relation_reasoning_code_use_multi_gpu.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Relation Networks for VQA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:11.225645Z",
     "start_time": "2018-07-16T04:07:08.641561Z"
    }
   },
   "outputs": [],
   "source": [
    "import mxnet as mx\n",
    "from mxnet import gluon, autograd, nd\n",
    "from mxnet.gluon import nn,utils\n",
    "from mxnet.gluon.data import ArrayDataset, DataLoader\n",
    "import numpy as np\n",
    "import os, sys, time\n",
    "from tqdm import tqdm\n",
    "import pickle\n",
    "import random\n",
    "from tqdm import tnrange, tqdm_notebook\n",
    "import matplotlib as mpl\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "import collections\n",
    "import pandas as pd\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'1.3.0'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mx.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data Preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "directory ./data already exists\n",
      "building test datasets...\n",
      "building train datasets...\n",
      "saving datasets...\n",
      "datasets saved at ./data/sort-of-clevr.pickle\n"
     ]
    }
   ],
   "source": [
    "##generate input data\n",
    "!python sort_of_clevr_gen.py"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model Definition"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:14.157815Z",
     "start_time": "2018-07-16T04:07:14.155331Z"
    }
   },
   "outputs": [],
   "source": [
    "GPU_COUNT = 2 # Set gpu count\n",
    "def setting_ctx(GPU_COUNT):\n",
    "    if GPU_COUNT > 0 :\n",
    "        ctx = [mx.gpu(i) for i in range(GPU_COUNT)]\n",
    "    else :\n",
    "        ctx = [mx.cpu()]\n",
    "    return ctx\n",
    "        \n",
    "ctx = setting_ctx(GPU_COUNT)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:14.825396Z",
     "start_time": "2018-07-16T04:07:14.795850Z"
    }
   },
   "outputs": [],
   "source": [
    "class ConvInputModel(nn.HybridSequential):\n",
    "    def __init__(self,**kwargs):\n",
    "        super(ConvInputModel,self).__init__(**kwargs)                \n",
    "        with self.name_scope():\n",
    "            self.add(\n",
    "                nn.Conv2D(channels=24, kernel_size=3, strides=2, padding=1, activation='relu'),\n",
    "                nn.BatchNorm(),\n",
    "                nn.Conv2D(channels=24, kernel_size=3, strides=2, padding=1, activation='relu'),\n",
    "                nn.BatchNorm(),\n",
    "                nn.Conv2D(channels=24, kernel_size=3, strides=2, padding=1, activation='relu'),\n",
    "                nn.BatchNorm(),\n",
    "                nn.Conv2D(channels=24, kernel_size=3, strides=2, padding=1, activation='relu'),\n",
    "                nn.BatchNorm()\n",
    "            )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:15.633401Z",
     "start_time": "2018-07-16T04:07:15.623210Z"
    }
   },
   "outputs": [],
   "source": [
    "class FCOutputModel(nn.HybridSequential):\n",
    "    def __init__(self,**kwargs):\n",
    "        super(FCOutputModel,self).__init__(**kwargs)\n",
    "        \n",
    "        with self.name_scope():\n",
    "            self.add(\n",
    "                nn.Dense(256, activation='relu'),\n",
    "                nn.Dropout(0.5),\n",
    "                nn.Dense(10)\n",
    "            )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:16.413751Z",
     "start_time": "2018-07-16T04:07:16.256041Z"
    }
   },
   "outputs": [],
   "source": [
    "class RN_Model(nn.HybridBlock):\n",
    "    def __init__(self, **kwargs):\n",
    "        super(RN_Model, self).__init__(**kwargs)\n",
    "    \n",
    "        with self.name_scope():\n",
    "            self.conv = ConvInputModel()\n",
    "            \n",
    "            self.g_fc1 = nn.Dense(256, activation='relu', flatten=False)\n",
    "            self.g_fc2 = nn.Dense(256, activation='relu', flatten=False)\n",
    "            self.g_fc3 = nn.Dense(256, activation='relu', flatten=False)\n",
    "            self.g_fc4 = nn.Dense(256, activation='relu', flatten=False)\n",
    "            \n",
    "            self.f_fc1 = nn.Dense(256, activation='relu')\n",
    "            self.fcout = FCOutputModel()\n",
    "            \n",
    "\n",
    "    def hybrid_forward(self, F, x, qst, coord_tensor):\n",
    "        \n",
    "        #input size = (64 * 3 * 75 * 75)\n",
    "        x = self.conv(x) ## x = (64 * 24 * 5 * 5)\n",
    "                \n",
    "        \n",
    "        x_flat = x.reshape(shape=(0, 0, -3))\n",
    "        x_flat = F.swapaxes(x_flat,1,2) ## (64 * 25 * 24)\n",
    "        \n",
    "        ##add coordinates\n",
    "        x_flat = F.concat(x_flat, coord_tensor,dim=2)\n",
    "        \n",
    "        ##add question\n",
    "        qst = qst.expand_dims(1)\n",
    "        qst = F.repeat(qst,repeats=25,axis=1)\n",
    "        qst = qst.expand_dims(2)\n",
    "        \n",
    "        # cast all pairs against each other\n",
    "        x_i = x_flat.expand_dims(1)\n",
    "        x_i = F.repeat(x_i,repeats=25,axis=1)\n",
    "        \n",
    "        x_j = x_flat.expand_dims(2)\n",
    "        x_j = F.concat(x_j,qst,dim=3)\n",
    "        x_j = F.repeat(x_j,repeats=25,axis=2)\n",
    "        \n",
    "        #concatenate all\n",
    "        x_full = F.concat(x_i,x_j,dim=3)\n",
    "        \n",
    "        x_ = self.g_fc1(x_full.reshape((0, -1, 63)))\n",
    "        x_ = self.g_fc2(x_)\n",
    "        x_ = self.g_fc3(x_)\n",
    "        x_ = self.g_fc4(x_)\n",
    "        \n",
    "        x_g = x_.sum(1)\n",
    "\n",
    "        \n",
    "        ##### f part #######\n",
    "        x_f = self.f_fc1(x_g)\n",
    "\n",
    "\n",
    "        \n",
    "        return self.fcout(x_f)\n",
    "            "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data Loading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:17.078333Z",
     "start_time": "2018-07-16T04:07:17.052989Z"
    }
   },
   "outputs": [],
   "source": [
    "def load_data():\n",
    "    print('loading data...')\n",
    "    dirs = os.getcwd()\n",
    "    filename = os.path.join(dirs,'data','sort-of-clevr.pickle')\n",
    "    with open(filename, 'rb') as f:\n",
    "        train_datasets, test_datasets = pickle.load(f)\n",
    "    rel_train = []\n",
    "    rel_test = []\n",
    "    norel_train = []\n",
    "    norel_test = []\n",
    "    print('processing data...')\n",
    "\n",
    "    for img, relations, norelations in train_datasets:\n",
    "        img = np.swapaxes(img,0,2)\n",
    "        for qst,ans in zip(relations[0], relations[1]):\n",
    "            rel_train.append((img,qst,ans))\n",
    "        for qst,ans in zip(norelations[0], norelations[1]):\n",
    "            norel_train.append((img,qst,ans))\n",
    "\n",
    "    for img, relations, norelations in test_datasets:\n",
    "        img = np.swapaxes(img,0,2)\n",
    "        for qst,ans in zip(relations[0], relations[1]):\n",
    "            rel_test.append((img,qst,ans))\n",
    "        for qst,ans in zip(norelations[0], norelations[1]):\n",
    "            norel_test.append((img,qst,ans))\n",
    "    \n",
    "    return (rel_train, rel_test, norel_train, norel_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:19.062021Z",
     "start_time": "2018-07-16T04:07:17.613339Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loading data...\n",
      "processing data...\n"
     ]
    }
   ],
   "source": [
    "##load input data\n",
    "rel_train, rel_test, norel_train, norel_test = load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:20.115834Z",
     "start_time": "2018-07-16T04:07:19.983675Z"
    }
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:20.568945Z",
     "start_time": "2018-07-16T04:07:20.565752Z"
    }
   },
   "outputs": [],
   "source": [
