\nChgD3jPeE23/lCoR/WWTDYwYFklCETO3zlDqulttMVWy6R1mfRHwQ548RLvbIuDxPtsjNjia6qJ2\nkjYCvg3cZ/tQSVsClwA7ACuBI+tfcUg6FTiW6h/SCbavaaDtERExy03nTOgEYHnX/VOAa23vDlwH\nnAog6QXAkcAeVENVz5akwTQ3IiI2JFNKQvXch9cA53ZtPoyqy4H6v4fXtw8FLrb9uO2VwApgwUBa\nGxERG5Spngn9DXAy6/ZRb2N7NYDtVcDW9fZtgXu7Hnd/vS0iImIdkyYhSa8FVtteSj0MdRy5iBoR\nEdMybwqP2R84VNJrgM2AZ0q6AFglaRvbqyWNAA/Uj78f2K7r+fPrbeuQlKQVETFH2O57EjPpmZDt\n99re3vbOwNHAdbb/CLgS+JP6YW8GLq9vXwEcLekpdTHGXYEbx3nt9f5buHDhjJ4/07+5HH8uH3vp\n+HP52EvHn8vHPtP4E5nKmdB4PgYskXQscDfViDhsL5e0hGok3WPAcZ6sFRERMSdNKwnZvh64vr79\nMPC74zzur4C/mnHrIiJigzZrKyaMjo4m/hyMPdfjz+VjLx1/Lh97k/GnXDFh4IGl9NJFRMwBkvD6\nDkyIiIhoSpJQREQUkyQUERHFJAlFREQxSUIREVFMklBERBSTJBQREcUkCUVERDFJQhERUUySUERE\nFJMkFBERxSQJRUREMUlCERFRTJJQREQUkyQUERHFJAlFREQxSUIREVFMklBERBSTJBQREcVMmoQk\nbSrpBkk3S1om6aP19oWS7pN0U/13UNdzTpW0QtLtkl7d5AFETNfIyI5IKvI3MrJj6cOPGCqyPfmD\npKfZflTSxsDXgXcDvws8YvvjPY/dA/gssC8wH7gW2M09gST1bopohSSg1GdP5HMfc40kbKvfvil1\nx9l+tL65af2cH6997T4PPwy42PbjtlcCK4AF02pxRETMCVNKQpI2knQzsAro2F5e7zpe0lJJ50ra\not62LXBv19Pvr7dFRESsY95UHmR7DfBiSZsD10h6BXA2cJptS/owcCbwlukEHxsbe+L26Ogoo6Oj\n03l6REQMoU6nQ6fTmdJjp3RNaJ0nSO8HHrV9Zte2HYArbe8l6RTAtk+v910NLLR9Q8/r5JpQFJFr\nQhHtmtE1IUm/sbarTdJmwKuApZJGuh52BHBbffsK4GhJT5G0E7ArcONMDiAiIjZMU+mOey6wWNXP\nx42AC2z/o6TPSNobWAOsBN4OYHu5pCXAcuAx4Lic8kRERD/T7o4bWOB0x0Uh6Y6LaNeMh2hHREQ0\nIUkoIiKKSRKKiIhikoQiIqKYJKGIiCgmSSgiIopJEoqIiGKShCIiopgkoYiIKCZJKCIiikkSioiI\nYpKEIiKimCShiIgoJkkoIiKKSRKKiIhikoQiIqKYJKGIiCgmSSgiIopJEoqIiGKShCIiopgkoYiI\nKGbSJCRpU0k3SLpZ0jJJH623bynpGkl3SPqqpC26nnOqpBWSbpf06iYPICIiZi/ZnvxB0tNsPypp\nY+DrwLuBQ4GHbJ8h6T3AlrZPkfQC4EJgX2A+cC2wm3sCSerdFNEKSUCpz57I5z7mGknYVr99U+qO\ns/1ofXPT+jk/Bg4DFtfbFwOH17cPBS62/bjtlcAKYMH6NT0iIjZkU0pCkjaSdDOwCujYXg5sY3s1\ngO1VwNb1w7cF7u16+v31toiIiHXMm8qDbK8BXixpc+CrkkZ5cn/GtPsYxsbGnrg9OjrK6OjodF8i\nIiKGTKfTodPpTOmxU7omtM4TpPcDPwf+FBi1vVrSCPBPtveQdApg26fXj78aWGj7hp7XyTWhKCLX\nhCLaNaNrQpJ+Y+3IN0mbAa8CbgauAP6kftibgcvr21cAR0t6iqSdgF2BG2d0BBERsUGaSnfcc4HF\nqn4+bgRcYPsf62tESyQdC9wNHAlge7mkJcBy4DHguJzyREREP9PujhtY4HTHRSHpjoto14yHaEdE\nRDQhSSgiIopJEoqIiGKShCIiopgkoYiIKCZJKCIiikkSioiIYpKEIiKimCShiIgoJkkoIiKKSRKK\niIhikoQiIqKYJKGIiCgmSSgiIopJEoqIiGKShCIiopgkoYiIKCZJKCIiikkSioiIYpKEIiKimEmT\nkKT5kq6TtEzSrZL+vN6+UNJ9km6q/w7qes6pklZIul3Sq5s8gIiImL1ke+IHSCPAiO2lkp4BfAc4\nDDgKeMT2x3sevwfwWWBfYD5wLbCbewJJ6t0U0QpJQKnPnsjnPuYaSdhWv32TngnZXmV7aX37Z8Dt\nwLZrX7vPUw4DLrb9uO2VwApgwfo0PCIiNmzTuiYkaUdgb+CGetPxkpZKOlfSFvW2bYF7u552P79O\nWhEREU+YchKqu+I+D5xQnxGdDexse29gFXBmM02MiIgN1bypPEjSPKoEdIHtywFsP9j1kE8CV9a3\n7we269o3v972JGNjY0/cHh0dZXR0dIrNjoiIYdXpdOh0OlN67KQDEwAkfQb4ke2TuraN2F5V3z4R\n2Nf2GyW9ALgQ2I+qG+5rZGBCDJEMTIho10QDEyY9E5K0P/Am4FZJN1P9630v8EZJewNrgJXA2wFs\nL5e0BFgOPAYcl2wTERH9TOlMqJHAOROKQnImFNGuGQ3RjoiIaEqSUEREFJMkFBERxSQJRUREMUlC\nERFRTJJQREQUkyQUERHFJAlFREQxSUIREVFMklBERBSTJBQREcUkCUVERDFJQhERUUySUEREFJMk\nFBGtGRnZEUlF/kZGdix9+NFH1hOKOSfrCZWT935uynpCERExlJKEIiKimCShiIgoJkkoIiKKSRKK\niIhiJk1CkuZLuk7SMkm3SnpnvX1LSddIukPSVyVt0fWcUyWtkHS7pFc3eQARETF7TTpEW9IIMGJ7\nqaRnAN8BDgOOAR6yfYak9wBb2j5F0guAC4F9gfnAtcBuveOxM0Q7Sskw4XLy3s9NMxqibXuV7aX1\n7Z8Bt1Mll8OAxfXDFgOH17cPBS62/bjtlcAKYMGMjiAiIjZI07omJGlHYG/gm8A2tldDlaiAreuH\nbQvc2/W0++ttERER65hyEqq74j4PnFCfEfWe1+Y8NyIipmXeVB4kaR5VArrA9uX15tWStrG9ur5u\n9EC9/X5gu66nz6+3PcnY2NgTt0dHRxkdHZ1W4yMiYvh0Oh06nc6UHjul2nGSPgP8yPZJXdtOBx62\nffo4AxP2o+qG+xoZmPAkIyM7snr13UVib7PNDqxatbJI7GGQi+Pl5L2fmyYamDCV0XH7A/8M3Er1\n6THwXuBGYAnVWc/dwJG2f1I/51TgT4HHqLrvrunzunM6CeUfYzl578vJez83zSgJNSVJKP8YS8l7\nX07e+7kpVbQjImIoJQlFREQxSUIREVFMklBERBSTJBQREcUkCUVERDFJQhERUUySUEREFJMkFBER\nxSQJRUREMUlCERFRTJJQREQUkyQUERHFTGlRu4gYlE3rStJlzPW1pGL4ZCmHQlLSvpzS73252FX8\nkv/vS7/3c/lzX1KWcoiIiKGUJBQREcUkCUVERDFJQhERUUySUEREFJMkFBERxUyahCR9StJqSbd0\nbVso6T5JN9V/B3XtO1XSCkm3S3p1Uw2PiIjZbypnQucDB/bZ/nHb+9R/VwNI2gM4EtgDOBg4WyVn\n5kVExFCbNAnZ/hfgx3129UsuhwEX237c9kpgBbBgRi2MiIgN1kyuCR0vaamkcyVtUW/bFri36zH3\n19siIiKeZH1rx50NnGbbkj4MnAm8ZbovMjY29sTt0dFRRkdH17M5ERExLDqdDp1OZ0qPnVLtOEk7\nAFfa3muifZJOAWz79Hrf1cBC2zf0eV5qx6WGVhGl3/vUjsvnfq4ZRO040XUNSNJI174jgNvq21cA\nR0t6iqSdgF2BG6ff5IiImAsm7Y6T9FlgFHiOpHuAhcABkvYG1gArgbcD2F4uaQmwHHgMOG5On+5E\nRMSEspRDIemWKKf0e5/uuHzu55os5RAREUMpSSgiIopJEoqIiGKShCIiopgkoYiIKCZJKCIiikkS\nioiIYpKEIiKimCShiIgoJkkoIiKKSRKKiIhikoQiIqKYJKGIiCgmSSgiIopJEoqIiGKShCIiopgk\noYiIKCZJKCIiikkSioiIYpKEIiKimEmTkKRPSVot6ZaubVtKukbSHZK+KmmLrn2nSloh6XZJr26q\n4RERMftN5UzofODAnm2nANfa3h24DjgVQNILgCOBPYCDgbMlaXDNjYiIDcmkScj2vwA/7tl8GLC4\nvr0YOLy+fShwse3Hba8EVgALBtPUiIjY0KzvNaGtba8GsL0K2Lrevi1wb9fj7q+3RUREPMmgBiZ4\nQK8TERFzyLz1fN5qSdvYXi1pBHig3n4/sF3X4+bX2/oaGxt74vbo6Cijo6Pr2ZyIiBgWnU6HTqcz\npcfKnvwkRtKOwJW296zvnw48bPt0Se8BtrR9Sj0w4UJgP6puuK8Bu7lPEEn9Ns8Z1XiNUscv8t6X\ne+/LdhyU/X9f+r2fy5/7kiRhu+8gtUnPhCR9FhgFniPpHmAh8DHgc5KOBe6mGhGH7eWSlgDLgceA\n4+Z0pomIiAlN6UyokcA5EyK/CMso/d7nTCif+7lmojOhVEyIiIhikoQiIqKYJKGIiCgmSSgiIopJ\nEoqIiGKShCIiopgkoYiIKCZJKCIiikkSioiIYpKEIiKimCShiIgoJkkoIiKKSRKKiIhikoQiIqKY\nJKGIiCgmSSgiIopJEoqIiGKShCIiopgkoYiIKCZJKCIiikkSioiIYubN5MmSVgI/BdYAj9leIGlL\n4BJgB2AlcKTtn86wnRERsQGa6ZnQGmDU9ottL6i3nQJca3t34Drg1BnGiIiIDdRMk5D6vMZhwOL6\n9mLg8BnGiIiIDdRMk5CBr0n6lqS31Nu2sb0awPYqYOsZxoiIiA3UjK4JAfvb/qGkrYBrJN1BlZi6\n9d6PiIgAZpiEbP+w/u+Dki4DFgCrJW1je7WkEeCB8Z4/Njb2xO3R0VFGR0dn0pyIiBgCnU6HTqcz\npcfKXr8TFUlPAzay/TNJTweuAT4I/A7wsO3TJb0H2NL2KX2e7/WNvSGQRLmTRJH3vtx7X7ZzoOz/\n+9Lv/Vz+3JckCdvqt28mZ0LbAF+U5Pp1LrR9jaRvA0skHQvcDRw5gxgREbEBW+8zoRkHzpkQ+UVY\nRun3PmdC+dzPNROdCaViQkREFJMkFBERxSQJRUREMUlCERFRTJJQREQUkyQUERHFJAlFREQxM60d\nFxERs8DIyI6sXn136WY8SSarFpJJe+WUfu8zWTWf+xKG4L3PZNWIiBguSUIREVFM0WtC22yzS7HY\n55xzJocfnkVfIyJKKpqEHnjgmkKRF3HTTTcnCUVEFFZ4dFypM6FnA2sKxY6IiLVyTSgiIopJEoqI\niGKShCIiopgkoYiIKCZJKCIiikkSioiIYpKEIiKimMaSkKSDJH1X0p2S3jP4CJ3Bv+R0onfKxi95\n/KWPvXT8sp+9krHn9ntf+thLx2/qvW8kCUnaCDgLOBB4IfAGSc8fbJTOYF9uutE30A/ElCLnH+Mc\njT233/vSx146/qxKQsACYIXtu20/BlwMHNZQrIiImKWaKtuzLXBv1/37qBJTj+/MIMQPZvD8HwAj\nM4gdERGD0MiidpJeBxxo+231/T8EFth+Z9dj5u7qUhERc8x4i9o1dSZ0P7B91/359bZJGxQREXNH\nU9eEvgXsKmkHSU8BjgauaChWRETMUo2cCdn+laTjgWuoEt2nbN/eRKyIiJi9GrkmFBERMRWpmBAR\nEcUkCU2DpH1Kt2GYSHpX4fj7lYw/l0maX7oNsWGYdUlI0laStioU/kxJt0v6kKQXtRlY0jxJvyfp\n5PrvEEmFl2fnpMLxP9dGEEkHSPqCpGX13+cljbYRe4I2fb2lOPtIer2kF9b350v6BNBK/GEj6enN\nlCGbVhsGXH1m2vE/MMjXmxVJSJUxST8C7gDulPTgoN+Mydg+ADgAeBA4R9Ktkt7XdFxJ2wLLgHcD\nz6OaDHwysEzS85qOP1HTCsZuJb6k1wLnAVcCbwTeBHwZOE/Sa5qOP4HtJ3/IzEj6MHAhcARwuaQz\ngX+m+izu3nT8Sdr2lYZf/7mSFkn6kqQzJD1D0olU3z8l/81BNeCrpLcM8sVmxcAESScBBwNvs/39\netvOwP8Grrb9NwXatCfwl8BRtp/ScKxPA0tt/23P9ncCL7H95ibjT9Cue2w3/mVYMr6kDnCC7X/r\n2b4X8Pe2X9Fk/Ana1caxL6P6fP1C0pZUVVBeZHtlk3G74o/X/S3gKtvPbTD216jO9r4BHAQcDnwT\nONH2qqbidsVfNN4u4M22N284/n9MEH8z2wPrhZktSehm4FW2f9SzfSvgGtsvbqkdewBHAa8DHgIu\nAS61/UDDcb9ru+8puKQ7bDf2q1TSI4CpPnzdH5aBfxjHiX9lT9zu+K+0/fSG40/03o+7b0Cxjxhv\nF/B/bDfaLS3pJtv7dN2/ua1/a3W8XwHX0/+M96W2N2sw9lLbe3fdvw/Y3vaapmL2xH+Equfjl312\nn2n7NxqOfw+wr+3Vffbda3u7QcUqfU1hqjbpTUAAth+UtEmL7TiPqhjrgbZ/0GLcn0+w79EmA9t+\nZpOvPwV/vZ77BuU/13PfIPzeBPuuajg2wM6S1k4yF7BT131sH9pw/NuBt9te0btD0r19Hj9Iqs/+\n1ibAh4AtJAnA9sMNx/8WcJvtf+3TsLGGYwN8BtgBeFISAj47yECz5UxonV9kU93XUFu2gioBthjz\ne8Bf9NsFnGF7lxbasAi4yPY3mo41TvwjgC/Z7vfLsMm4P6G6DvKkXcDLbW/ZQhs2tv2rpuP0iTth\nV6Pt6xuO/wfArbbv6LPvcNuXNRh7JbCG/mdhtr1zU7Hr+M8GfmG70R+Zw2C2JKFf0f9Xp4Cn2m70\nbKj+9bMQ+DNg4zru41TXBE5rMnYd//yJ9ts+poU2vJmqK3J34IvAxba/3XTcrvjnA6+kSgiXUF0L\nfLyFuEW/iOs23ANcTXXc17nlf7SlfgBEf5IW2L6xUOztgPfa/h8De1Hb+Zvkj2oo8teAnbq27Qx8\nlepCZfE2tvhePBt4K/CPVGtGtRl7E+BQqhFbdwPnFnoP9mk53tOAI4EvACupFox8eYvxz6/f7wuA\nQ4B5LR//VtQ/mOv7b6Tqqmoj9hXAG4Cnt3zMohqVeDLwmnrbS6hGxt3cQvw9qEaELgOWUBWh/rv6\n8zfQ77xZMUR7CPwR8AbXI/MAbH8P+EPgj0s0SFIb1wT62RV4PlV/8XfbDOxqgcSvUF2X+w7ViKUS\nzm0zmO1HbS+xfQTwYmBzqgv2bcU/hur/++eovpDvktT4eyDpiHpaxi3AvZIOlXQT8Hra+3d3JvDb\nwPJ6ftgfSHpqC3E/QdXz8mzgfZIuBRYDZ7udwSGfAi6lSoT/SvX/4JfA8z3g0cizojuuNEm32e47\nOXWifQ23qe2RSmcAvw/cRdUt9EXbP2kx/sFU3YGjVOsML6EaGdl4l1yftrT63tcxX0F1/AcB3wYu\nsX1py23YpI5/DPDf3fwIrVuAI2z/ez1c+wbg9223/gNM0sZU3cFvBQ5y80OkbwV+0/aaOumtAnax\n/VCTcbvi944O/J4bug42W0bHlfZf67mvSTe3HO8u4GXuM0qxJX9MdQb0dpe/NvHBNoPVF8lvpkq8\nJ9tuelReb/zeHwDnUnUPNu1x2/8OYPumejpCiQS0GdVIxaOAfajOSJr2mOvh4K7maX2vrQRUe6qk\nF/PrgRm/7L5v+6ZBBcqZ0BSUHhjR05YSo/PmUU0WXjsn5nZaGhjQ1YZnAbvVd++0/dMWYxc9fkmb\n2x5v8mAb8S+i+gFwdZs/AOq5OR/v2nRS933bH3/SkwbfhiXAflTdwJcA17uFuUKSHgX+fe1dYJf6\nvqhG5+3VcPx/mmC3bb9yYLGShIZf1+i846lKLbU2Oq8uGXQd8EOqX+Oiui4xAhzghudLSdoUOAc4\nDPg+1fHvQDVC7x22Gz0THYLjP4Dq/3t3AjzLdqfJuD1tKPIDQNLCifbbbvyMVNKBwLVueYi8pB0m\n2m/77rba0rQkoVmgZNmi0iWDJJ1G9SvwHbYfqbc9E/gH4G7b7284/qcpdPyq6tadBZwG3ESVAPcB\n3gccb/vLTcWu4xf9AdDVjq3aPPPvivscqpF43T8ALmqzW6z+t39x0z92Joh/C3ARsMT2XY3ESBIa\nfiXLFpUsGVTHuA1Y4J5Je5KeAXyz6UEhhUsmdShYt670D4CudtxJNTT4EuALtn/cQsw9qM6Av8q6\nZ8CvoioX1crI0Pps8EjgYarj/5z7lNJpMP4OVNfCjqKavHsJVUK6Z2AxkoSGX8nReRONBGtjlJik\nW8br/5Z0q+09G45f7PgnSYCN1q2rYxT9AdATcwFwNNWw/OVUZwf/t8F4n6f6sl3Ss/11wBttv66p\n2OO0Zy9+XbfyPtu/22b8ug27Ae8H3mR740G9bkbHzQ4lR+dtof6FNEU1X6Vp1ro1vLq1UUyy5PGX\nrFsHsKY3AQHY/