    "def display_image(data):\n",
    "    img = data[0]\n",
    "    img = np.swapaxes(img,0,2)\n",
    "    plt.imshow((img * 255.0).astype('uint8'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:21.517133Z",
     "start_time": "2018-07-16T04:07:21.415086Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAD8CAYAAABXXhlaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAADMtJREFUeJzt3W+oZPV9x/H3p6uSkrQx6lbE1V5LRPGJu+klVQyl1WyxqWgfBFGSEoLgk7QoTUk1T0RoIXmSxAclIGrqAxu1JlKRYCrG0BbK1jWaJu5qNXbFFXXXRDG1kLLJtw/mbJ1s9+49d+ffPfN7v5bhzjlz7pzf3NnP/H5z5jfnm6pCUlt+ZdENkDR/Bl9qkMGXGmTwpQYZfKlBBl9qkMGXGjRR8JNcnuS5JC8kuWlajZI0WzneCTxJtgD/AewE9gNPANdW1Z7pNU/SLJwwwe9+GHihql4ESHIvcBWwZvBPO+20WllZmWCXko5l3759vPHGG1lvu0mCfybw8tjyfuB3jvULKysr7N69e4JdSjqW1dXVXtvN/OBekuuT7E6y++DBg7PenaQeJgn+K8BZY8vbunW/pKpur6rVqlrdunXrBLuTNC2TBP8J4Nwk5yQ5CbgGeGg6zZI0S8f9Hr+qDiX5U+DbwBbgrqp6ZmotkzQzkxzco6q+BXxrSm2RNCfO3JMaZPClBhl8qUEGX2qQwZcaNNFR/c0k685Oni1PVqwhsceXGmTwpQYZfKlBBl9qkMGXGmTwpQYZfKlBBl9qkMGXGmTwpQYZfKlBBl9qkMGXGmTwpQatG/wkdyU5kOSHY+tOSfJokue7nx+YbTMlTVOfHv9vgcuPWHcT8FhVnQs81i1LGoh1g19V/wT85IjVVwF3d9fvBv54yu2SNEPH+x7/9Kp6tbv+GnD6lNojaQ4mPrhXVQWseeIpi2ZKm8/xBv/1JGcAdD8PrLWhRTOlzed4g/8Q8Knu+qeAf5hOcyTNQ5+P874O/CtwXpL9Sa4DvgDsTPI88NFuWdJArHt67aq6do2bLptyWyTNiTP3pAYZfKlBBl9qkMGXGmTwpQYZfKlBBl9qkMGXGrTuBJ6hsD691J89vtQggy81yOBLDTL4UoMMvtQggy81yOBLDTL4UoMMvtQggy81yOBLDepzlt2zkjyeZE+SZ5Lc0K23cKY0UH16/EPAZ6vqAuAi4DNJLsDCmdJg9Sma+WpVfa+7/lNgL3AmFs6UBmtD7/GTrAA7gF1YOFMarN7BT/I+4BvAjVX19vhtxyqcadFMafPpFfwkJzIK/T1V9c1uda/CmRbNlDafPkf1A9wJ7K2qL43dZOFMaaD6nHrrEuBPgB8kebpb93lGhTLv74povgRcPZsmSpq2PkUz/wXIGjdbOFMaIGfuSQ0y+FKDDL7UIIMvNcjgSw0y+FKDDL7UIIMvNcjgSw0y+FKDDL7UIIMvNcjgSw0y+FKDDL7UIIMvNcjgSw0y+FKDDL7UIIMvNajP6bXfk+Tfkny/K5p5a7f+nCS7kryQ5L4kJ82+uZKmoU+P/zPg0qq6ENgOXJ7kIuCLwJer6oPAm8B1s2umpGnqUzSzquq/usUTu0sBlwIPdOstmikNSN8SWlu6YhoHgEeBHwFvVdWhbpP9jCroShqAXsGvqp9X1XZgG/Bh4Py+O7BoprT5bOioflW9BTwOXAycnORwJZ5twCtr/I5FM6VNps9R/a1JTu6u/yqwE9jL6AXg491mFs2UBqRP0cwzgLuTbGH0QnF/VT2cZA9wb5K/Ap5iVFFX0gD0KZr578COo6x/kdH7fW0yt9566/9dv+WWWxbYEm1WztyTGmTwpQb1eY+vTWR8GD+t7X070B57fKlBBl9qkEP9Adjo8H6S+3fY3wZ7fKlBBl9qkMGXGmTwpQYZfKlBBl9qkMGXGmTwpQY5gWcTm/XEnfX26WSe5WWPLzXIHn8TO9zjzrPnt5dvgz2+1CCDLzXI4EsN6h38rprOU0ke7pYtmikN1EZ6/BsYnU//MItmSgPVt3beNuCPgDu65WDRTGmw+vb4XwE+B/yiWz4Vi2ZKg9WnhNYVwIGqevJ4dmDRTGnz6TOB5xLgyiQfA94D/DpwG13RzK7XP2bRTOB2gNXV1ZpKqzcg5N22MPfdT8X4pJpZTOZx0k571u3xq+rmqtpWVSvANcB3quoTWDRTGqxJPsf/S+DPk7zA6D2/RTOlgdjQXP2q+i7w3e76wotmjg/jp7X9Zn870GdY7jfstB5n7kkNMvhSgwb3tdyNDu8nuf/NPuxfi8N7rcceX2qQwZcaZPClBhl8qUEGX2qQwZcaZPClBhl8qUGDmcAz64k76+1zqJN5pKOxx5caZPClBg1mqH94qD3PIb/Dey0re3ypQQZfapDBlxpk8KUGGXypQb2O6ifZB/wU+DlwqKpWk5wC3AesAPuAq6vqzdk0U9I0baTH//2q2l5Vq93yTcBjVXUu8Fi3LGkAJhnqX8WoWCbMsWhmjf0b4v1Lm0Hf4Bfwj0meTHJ9t+70qnq1u/4acPrUWydpJvrO3PtIVb2S5DeAR5M8O35jVVWSo3aR3QvF9QBnn332RI2VNB29evyqeqX7eQB4kFEFndeTnAHQ/Tywxu/eXlWrVbW6devW6bT68H33+DfJ9sctefeiucjYRevrUyb7vUl+7fB14A+AHwIPMSqWCRbNlAalz1D/dODBjHqvE4C/q6pHkjwB3J/kOuAl4OrZNVPSNK0b/K445oVHWf9j4LJZNGqapn50fqPD9z7bl58gHMtGh+99tm/9L+7MPalBBl9q0GBOxLFwszxCP37fcxr258657GZNdd3628zyCP34fbc47LfHlxpk8KUGGXypQQZfapDBlxpk8KUGGXypQQZfapATeI5lEV+rXcBkns1kEV+rbXEyjz2+1CCDLzXIof6xjA+15zXsb3B4P2780c9r2N/iX9weX2qQwZcaZPClBhl8qUEGX2pQr+AnOTnJA0meTbI3ycVJTknyaJLnu58fmHVjJU1H3x7/NuCRqjqf0Rl392LRTGmw+hTUeD/wu8CdAFX1P1X1FgsqmrkwVbP7jP3wfTf+Gf6Ritl9xl7M9v43uz49/jnAQeBrSZ5KckdXUceimdJA9Qn+CcCHgK9W1Q7gHY4Y1lfVmi+eSa5PsjvJ7oMHD07aXklT0Cf4+4H9VbWrW36A0QvBwotmLsT4sHytyyTb6/+pHpdJtm/RusGvqteAl5Oc1626DNiDRTOlwer7JZ0/A+5JchLwIvBpRi8aFs2UBqhX8KvqaWD1KDdt+qKZC+Hwfe78i2+MM/ekBhl8qUEGX2qQwZcaZPClBnnOvUb1qU+v5WWPLzXI4EsNMvhSgwy+1CCDLzXI4EsNMvhSgwy+1CCDLzXI4EsNMvhSgwy+1CCDLzXI4EsN6lNC67wkT49d3k5yo0UzpeHqc17956pqe1VtB34b+G/gQSyaKQ3WRof6lwE/qqqXaK1oprRENhr8a4Cvd9ctmikNVO/gd1V0rgT+/sjbLJopDctGevw/BL5XVa93y20WzZSWwEaCfy3vDvPBopnSYPUKfpL3AjuBb46t/gKwM8nzwEe7ZUkD0Ldo5jvAqUes+zEWzZQGyZl7UoMMvtQggy81yOBLDTL4UoMMvtQggy81yOBLDcro+zVz2llyEHgHeGNuO12c0/BxLpOhPM7frKp1vxQz1+ADJNldVatz3ekC+DiXy7I9Tof6UoMMvtSgRQT/9gXscxF8nMtlqR7n3N/jS1o8h/pSg+Ya/CSXJ3kuyQtJluZ03EnOSvJ4kj1JnklyQ7d+6WoPJNmS5KkkD3fL5yTZ1T2n93XnZhy8JCcneSDJs0n2Jrl4mZ7PuQU/yRbgbxidu+8C4NokF8xr/zN2CPhsVV0AXAR8pntsy1h74AZg79jyF4EvV9UHgTeB6xbSqum7DXikqs4HLmT0mJfn+ayquVyAi4Fvjy3fDNw8r/3P88Lo/IM7geeAM7p1ZwDPLbptEz6ubYz+w18KPAyE0aSWE472HA/1Arwf+E+6Y2Bj65fm+ZznUP9M4OWx5f3duqWSZAXYAexi+WoPfAX4HPCLbvlU4K2qOtQtL8tzeg5wEPha97bmju68k0vzfHpwb4qSvA/4BnBjVb09fluNuonBfoSS5ArgQFU9uei2zMEJwIeAr1bVDkbTzH9pWD/053OewX8FOGtseVu3bikkOZFR6O+pqsNnI+5Ve2AgLgGuTLIPuJfRcP824OQkh0/auizP6X5gf1Xt6pYfYPRCsDTP5zyD/wRwbncU+CRG5bgemuP+ZyZJgDuBvVX1pbGblqb2QFXdXFXbqmqF0XP3nar6BPA48PFus0E/xsOq6jXg5STndasuA/awRM/nvL+d9zFG7xO3AHdV1V/PbeczlOQjwD8DP+Dd97+fZ/Q+/37gbOAl4Oqq+slCGjlFSX4P+IuquiLJbzEaAZwCPAV8sqp+tsj2TUOS7cAdwEnAi8CnGXWUS/F8OnNPapAH96QGGXypQQZfapDBlxpk8KUGGXypQQZfapDBlxr0v+G9weaYQTg2AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "display_image(rel_train[20])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:23.136481Z",
     "start_time": "2018-07-16T04:07:23.133921Z"
    }
   },
   "outputs": [],
   "source": [
    "##setting hyper-parameters\n",
    "args = dict()\n",
    "args['batch_size'] = 128\n",
    "args['epochs'] = 50"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:07:23.689994Z",
     "start_time": "2018-07-16T04:07:23.684723Z"
    }
   },
   "outputs": [],
   "source": [
    "model = RN_Model()\n",
    "#parameter initialozation\n",
    "model.collect_params().initialize(mx.init.Xavier(magnitude=2.3), ctx=ctx)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n",
       "<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n",
       " \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n",
       "<!-- Generated by graphviz version 2.38.0 (20140413.2041)\n",
       " -->\n",
       "<!-- Title: plot Pages: 1 -->\n",
       "<svg width=\"351pt\" height=\"3967pt\"\n",
       " viewBox=\"0.00 0.00 351.00 3967.00\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g id=\"graph0\" class=\"graph\" transform=\"scale(1 1) rotate(0) translate(4 3963)\">\n",
       "<title>plot</title>\n",
       "<polygon fill=\"white\" stroke=\"none\" points=\"-4,4 -4,-3963 347,-3963 347,4 -4,4\"/>\n",
       "<!-- data -->\n",
       "<g id=\"node1\" class=\"node\"><title>data</title>\n",
       "<ellipse fill=\"#8dd3c7\" stroke=\"black\" cx=\"47\" cy=\"-29\" rx=\"47\" ry=\"29\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-25.3\" font-family=\"Times,serif\" font-size=\"14.00\">data</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv0_fwd -->\n",
       "<g id=\"node2\" class=\"node\"><title>rn_model2_convinputmodel0_conv0_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"94,-167 -7.10543e-15,-167 -7.10543e-15,-109 94,-109 94,-167\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-141.8\" font-family=\"Times,serif\" font-size=\"14.00\">Convolution</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-126.8\" font-family=\"Times,serif\" font-size=\"14.00\">3x3/2x2, 24</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv0_fwd&#45;&gt;data -->\n",
       "<g id=\"edge1\" class=\"edge\"><title>rn_model2_convinputmodel0_conv0_fwd&#45;&gt;data</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-98.5824C47,-85.2841 47,-70.632 47,-58.2967\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-108.887 42.5001,-98.887 47,-103.887 47.0001,-98.887 47.0001,-98.887 47.0001,-98.887 47,-103.887 51.5001,-98.8871 47,-108.887 47,-108.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"71\" y=\"-79.8\" font-family=\"Times,serif\" font-size=\"14.00\">3x75x75</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv0_relu_fwd -->\n",
       "<g id=\"node3\" class=\"node\"><title>rn_model2_convinputmodel0_conv0_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"94,-276 -7.10543e-15,-276 -7.10543e-15,-218 94,-218 94,-276\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-250.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-235.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv0_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv0_fwd -->\n",
       "<g id=\"edge2\" class=\"edge\"><title>rn_model2_convinputmodel0_conv0_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv0_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-207.582C47,-194.284 47,-179.632 47,-167.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-217.887 42.5001,-207.887 47,-212.887 47.0001,-207.887 47.0001,-207.887 47.0001,-207.887 47,-212.887 51.5001,-207.887 47,-217.887 47,-217.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-188.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x38x38</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm0_fwd -->\n",
       "<g id=\"node4\" class=\"node\"><title>rn_model2_convinputmodel0_batchnorm0_fwd</title>\n",
       "<polygon fill=\"#bebada\" stroke=\"black\" points=\"94,-385 -7.10543e-15,-385 -7.10543e-15,-327 94,-327 94,-385\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-352.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_convinputmodel0_batchnorm0_fwd</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm0_fwd&#45;&gt;rn_model2_convinputmodel0_conv0_relu_fwd -->\n",
       "<g id=\"edge3\" class=\"edge\"><title>rn_model2_convinputmodel0_batchnorm0_fwd&#45;&gt;rn_model2_convinputmodel0_conv0_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-316.582C47,-303.284 47,-288.632 47,-276.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-326.887 42.5001,-316.887 47,-321.887 47.0001,-316.887 47.0001,-316.887 47.0001,-316.887 47,-321.887 51.5001,-316.887 47,-326.887 47,-326.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-297.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x38x38</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv1_fwd -->\n",
       "<g id=\"node5\" class=\"node\"><title>rn_model2_convinputmodel0_conv1_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"94,-494 -7.10543e-15,-494 -7.10543e-15,-436 94,-436 94,-494\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-468.8\" font-family=\"Times,serif\" font-size=\"14.00\">Convolution</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-453.8\" font-family=\"Times,serif\" font-size=\"14.00\">3x3/2x2, 24</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv1_fwd&#45;&gt;rn_model2_convinputmodel0_batchnorm0_fwd -->\n",
       "<g id=\"edge4\" class=\"edge\"><title>rn_model2_convinputmodel0_conv1_fwd&#45;&gt;rn_model2_convinputmodel0_batchnorm0_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-425.582C47,-412.284 47,-397.632 47,-385.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-435.887 42.5001,-425.887 47,-430.887 47.0001,-425.887 47.0001,-425.887 47.0001,-425.887 47,-430.887 51.5001,-425.887 47,-435.887 47,-435.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-406.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x38x38</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv1_relu_fwd -->\n",
       "<g id=\"node6\" class=\"node\"><title>rn_model2_convinputmodel0_conv1_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"94,-603 -7.10543e-15,-603 -7.10543e-15,-545 94,-545 94,-603\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-577.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-562.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv1_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv1_fwd -->\n",
       "<g id=\"edge5\" class=\"edge\"><title>rn_model2_convinputmodel0_conv1_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv1_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-534.582C47,-521.284 47,-506.632 47,-494.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-544.887 42.5001,-534.887 47,-539.887 47.0001,-534.887 47.0001,-534.887 47.0001,-534.887 47,-539.887 51.5001,-534.887 47,-544.887 47,-544.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-515.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x19x19</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm1_fwd -->\n",
       "<g id=\"node7\" class=\"node\"><title>rn_model2_convinputmodel0_batchnorm1_fwd</title>\n",
       "<polygon fill=\"#bebada\" stroke=\"black\" points=\"94,-712 -7.10543e-15,-712 -7.10543e-15,-654 94,-654 94,-712\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-679.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_convinputmodel0_batchnorm1_fwd</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm1_fwd&#45;&gt;rn_model2_convinputmodel0_conv1_relu_fwd -->\n",
       "<g id=\"edge6\" class=\"edge\"><title>rn_model2_convinputmodel0_batchnorm1_fwd&#45;&gt;rn_model2_convinputmodel0_conv1_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-643.582C47,-630.284 47,-615.632 47,-603.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-653.887 42.5001,-643.887 47,-648.887 47.0001,-643.887 47.0001,-643.887 47.0001,-643.887 47,-648.887 51.5001,-643.887 47,-653.887 47,-653.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-624.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x19x19</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv2_fwd -->\n",
       "<g id=\"node8\" class=\"node\"><title>rn_model2_convinputmodel0_conv2_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"94,-821 -7.10543e-15,-821 -7.10543e-15,-763 94,-763 94,-821\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-795.8\" font-family=\"Times,serif\" font-size=\"14.00\">Convolution</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-780.8\" font-family=\"Times,serif\" font-size=\"14.00\">3x3/2x2, 24</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv2_fwd&#45;&gt;rn_model2_convinputmodel0_batchnorm1_fwd -->\n",
       "<g id=\"edge7\" class=\"edge\"><title>rn_model2_convinputmodel0_conv2_fwd&#45;&gt;rn_model2_convinputmodel0_batchnorm1_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-752.582C47,-739.284 47,-724.632 47,-712.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-762.887 42.5001,-752.887 47,-757.887 47.0001,-752.887 47.0001,-752.887 47.0001,-752.887 47,-757.887 51.5001,-752.887 47,-762.887 47,-762.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-733.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x19x19</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv2_relu_fwd -->\n",
       "<g id=\"node9\" class=\"node\"><title>rn_model2_convinputmodel0_conv2_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"94,-930 -7.10543e-15,-930 -7.10543e-15,-872 94,-872 94,-930\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-904.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-889.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv2_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv2_fwd -->\n",
       "<g id=\"edge8\" class=\"edge\"><title>rn_model2_convinputmodel0_conv2_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv2_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-861.582C47,-848.284 47,-833.632 47,-821.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-871.887 42.5001,-861.887 47,-866.887 47.0001,-861.887 47.0001,-861.887 47.0001,-861.887 47,-866.887 51.5001,-861.887 47,-871.887 47,-871.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-842.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x10x10</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm2_fwd -->\n",
       "<g id=\"node10\" class=\"node\"><title>rn_model2_convinputmodel0_batchnorm2_fwd</title>\n",
       "<polygon fill=\"#bebada\" stroke=\"black\" points=\"94,-1039 -7.10543e-15,-1039 -7.10543e-15,-981 94,-981 94,-1039\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1006.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_convinputmodel0_batchnorm2_fwd</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm2_fwd&#45;&gt;rn_model2_convinputmodel0_conv2_relu_fwd -->\n",