pmkVn+9uqoQcKOkj1INTlgMNJaEgD1t/0Gfdlxat6FtD1AN034I2LrNwD1nQ7+i\nWj1gYJKEZoffVP/S6gKanjh3PeMX0uxXU23QtqCamNq3hlcL8Use/y7qKhjaRVQVO5pW+gcAUI0O\npJqjdjRV9+AXgQUNhy39AwAAScdRdcdtRTVZ+K22l7cY/waqSiWfA15fT9IfbIx0x0UMp9J160oX\n8exqx/eBy6i6x1opoNtnePgTu4B3eYBLGUzSjr+impi8tI14feLv7j4FZAcaI0kopkvSVbYPKRh/\nzPZYwfjFjl/SPoOcKDgbSJJb/qIahuHhdTv2pGt0nu3b2ohbx94deBvrjg785KCTUpJQTFuJsjU9\n8VtdvqNP/GLHPwTH3uoPAFXV29/Jul+Ei2x/pq02lCBpC+ByYDuqum0C9gTuAQ5revKypJdRFcw9\nh3VHB76VqpTSNwcVK9eEYn20XTKoV7/uoTaVPP7Sx34oMNZGoDoBvYuqUkL3PKn/VZ8cXdBGO7ra\n0+YPgA9R1Qh85doKDZI2Aj4GfAT484bjf4CqaHOna9tlkq6jmjh/8KAC5UwopqxEyaBx2rGRq8KO\nT+s3eqvBuMWPXw0v5jaF+K2dBUr6JnC07ZU923ekGqL90jba0RW3zWNfDuzVWxqqLiF1q+09Go5/\np+3/Ns6+gc6Py1IOMSFVxlSV1L8DuFPSg5I+0GIbtpK0r6rVJgE2k/ReWlhKovTxS5on6fcknSzp\nZODx+ouolJfU7XpXC7E2701AAPW2NqYH9PpSi7H+q19twnpbG/X7Hplg30BHB6Y7LiZzIrA/sG9v\nySBJJzZZMqiOdRzV6f/3gJ0kfYSqi+YKqsKSTSt2/Opft+4Q4ExJjdet66ereOdJwN9O9NgB+Pl6\n7muE7fcBSPq67f0bDtdbxXotAZs2HBtgO0mL+mwXsO0gA6U7LiZUsmRQHWcZ8Nu2H5a0PXAnsL/t\n7zQZtyt+yZJJn6Zg3b6JSLq36WHKWreS9Dq7gJ1tP73J+ONp6dgnqmKN7QMajj/hZ8v2wJazyJlQ\nTGaT3i9gqK6LqFrkrGm/tP1wHfOeuj+6lQRUK3n8L7X9J31iL5LU6NyNKWjj12uj1z1moPFjbzrJ\nTCH+k5KMpBHbqwYdK0koJlN6Qb/5Pd0Cz+2+b/udDccvefwTdTk1PiBD0iNUX7hi3S9eAZs1Hd99\nliuQdIhbWNhunFJN0NKx9w0sfcL220rErn2ZanTiQCUJxWRKlgwCOLnnfptnQVD2+IvW7bP9zKZj\nrIfTgDZWVx2vVBMtxe/ntwrFXauR6QFJQjGhQVbLXc/4i6FcpYDCx1+6bh8A9ZnnRW2VzJlEK/Ok\nbB8DIGljt7yg3QQeKBz/k028aAYmxKxQX6gdAT5PVUurtfIlc119kfooYHeq4qEX2/52obYsqCtq\ntxXvHuBqqnV0rmu7fFA/kubbvq9Q7KdTLah4+qBeM/OEYlaoL9QeADwInCPpVknvK9ys1klqvSvI\n9mLbrwH2pZordbqkFW23o27LjZJeKukrLYV8PnAt8GfA9yWdJenlbQSWtI+k10t6YX1/vqRPAF9v\nIfZzJS2S9CVJZ0h6hqQTqf7/P2+QsZKEYtawvcr2IuAdwFKq0iJzzUDnaEzTrlRfyjvQzkThl9c/\nNv5T0o2SXiLpcqqVXRvpGupl+1HbS2wfQVU7bXOqbtJGSfowcCFwBHC5pDOB/wcsozojbdpnqFZz\nXUS1lMOtVMtn/JbtEwYZKN1xMSuoWm557cqSD1F1j1xqu3Q/easknWf72JZjnkG1ns9dVO/7F23/\npIW43wFXkSjuAAAIi0lEQVTeDXyDqlbZRcBf2P6HpmP3tOMVVJ+9g6jquV1i+9KGYy6jmgv2C1Vr\nOt0LvKhfBYmG4i+1vXfX/fuA7bsmKw9MBibEbHEecDFwYIlKAaV11a1rNQHV7gJe1m++VMPUVUDz\nMkn3FEhAK6mqVSwBTrbd1oJ2v7T9CwDbP5a0oq0EVJPWXdDwIarRmqrb9PCgAiUJxaxg+2Vrv4jn\nivof/ELgeKquc0l6HPh726e11IZ5wA+AY+rvn9uBq/vVNWvAs3qGqM/rvm/7Cy20Ya+ml00Yx876\n9aq6oipZ9cQqu7YPbTh+vxWN145ONQNc2TfdcTHUur6I/wzYmOofRatfxKVIOomqG+ptvXXrqBJB\n03X7+tWuezHVKMXGa9dJOn+C3W76rFDSAVQ/ALrXMjqrZ3mDpmIXXVW3TUlCMdRKfxGXNAR1+z7N\nENSuKzFXR9JrgbOoJsd2r2X0Pqohyl9uqR1HAF+y3Ubl7H7xr6C6FndFU12RSUIx1Ep/EZck6Tbb\nL5ruvgHG/67t54+zb6BrykzSjtbn6kjqACfY/ree7XtRnYVPeKYywHacD7ySanLyJbTXFbo2/tpB\nGa8FvkV1XfaqtderBiFDtGPYjVtAlGro6IasdN2+orXrupSYqzPSm4AAbN8CbNNw7O54x1ANjf8c\n8AbgLknnthj/etvHUV0DOgc4kgFXbsjAhBh2pb+ISypdt69o7bq1XK2euwRYUo/Y+juquTpNllSa\nqOuprRFyANh+rJ6ca6riqYcDb2krvqTNqMpHHUXVJTmwZRwgSSiGX+kv4mJK1+1jSGrXQd+5Okc2\nHHKX7tFo3U1hgCPDJiPpYKrjHgU6wLk0f+zd8ZdQLR75FaprZNcPeq5QrglFxFDrmavT2AXynphD\nMTpN0kVU12GuLjE4QdKBwLVNDgxJEoqIKZN0le1DWo65eaG5Or3taL2Su6RnAbvVd++0/dMWYz8H\neCPrDlG/yPZDg4yTgQkRMR2t1q6r5+qcL2lZ/fd5SaNttqFLawMCJG1aD5H/PtWAgE8CKyWdJ+kp\nLcTfA7gNeAlwJ7CCqoDtrZL6jphcX7kmFBHTcXNbgXrm6pzGr+fqnCeptbk63U1qMdb/pBr9ub3t\nRwAkPZOqeOv7678mfYhqiPqS7o2SXgd8hKqG40CkOy4iJtVVu+7BFmN2GIK5Ol1xD7d9WUuxbgMW\n1CMDu7c/A/hmC3PExp0HNug5YjkTioi+hqB23bhzdSQ1Plenrpt3MF3XRCTNa2my6JreBARg+2eS\n2jhzaG2IepJQRIznRGB/YN/ekkmSTmyhZFKxuTrj1M07BDhTUuN18wD3VLHuNvDlFPrYui6Z1UvA\nQAsJpzsuIvoqXTJJ0k/oPx9JwMttb9lg7E9TsG5ePSx9Df2TkG03OldJ0sKJ9tv+4MBiJQlFRD9D\nULuu2FydYambNxekOy4ixlO0ZFK/JNPiXJ1hqZv3BEljtsdKxK7j32R7n0G/bpJQRIxnGEsmnUs1\nTLtpQ1E3r8ehwFih2NDQEPUkoYjoawhq1/XT1lydoamb16XNeUr9fKmJF801oYiYNdbO1ZH0vBZG\nqA0VSRvZXiPpXb0DJlpux9dt7z+w10sSiojZRtI9trdvOWbrdfPGaUfrx94T/17b2w3q9VI7LiJm\noxJdU63WzZtA6W65gZ655JpQRMxGJbpwWqubN4k2ljfvNygDqgS42SBjJQlFxFCS9Pf0/8IV8KwW\n27G2bt6xLcZ8hOrYxbrvwcCTwDjGG5QBcNUgA+WaUEQMJUkTViWwPdBlpntiP6luHtBm3byhIGnj\nJhe0gyShiBhyhRaTO4mqeOnbeuvmUa1y2nTdvLXtWES1kNw32ojXJ/49wNXAJcB1biBhJAlFxFCT\n9E/ACPB54BLbt7UQs2jdvK54bwaOAnYHvghcbPvbbcSu4z+NqnDr0VSThK+q2/AvA4uRJBQRw07S\nCHAk1Rfy5lTJ6MMNxitaN69PzGdTLSR3NNVCd7tN8pQm2rAl8HfAmwY5kTlDtCNi6NleZXsR8A5g\nKfCBhkMWrZvXx65U6xrtAHy3zcCSXiHpbOA7VOWajhzo6+dMKCKGmaQ9qM6AXgc8RHV94lLbDzQY\n81f0X7NIwFNtb9JU7J52nAH8PnAX1XF/0fZP2ohdx19JNTR9CXCF7YGv45Qh2hEx7M4DLgYObKtU\nzxDVzbsLeFnvtakW7WW7XxHbgcmZUEQMva65Og+Wbktb+i0vTjUyr43lxZF0ANUQ9e74Z9nuDDJO\nrglFxFBSZUzSg8AdwJ2SHpTU9PWg4urlxZcB7waeR1Uy6GRgmaTntRD/tVRnoFcBbwTeBHwZOE/S\nawYaK2dCETGMhmWuTglDsLx4BzjB9r/1bN+LasLuhKveTitWklBEDKNhmatTQunlxSeJP+6+9ZHu\nuIgYVpv0uyBfXxdqZXRaQaWXF59oFNxAR8hldFxEDKthm6vTptLLi+8i6Ypx4u88yEDpjouIoTQs\nc3VKkHT+RPttH9Nw/Amv+di+fmCxkoQiImIyTRWSzTWhiIhZQNJA1/FZD+c28aJJQhERs0Pp5cUb\nWVY8SSgiYnYovbz4B5t40VwTiogYYqVKFrVVNihnQhERQ6arZNGPKFCyqM2yQTkTiogYMqVLFrVZ\nNihJKCJiyJQuWdRm2aB0x0VEDJ/SJYtaKxuUsj0REcOndMmi1soGpTsuImLIlC5Z1GbZoCShiIgo\nJteEIiJiUk2VDUoSioiIqWikbFCSUERETEUjZYNyTSgiIsbVdNmgnAlFRMQ62iwblCQUERG9TgT2\nB/a1/WzbWwL7AftLOnGQgdIdFxER62izbFDOhCIioldrZYOShCIioldrZYPSHRcREetos2xQklBE\nRBST7riIiCgmSSgiIopJEoqIiGKShCIiopgkoYiIKOb/AzaT9NsOymmxAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f3cfc762ad0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from Tools import M0\n",
    "title(\"RNN\")\n",
    "hist(np.argmax(ps[\"category\"],axis=-1),range=[-0.5,11.5],bins=12)\n",
    "xlim(-0.5,11.5)\n",
    "t = xticks(np.arange(0,12,1),M0,rotation=90)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}