       "<g id=\"edge9\" class=\"edge\"><title>rn_model2_convinputmodel0_batchnorm2_fwd&#45;&gt;rn_model2_convinputmodel0_conv2_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-970.582C47,-957.284 47,-942.632 47,-930.297\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-980.887 42.5001,-970.887 47,-975.887 47.0001,-970.887 47.0001,-970.887 47.0001,-970.887 47,-975.887 51.5001,-970.887 47,-980.887 47,-980.887\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-951.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x10x10</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv3_fwd -->\n",
       "<g id=\"node11\" class=\"node\"><title>rn_model2_convinputmodel0_conv3_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"94,-1148 -7.10543e-15,-1148 -7.10543e-15,-1090 94,-1090 94,-1148\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1122.8\" font-family=\"Times,serif\" font-size=\"14.00\">Convolution</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1107.8\" font-family=\"Times,serif\" font-size=\"14.00\">3x3/2x2, 24</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv3_fwd&#45;&gt;rn_model2_convinputmodel0_batchnorm2_fwd -->\n",
       "<g id=\"edge10\" class=\"edge\"><title>rn_model2_convinputmodel0_conv3_fwd&#45;&gt;rn_model2_convinputmodel0_batchnorm2_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-1079.58C47,-1066.28 47,-1051.63 47,-1039.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-1089.89 42.5001,-1079.89 47,-1084.89 47.0001,-1079.89 47.0001,-1079.89 47.0001,-1079.89 47,-1084.89 51.5001,-1079.89 47,-1089.89 47,-1089.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"74.5\" y=\"-1060.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x10x10</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv3_relu_fwd -->\n",
       "<g id=\"node12\" class=\"node\"><title>rn_model2_convinputmodel0_conv3_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"94,-1257 -7.10543e-15,-1257 -7.10543e-15,-1199 94,-1199 94,-1257\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1231.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1216.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_conv3_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv3_fwd -->\n",
       "<g id=\"edge11\" class=\"edge\"><title>rn_model2_convinputmodel0_conv3_relu_fwd&#45;&gt;rn_model2_convinputmodel0_conv3_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-1188.58C47,-1175.28 47,-1160.63 47,-1148.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-1198.89 42.5001,-1188.89 47,-1193.89 47.0001,-1188.89 47.0001,-1188.89 47.0001,-1188.89 47,-1193.89 51.5001,-1188.89 47,-1198.89 47,-1198.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"67.5\" y=\"-1169.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x5x5</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm3_fwd -->\n",
       "<g id=\"node13\" class=\"node\"><title>rn_model2_convinputmodel0_batchnorm3_fwd</title>\n",
       "<polygon fill=\"#bebada\" stroke=\"black\" points=\"94,-1366 -7.10543e-15,-1366 -7.10543e-15,-1308 94,-1308 94,-1366\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1333.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_convinputmodel0_batchnorm3_fwd</text>\n",
       "</g>\n",
       "<!-- rn_model2_convinputmodel0_batchnorm3_fwd&#45;&gt;rn_model2_convinputmodel0_conv3_relu_fwd -->\n",
       "<g id=\"edge12\" class=\"edge\"><title>rn_model2_convinputmodel0_batchnorm3_fwd&#45;&gt;rn_model2_convinputmodel0_conv3_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-1297.58C47,-1284.28 47,-1269.63 47,-1257.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-1307.89 42.5001,-1297.89 47,-1302.89 47.0001,-1297.89 47.0001,-1297.89 47.0001,-1297.89 47,-1302.89 51.5001,-1297.89 47,-1307.89 47,-1307.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"67.5\" y=\"-1278.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x5x5</text>\n",
       "</g>\n",
       "<!-- rn_model2_reshape0 -->\n",
       "<g id=\"node14\" class=\"node\"><title>rn_model2_reshape0</title>\n",
       "<polygon fill=\"#fdb462\" stroke=\"black\" points=\"94,-1475 -7.10543e-15,-1475 -7.10543e-15,-1417 94,-1417 94,-1475\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1442.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_reshape0</text>\n",
       "</g>\n",
       "<!-- rn_model2_reshape0&#45;&gt;rn_model2_convinputmodel0_batchnorm3_fwd -->\n",
       "<g id=\"edge13\" class=\"edge\"><title>rn_model2_reshape0&#45;&gt;rn_model2_convinputmodel0_batchnorm3_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-1406.58C47,-1393.28 47,-1378.63 47,-1366.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-1416.89 42.5001,-1406.89 47,-1411.89 47.0001,-1406.89 47.0001,-1406.89 47.0001,-1406.89 47,-1411.89 51.5001,-1406.89 47,-1416.89 47,-1416.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"67.5\" y=\"-1387.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x5x5</text>\n",
       "</g>\n",
       "<!-- rn_model2_swapaxes0 -->\n",
       "<g id=\"node15\" class=\"node\"><title>rn_model2_swapaxes0</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"94,-1584 -7.10543e-15,-1584 -7.10543e-15,-1526 94,-1526 94,-1584\"/>\n",
       "<text text-anchor=\"middle\" x=\"47\" y=\"-1551.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_swapaxes0</text>\n",
       "</g>\n",
       "<!-- rn_model2_swapaxes0&#45;&gt;rn_model2_reshape0 -->\n",
       "<g id=\"edge14\" class=\"edge\"><title>rn_model2_swapaxes0&#45;&gt;rn_model2_reshape0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M47,-1515.58C47,-1502.28 47,-1487.63 47,-1475.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"47,-1525.89 42.5001,-1515.89 47,-1520.89 47.0001,-1515.89 47.0001,-1515.89 47.0001,-1515.89 47,-1520.89 51.5001,-1515.89 47,-1525.89 47,-1525.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"64\" y=\"-1496.8\" font-family=\"Times,serif\" font-size=\"14.00\">24x25</text>\n",
       "</g>\n",
       "<!-- coord_tensor -->\n",
       "<g id=\"node16\" class=\"node\"><title>coord_tensor</title>\n",
       "<ellipse fill=\"#8dd3c7\" stroke=\"black\" cx=\"159\" cy=\"-1555\" rx=\"47\" ry=\"29\"/>\n",
       "<text text-anchor=\"middle\" x=\"159\" y=\"-1551.3\" font-family=\"Times,serif\" font-size=\"14.00\">coord_tensor</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat0 -->\n",
       "<g id=\"node17\" class=\"node\"><title>rn_model2_concat0</title>\n",
       "<polygon fill=\"#fdb462\" stroke=\"black\" points=\"181,-1693 87,-1693 87,-1635 181,-1635 181,-1693\"/>\n",
       "<text text-anchor=\"middle\" x=\"134\" y=\"-1660.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_concat0</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat0&#45;&gt;rn_model2_swapaxes0 -->\n",
       "<g id=\"edge15\" class=\"edge\"><title>rn_model2_concat0&#45;&gt;rn_model2_swapaxes0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M104.678,-1626.94C93.3728,-1613.03 80.633,-1597.36 70.0076,-1584.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"111.142,-1634.89 101.342,-1629.97 107.987,-1631.01 104.833,-1627.13 104.833,-1627.13 104.833,-1627.13 107.987,-1631.01 108.325,-1624.29 111.142,-1634.89 111.142,-1634.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"113\" y=\"-1605.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x24</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat0&#45;&gt;coord_tensor -->\n",
       "<g id=\"edge16\" class=\"edge\"><title>rn_model2_concat0&#45;&gt;coord_tensor</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M142.903,-1624.9C146.061,-1611.38 149.553,-1596.43 152.475,-1583.93\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"140.568,-1634.89 138.462,-1624.13 141.706,-1630.02 142.844,-1625.15 142.844,-1625.15 142.844,-1625.15 141.706,-1630.02 147.226,-1626.17 140.568,-1634.89 140.568,-1634.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"161\" y=\"-1605.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x2</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims2 -->\n",
       "<g id=\"node18\" class=\"node\"><title>rn_model2_expand_dims2</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"132,-1878 38,-1878 38,-1820 132,-1820 132,-1878\"/>\n",
       "<text text-anchor=\"middle\" x=\"85\" y=\"-1845.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_expand_dims2</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims2&#45;&gt;rn_model2_concat0 -->\n",
       "<g id=\"edge17\" class=\"edge\"><title>rn_model2_expand_dims2&#45;&gt;rn_model2_concat0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M84.8525,-1809.44C85.6448,-1789.56 87.9816,-1765.13 94,-1744 99.054,-1726.26 108.159,-1707.79 116.407,-1693.18\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"84.5625,-1819.67 80.3478,-1809.54 84.7043,-1814.67 84.846,-1809.67 84.846,-1809.67 84.846,-1809.67 84.7043,-1814.67 89.3442,-1809.8 84.5625,-1819.67 84.5625,-1819.67\"/>\n",
       "<text text-anchor=\"middle\" x=\"111\" y=\"-1769.3\" font-family=\"Times,serif\" font-size=\"14.00\">25x26</text>\n",
       "</g>\n",
       "<!-- rn_model2_repeat1 -->\n",
       "<g id=\"node19\" class=\"node\"><title>rn_model2_repeat1</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"133,-2030 39,-2030 39,-1972 133,-1972 133,-2030\"/>\n",
       "<text text-anchor=\"middle\" x=\"86\" y=\"-1997.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_repeat1</text>\n",
       "</g>\n",
       "<!-- rn_model2_repeat1&#45;&gt;rn_model2_expand_dims2 -->\n",
       "<g id=\"edge18\" class=\"edge\"><title>rn_model2_repeat1&#45;&gt;rn_model2_expand_dims2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M85.7434,-1961.51C85.5697,-1935.45 85.3441,-1901.62 85.1872,-1878.09\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"85.8119,-1971.79 81.2452,-1961.82 85.7785,-1966.79 85.7451,-1961.79 85.7451,-1961.79 85.7451,-1961.79 85.7785,-1966.79 90.245,-1961.76 85.8119,-1971.79 85.8119,-1971.79\"/>\n",
       "<text text-anchor=\"middle\" x=\"110\" y=\"-1921.3\" font-family=\"Times,serif\" font-size=\"14.00\">1x25x26</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims3 -->\n",
       "<g id=\"node20\" class=\"node\"><title>rn_model2_expand_dims3</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"231,-1802 137,-1802 137,-1744 231,-1744 231,-1802\"/>\n",
       "<text text-anchor=\"middle\" x=\"184\" y=\"-1769.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_expand_dims3</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims3&#45;&gt;rn_model2_concat0 -->\n",
       "<g id=\"edge19\" class=\"edge\"><title>rn_model2_expand_dims3&#45;&gt;rn_model2_concat0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M166.6,-1734.76C160.241,-1721.16 153.158,-1706 147.223,-1693.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"170.863,-1743.89 162.553,-1736.73 168.746,-1739.36 166.63,-1734.83 166.63,-1734.83 166.63,-1734.83 168.746,-1739.36 170.706,-1732.92 170.863,-1743.89 170.863,-1743.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"178\" y=\"-1714.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x26</text>\n",
       "</g>\n",
       "<!-- qst -->\n",
       "<g id=\"node21\" class=\"node\"><title>qst</title>\n",
       "<ellipse fill=\"#8dd3c7\" stroke=\"black\" cx=\"296\" cy=\"-1446\" rx=\"47\" ry=\"29\"/>\n",
       "<text text-anchor=\"middle\" x=\"296\" y=\"-1442.3\" font-family=\"Times,serif\" font-size=\"14.00\">qst</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims0 -->\n",
       "<g id=\"node22\" class=\"node\"><title>rn_model2_expand_dims0</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"343,-1584 249,-1584 249,-1526 343,-1526 343,-1584\"/>\n",
       "<text text-anchor=\"middle\" x=\"296\" y=\"-1551.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_expand_dims0</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims0&#45;&gt;qst -->\n",
       "<g id=\"edge20\" class=\"edge\"><title>rn_model2_expand_dims0&#45;&gt;qst</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M296,-1515.58C296,-1502.28 296,-1487.63 296,-1475.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"296,-1525.89 291.5,-1515.89 296,-1520.89 296,-1515.89 296,-1515.89 296,-1515.89 296,-1520.89 300.5,-1515.89 296,-1525.89 296,-1525.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"303\" y=\"-1496.8\" font-family=\"Times,serif\" font-size=\"14.00\">11</text>\n",
       "</g>\n",
       "<!-- rn_model2_repeat0 -->\n",
       "<g id=\"node23\" class=\"node\"><title>rn_model2_repeat0</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"343,-1693 249,-1693 249,-1635 343,-1635 343,-1693\"/>\n",
       "<text text-anchor=\"middle\" x=\"296\" y=\"-1660.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_repeat0</text>\n",
       "</g>\n",
       "<!-- rn_model2_repeat0&#45;&gt;rn_model2_expand_dims0 -->\n",
       "<g id=\"edge21\" class=\"edge\"><title>rn_model2_repeat0&#45;&gt;rn_model2_expand_dims0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M296,-1624.58C296,-1611.28 296,-1596.63 296,-1584.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"296,-1634.89 291.5,-1624.89 296,-1629.89 296,-1624.89 296,-1624.89 296,-1624.89 296,-1629.89 300.5,-1624.89 296,-1634.89 296,-1634.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"310\" y=\"-1605.8\" font-family=\"Times,serif\" font-size=\"14.00\">1x11</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims1 -->\n",
       "<g id=\"node24\" class=\"node\"><title>rn_model2_expand_dims1</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"343,-1802 249,-1802 249,-1744 343,-1744 343,-1802\"/>\n",
       "<text text-anchor=\"middle\" x=\"296\" y=\"-1769.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_expand_dims1</text>\n",
       "</g>\n",
       "<!-- rn_model2_expand_dims1&#45;&gt;rn_model2_repeat0 -->\n",
       "<g id=\"edge22\" class=\"edge\"><title>rn_model2_expand_dims1&#45;&gt;rn_model2_repeat0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M296,-1733.58C296,-1720.28 296,-1705.63 296,-1693.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"296,-1743.89 291.5,-1733.89 296,-1738.89 296,-1733.89 296,-1733.89 296,-1733.89 296,-1738.89 300.5,-1733.89 296,-1743.89 296,-1743.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"313\" y=\"-1714.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x11</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat1 -->\n",
       "<g id=\"node25\" class=\"node\"><title>rn_model2_concat1</title>\n",
       "<polygon fill=\"#fdb462\" stroke=\"black\" points=\"245,-1954 151,-1954 151,-1896 245,-1896 245,-1954\"/>\n",
       "<text text-anchor=\"middle\" x=\"198\" y=\"-1921.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_concat1</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat1&#45;&gt;rn_model2_expand_dims3 -->\n",
       "<g id=\"edge23\" class=\"edge\"><title>rn_model2_concat1&#45;&gt;rn_model2_expand_dims3</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M194.407,-1885.51C191.975,-1859.45 188.818,-1825.62 186.621,-1802.09\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"195.367,-1895.79 189.957,-1886.25 194.902,-1890.81 194.437,-1885.83 194.437,-1885.83 194.437,-1885.83 194.902,-1890.81 198.918,-1885.41 195.367,-1895.79 195.367,-1895.79\"/>\n",
       "<text text-anchor=\"middle\" x=\"217\" y=\"-1845.3\" font-family=\"Times,serif\" font-size=\"14.00\">25x1x26</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat1&#45;&gt;rn_model2_expand_dims1 -->\n",
       "<g id=\"edge24\" class=\"edge\"><title>rn_model2_concat1&#45;&gt;rn_model2_expand_dims1</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M236.498,-1888.58C239.526,-1885.14 242.409,-1881.58 245,-1878 262.085,-1854.37 276.178,-1824.14 285.187,-1802.37\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"229.698,-1895.94 233.182,-1885.54 233.092,-1892.27 236.487,-1888.6 236.487,-1888.6 236.487,-1888.6 233.092,-1892.27 239.791,-1891.65 229.698,-1895.94 229.698,-1895.94\"/>\n",
       "<text text-anchor=\"middle\" x=\"301\" y=\"-1845.3\" font-family=\"Times,serif\" font-size=\"14.00\">25x1x11</text>\n",
       "</g>\n",
       "<!-- rn_model2_repeat2 -->\n",
       "<g id=\"node26\" class=\"node\"><title>rn_model2_repeat2</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"241,-2106 147,-2106 147,-2048 241,-2048 241,-2106\"/>\n",
       "<text text-anchor=\"middle\" x=\"194\" y=\"-2073.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_repeat2</text>\n",
       "</g>\n",
       "<!-- rn_model2_repeat2&#45;&gt;rn_model2_concat1 -->\n",
       "<g id=\"edge25\" class=\"edge\"><title>rn_model2_repeat2&#45;&gt;rn_model2_concat1</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M195.026,-2037.51C195.721,-2011.45 196.623,-1977.62 197.251,-1954.09\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"194.752,-2047.79 190.521,-2037.67 194.886,-2042.79 195.019,-2037.79 195.019,-2037.79 195.019,-2037.79 194.886,-2042.79 199.517,-2037.91 194.752,-2047.79 194.752,-2047.79\"/>\n",
       "<text text-anchor=\"middle\" x=\"220\" y=\"-1997.3\" font-family=\"Times,serif\" font-size=\"14.00\">25x1x37</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat2 -->\n",
       "<g id=\"node27\" class=\"node\"><title>rn_model2_concat2</title>\n",
       "<polygon fill=\"#fdb462\" stroke=\"black\" points=\"176,-2215 82,-2215 82,-2157 176,-2157 176,-2215\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2182.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_concat2</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat2&#45;&gt;rn_model2_repeat1 -->\n",
       "<g id=\"edge26\" class=\"edge\"><title>rn_model2_concat2&#45;&gt;rn_model2_repeat1</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M101.754,-2147.77C100.319,-2144.88 99.0428,-2141.95 98,-2139 85.2537,-2102.99 83.7528,-2058.48 84.4195,-2030.16\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"106.779,-2156.75 97.9684,-2150.22 104.337,-2152.39 101.895,-2148.02 101.895,-2148.02 101.895,-2148.02 104.337,-2152.39 105.822,-2145.83 106.779,-2156.75 106.779,-2156.75\"/>\n",
       "<text text-anchor=\"middle\" x=\"125.5\" y=\"-2127.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x25x26</text>\n",
       "</g>\n",
       "<!-- rn_model2_concat2&#45;&gt;rn_model2_repeat2 -->\n",
       "<g id=\"edge27\" class=\"edge\"><title>rn_model2_concat2&#45;&gt;rn_model2_repeat2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M151.382,-2148.16C159.709,-2134.45 169.02,-2119.12 176.81,-2106.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"146.078,-2156.89 147.424,-2146 148.674,-2152.61 151.27,-2148.34 151.27,-2148.34 151.27,-2148.34 148.674,-2152.61 155.116,-2150.68 146.078,-2156.89 146.078,-2156.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"192.5\" y=\"-2127.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x25x37</text>\n",
       "</g>\n",
       "<!-- rn_model2_reshape1 -->\n",
       "<g id=\"node28\" class=\"node\"><title>rn_model2_reshape1</title>\n",
       "<polygon fill=\"#fdb462\" stroke=\"black\" points=\"176,-2324 82,-2324 82,-2266 176,-2266 176,-2324\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2291.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_reshape1</text>\n",
       "</g>\n",
       "<!-- rn_model2_reshape1&#45;&gt;rn_model2_concat2 -->\n",
       "<g id=\"edge28\" class=\"edge\"><title>rn_model2_reshape1&#45;&gt;rn_model2_concat2</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2255.58C129,-2242.28 129,-2227.63 129,-2215.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2265.89 124.5,-2255.89 129,-2260.89 129,-2255.89 129,-2255.89 129,-2255.89 129,-2260.89 133.5,-2255.89 129,-2265.89 129,-2265.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"156.5\" y=\"-2236.8\" font-family=\"Times,serif\" font-size=\"14.00\">25x25x63</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense0_fwd -->\n",
       "<g id=\"node29\" class=\"node\"><title>rn_model2_dense0_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-2433 82,-2433 82,-2375 176,-2375 176,-2433\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2407.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2392.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense0_fwd&#45;&gt;rn_model2_reshape1 -->\n",
       "<g id=\"edge29\" class=\"edge\"><title>rn_model2_dense0_fwd&#45;&gt;rn_model2_reshape1</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2364.58C129,-2351.28 129,-2336.63 129,-2324.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2374.89 124.5,-2364.89 129,-2369.89 129,-2364.89 129,-2364.89 129,-2364.89 129,-2369.89 133.5,-2364.89 129,-2374.89 129,-2374.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"149.5\" y=\"-2345.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x63</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense0_relu_fwd -->\n",
       "<g id=\"node30\" class=\"node\"><title>rn_model2_dense0_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"176,-2542 82,-2542 82,-2484 176,-2484 176,-2542\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2516.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2501.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense0_relu_fwd&#45;&gt;rn_model2_dense0_fwd -->\n",
       "<g id=\"edge30\" class=\"edge\"><title>rn_model2_dense0_relu_fwd&#45;&gt;rn_model2_dense0_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2473.58C129,-2460.28 129,-2445.63 129,-2433.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2483.89 124.5,-2473.89 129,-2478.89 129,-2473.89 129,-2473.89 129,-2473.89 129,-2478.89 133.5,-2473.89 129,-2483.89 129,-2483.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-2454.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense1_fwd -->\n",
       "<g id=\"node31\" class=\"node\"><title>rn_model2_dense1_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-2651 82,-2651 82,-2593 176,-2593 176,-2651\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2625.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2610.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense1_fwd&#45;&gt;rn_model2_dense0_relu_fwd -->\n",
       "<g id=\"edge31\" class=\"edge\"><title>rn_model2_dense1_fwd&#45;&gt;rn_model2_dense0_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2582.58C129,-2569.28 129,-2554.63 129,-2542.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2592.89 124.5,-2582.89 129,-2587.89 129,-2582.89 129,-2582.89 129,-2582.89 129,-2587.89 133.5,-2582.89 129,-2592.89 129,-2592.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-2563.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense1_relu_fwd -->\n",
       "<g id=\"node32\" class=\"node\"><title>rn_model2_dense1_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"176,-2760 82,-2760 82,-2702 176,-2702 176,-2760\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2734.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2719.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense1_relu_fwd&#45;&gt;rn_model2_dense1_fwd -->\n",
       "<g id=\"edge32\" class=\"edge\"><title>rn_model2_dense1_relu_fwd&#45;&gt;rn_model2_dense1_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2691.58C129,-2678.28 129,-2663.63 129,-2651.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2701.89 124.5,-2691.89 129,-2696.89 129,-2691.89 129,-2691.89 129,-2691.89 129,-2696.89 133.5,-2691.89 129,-2701.89 129,-2701.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-2672.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense2_fwd -->\n",
       "<g id=\"node33\" class=\"node\"><title>rn_model2_dense2_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-2869 82,-2869 82,-2811 176,-2811 176,-2869\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2843.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2828.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense2_fwd&#45;&gt;rn_model2_dense1_relu_fwd -->\n",
       "<g id=\"edge33\" class=\"edge\"><title>rn_model2_dense2_fwd&#45;&gt;rn_model2_dense1_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2800.58C129,-2787.28 129,-2772.63 129,-2760.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2810.89 124.5,-2800.89 129,-2805.89 129,-2800.89 129,-2800.89 129,-2800.89 129,-2805.89 133.5,-2800.89 129,-2810.89 129,-2810.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-2781.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense2_relu_fwd -->\n",
       "<g id=\"node34\" class=\"node\"><title>rn_model2_dense2_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"176,-2978 82,-2978 82,-2920 176,-2920 176,-2978\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2952.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-2937.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense2_relu_fwd&#45;&gt;rn_model2_dense2_fwd -->\n",
       "<g id=\"edge34\" class=\"edge\"><title>rn_model2_dense2_relu_fwd&#45;&gt;rn_model2_dense2_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-2909.58C129,-2896.28 129,-2881.63 129,-2869.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-2919.89 124.5,-2909.89 129,-2914.89 129,-2909.89 129,-2909.89 129,-2909.89 129,-2914.89 133.5,-2909.89 129,-2919.89 129,-2919.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-2890.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense3_fwd -->\n",
       "<g id=\"node35\" class=\"node\"><title>rn_model2_dense3_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-3087 82,-3087 82,-3029 176,-3029 176,-3087\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3061.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3046.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense3_fwd&#45;&gt;rn_model2_dense2_relu_fwd -->\n",
       "<g id=\"edge35\" class=\"edge\"><title>rn_model2_dense3_fwd&#45;&gt;rn_model2_dense2_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3018.58C129,-3005.28 129,-2990.63 129,-2978.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3028.89 124.5,-3018.89 129,-3023.89 129,-3018.89 129,-3018.89 129,-3018.89 129,-3023.89 133.5,-3018.89 129,-3028.89 129,-3028.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-2999.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense3_relu_fwd -->\n",
       "<g id=\"node36\" class=\"node\"><title>rn_model2_dense3_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"176,-3196 82,-3196 82,-3138 176,-3138 176,-3196\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3170.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3155.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense3_relu_fwd&#45;&gt;rn_model2_dense3_fwd -->\n",
       "<g id=\"edge36\" class=\"edge\"><title>rn_model2_dense3_relu_fwd&#45;&gt;rn_model2_dense3_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3127.58C129,-3114.28 129,-3099.63 129,-3087.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3137.89 124.5,-3127.89 129,-3132.89 129,-3127.89 129,-3127.89 129,-3127.89 129,-3132.89 133.5,-3127.89 129,-3137.89 129,-3137.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-3108.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_sum0 -->\n",
       "<g id=\"node37\" class=\"node\"><title>rn_model2_sum0</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"176,-3305 82,-3305 82,-3247 176,-3247 176,-3305\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3272.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_sum0</text>\n",
       "</g>\n",
       "<!-- rn_model2_sum0&#45;&gt;rn_model2_dense3_relu_fwd -->\n",
       "<g id=\"edge37\" class=\"edge\"><title>rn_model2_sum0&#45;&gt;rn_model2_dense3_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3236.58C129,-3223.28 129,-3208.63 129,-3196.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3246.89 124.5,-3236.89 129,-3241.89 129,-3236.89 129,-3236.89 129,-3236.89 129,-3241.89 133.5,-3236.89 129,-3246.89 129,-3246.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"153\" y=\"-3217.8\" font-family=\"Times,serif\" font-size=\"14.00\">625x256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense4_fwd -->\n",
       "<g id=\"node38\" class=\"node\"><title>rn_model2_dense4_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-3414 82,-3414 82,-3356 176,-3356 176,-3414\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3388.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3373.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense4_fwd&#45;&gt;rn_model2_sum0 -->\n",
       "<g id=\"edge38\" class=\"edge\"><title>rn_model2_dense4_fwd&#45;&gt;rn_model2_sum0</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3345.58C129,-3332.28 129,-3317.63 129,-3305.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3355.89 124.5,-3345.89 129,-3350.89 129,-3345.89 129,-3345.89 129,-3345.89 129,-3350.89 133.5,-3345.89 129,-3355.89 129,-3355.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"139.5\" y=\"-3326.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense4_relu_fwd -->\n",
       "<g id=\"node39\" class=\"node\"><title>rn_model2_dense4_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"176,-3523 82,-3523 82,-3465 176,-3465 176,-3523\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3497.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3482.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_dense4_relu_fwd&#45;&gt;rn_model2_dense4_fwd -->\n",
       "<g id=\"edge39\" class=\"edge\"><title>rn_model2_dense4_relu_fwd&#45;&gt;rn_model2_dense4_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3454.58C129,-3441.28 129,-3426.63 129,-3414.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3464.89 124.5,-3454.89 129,-3459.89 129,-3454.89 129,-3454.89 129,-3454.89 129,-3459.89 133.5,-3454.89 129,-3464.89 129,-3464.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"139.5\" y=\"-3435.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dense0_fwd -->\n",
       "<g id=\"node40\" class=\"node\"><title>rn_model2_fcoutputmodel0_dense0_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-3632 82,-3632 82,-3574 176,-3574 176,-3632\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3606.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3591.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dense0_fwd&#45;&gt;rn_model2_dense4_relu_fwd -->\n",
       "<g id=\"edge40\" class=\"edge\"><title>rn_model2_fcoutputmodel0_dense0_fwd&#45;&gt;rn_model2_dense4_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3563.58C129,-3550.28 129,-3535.63 129,-3523.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3573.89 124.5,-3563.89 129,-3568.89 129,-3563.89 129,-3563.89 129,-3563.89 129,-3568.89 133.5,-3563.89 129,-3573.89 129,-3573.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"139.5\" y=\"-3544.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dense0_relu_fwd -->\n",
       "<g id=\"node41\" class=\"node\"><title>rn_model2_fcoutputmodel0_dense0_relu_fwd</title>\n",
       "<polygon fill=\"#ffffb3\" stroke=\"black\" points=\"176,-3741 82,-3741 82,-3683 176,-3683 176,-3741\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3715.8\" font-family=\"Times,serif\" font-size=\"14.00\">Activation</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3700.8\" font-family=\"Times,serif\" font-size=\"14.00\">relu</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dense0_relu_fwd&#45;&gt;rn_model2_fcoutputmodel0_dense0_fwd -->\n",
       "<g id=\"edge41\" class=\"edge\"><title>rn_model2_fcoutputmodel0_dense0_relu_fwd&#45;&gt;rn_model2_fcoutputmodel0_dense0_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3672.58C129,-3659.28 129,-3644.63 129,-3632.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3682.89 124.5,-3672.89 129,-3677.89 129,-3672.89 129,-3672.89 129,-3672.89 129,-3677.89 133.5,-3672.89 129,-3682.89 129,-3682.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"139.5\" y=\"-3653.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dropout0_fwd -->\n",
       "<g id=\"node42\" class=\"node\"><title>rn_model2_fcoutputmodel0_dropout0_fwd</title>\n",
       "<polygon fill=\"#fccde5\" stroke=\"black\" points=\"176,-3850 82,-3850 82,-3792 176,-3792 176,-3850\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3817.3\" font-family=\"Times,serif\" font-size=\"14.00\">rn_model2_fcoutputmodel0_dropout0_fwd</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dropout0_fwd&#45;&gt;rn_model2_fcoutputmodel0_dense0_relu_fwd -->\n",
       "<g id=\"edge42\" class=\"edge\"><title>rn_model2_fcoutputmodel0_dropout0_fwd&#45;&gt;rn_model2_fcoutputmodel0_dense0_relu_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3781.58C129,-3768.28 129,-3753.63 129,-3741.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3791.89 124.5,-3781.89 129,-3786.89 129,-3781.89 129,-3781.89 129,-3781.89 129,-3786.89 133.5,-3781.89 129,-3791.89 129,-3791.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"139.5\" y=\"-3762.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dense1_fwd -->\n",
       "<g id=\"node43\" class=\"node\"><title>rn_model2_fcoutputmodel0_dense1_fwd</title>\n",
       "<polygon fill=\"#fb8072\" stroke=\"black\" points=\"176,-3959 82,-3959 82,-3901 176,-3901 176,-3959\"/>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3933.8\" font-family=\"Times,serif\" font-size=\"14.00\">FullyConnected</text>\n",
       "<text text-anchor=\"middle\" x=\"129\" y=\"-3918.8\" font-family=\"Times,serif\" font-size=\"14.00\">10</text>\n",
       "</g>\n",
       "<!-- rn_model2_fcoutputmodel0_dense1_fwd&#45;&gt;rn_model2_fcoutputmodel0_dropout0_fwd -->\n",
       "<g id=\"edge43\" class=\"edge\"><title>rn_model2_fcoutputmodel0_dense1_fwd&#45;&gt;rn_model2_fcoutputmodel0_dropout0_fwd</title>\n",
       "<path fill=\"none\" stroke=\"black\" d=\"M129,-3890.58C129,-3877.28 129,-3862.63 129,-3850.3\"/>\n",
       "<polygon fill=\"black\" stroke=\"black\" points=\"129,-3900.89 124.5,-3890.89 129,-3895.89 129,-3890.89 129,-3890.89 129,-3890.89 129,-3895.89 133.5,-3890.89 129,-3900.89 129,-3900.89\"/>\n",
       "<text text-anchor=\"middle\" x=\"139.5\" y=\"-3871.8\" font-family=\"Times,serif\" font-size=\"14.00\">256</text>\n",
       "</g>\n",
       "</g>\n",
       "</svg>\n"
      ],
      "text/plain": [
       "<graphviz.dot.Digraph at 0x7f0f305c2518>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mx.viz.plot_network(model(mx.sym.var('data'), mx.sym.var('qst'), mx.sym.var('coord_tensor')), shape={'data':(128,3,75,75), 'qst':(128,11), 'coord_tensor': (128, 25, 2)})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Trainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:08:22.323271Z",
     "start_time": "2018-07-16T04:08:22.319317Z"
    }
   },
   "outputs": [],
   "source": [
    "#set optimizer\n",
    "trainer = gluon.Trainer(model.collect_params(),optimizer='adam',optimizer_params={'learning_rate':0.0001})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:08:22.327043Z",
     "start_time": "2018-07-16T04:08:22.324824Z"
    }
   },
   "outputs": [],
   "source": [
    "#define loss function\n",
    "loss = gluon.loss.SoftmaxCrossEntropyLoss()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Helper functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:08:22.331287Z",
     "start_time": "2018-07-16T04:08:22.328358Z"
    }
   },
   "outputs": [],
   "source": [
    "# prepare coord tensor\n",
    "def cvt_coord(i):\n",
    "    return [(i/5-2)/2., (i%5-2)/2.]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:08:22.336882Z",
     "start_time": "2018-07-16T04:08:22.332625Z"
    }
   },
   "outputs": [],
   "source": [
    "def cvt_data_axis_nd(data):\n",
    "    img = [e[0] for e in data]\n",
    "    qst = [e[1] for e in data]\n",
    "    ans = [e[2] for e in data]\n",
    "    return (nd.array(img), nd.array(qst), nd.array(ans))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:08:22.361207Z",
     "start_time": "2018-07-16T04:08:22.345308Z"
    }
   },
   "source": [
    "### Evaluation metric"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T04:08:22.361207Z",
     "start_time": "2018-07-16T04:08:22.345308Z"
    }
   },
   "outputs": [],
   "source": [
    "def evaluate_accuracy(data,  model, ctx):\n",
    "    acc_rel = mx.metric.Accuracy()\n",
    "    acc_norel = mx.metric.Accuracy()\n",
    "    for i, (rel_img, rel_qst, rel_label, norel_img, norel_qst, norel_label) in enumerate(data):\n",
    "\n",
    "        rel_img = rel_img.as_in_context(ctx)\n",
    "        rel_qst = rel_qst.as_in_context(ctx)\n",
    "        rel_label = rel_label.as_in_context(ctx)\n",
    "        \n",
    "        norel_img = norel_img.as_in_context(ctx)\n",
    "        norel_qst = norel_qst.as_in_context(ctx)\n",
    "        norel_label = norel_label.as_in_context(ctx)\n",
    "\n",
    "        output = model(rel_img, rel_qst, coord_tensors[0])        \n",
    "        predictions = nd.argmax(output,axis=1)        \n",
    "        acc_rel.update(preds=predictions, labels=rel_label)\n",
    "\n",
    "        output = model(norel_img, norel_qst, coord_tensors[0])        \n",
    "        predictions = nd.argmax(output,axis=1)        \n",
    "        acc_norel.update(preds=predictions, labels=norel_label)\n",
    "\n",
    "    return acc_rel, acc_norel"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data Loading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "rel_train_nd = cvt_data_axis_nd(rel_train)\n",
    "rel_test_nd = cvt_data_axis_nd(rel_test)\n",
    "norel_train_nd = cvt_data_axis_nd(norel_train)\n",
    "norel_test_nd = cvt_data_axis_nd(norel_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "batchsize = args['batch_size']*GPU_COUNT if GPU_COUNT > 0 else args['batch_size']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataset = ArrayDataset(rel_train_nd[0], rel_train_nd[1], rel_train_nd[2], norel_train_nd[0], norel_train_nd[1], norel_train_nd[2])\n",
    "test_dataset = ArrayDataset(rel_test_nd[0], rel_test_nd[1], rel_test_nd[2], norel_test_nd[0], norel_test_nd[1], norel_test_nd[2])\n",
    "\n",
    "dataloader_train = DataLoader(train_dataset, batch_size=batchsize, shuffle=True, last_batch='discard', num_workers=5)\n",
    "dataloader_test = DataLoader(test_dataset, batch_size=args['batch_size'], shuffle=False, last_batch='discard', num_workers=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Prepare coord_tensor\n",
    "coord_tensor = nd.zeros((batchsize, 25, 2))\n",
    "for i in range(25):\n",
    "     coord_tensor[:,i,:] = nd.array(cvt_coord(i))\n",
    "        \n",
    "coord_tensors = gluon.utils.split_and_load(coord_tensor,ctx)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training loop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.hybridize()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0], Rel loss: 0.1424, NoRel loss 0.0033, Acc Rel 0.8906, Acc NoRel 0.9844. 23.0805sec\n",
      "[1], Rel loss: 0.1408, NoRel loss 0.0036, Acc Rel 0.8938, Acc NoRel 0.9849. 23.0171sec\n",
      "[2], Rel loss: 0.1344, NoRel loss 0.0049, Acc Rel 0.8990, Acc NoRel 0.9833. 23.0294sec\n",
      "[3], Rel loss: 0.1268, NoRel loss 0.0037, Acc Rel 0.8844, Acc NoRel 0.9875. 22.9431sec\n",
      "[4], Rel loss: 0.1193, NoRel loss 0.0028, Acc Rel 0.8880, Acc NoRel 0.9859. 22.9057sec\n"
     ]
    }
   ],
   "source": [
    "rel_loss = list()\n",
    "norel_loss =  list()\n",
    "rel_acc = list()\n",
    "norel_acc = list()\n",
    "\n",
    "for epoch in range(args['epochs']):\n",
    "    cumulative_rel_loss = 0.0\n",
    "    cumulative_norel_loss = 0.0\n",
    "    tick = time.time()\n",
    "    \n",
    "    for i, (rel_img, rel_qst, rel_label, norel_img, norel_qst, norel_label) in enumerate(dataloader_train):\n",
    "\n",
    "        # Relational data        \n",
    "        rel_imgs = gluon.utils.split_and_load(rel_img, ctx)\n",
    "        rel_qsts = gluon.utils.split_and_load(rel_qst, ctx)\n",
    "        rel_labels = gluon.utils.split_and_load(rel_label, ctx)\n",
    "        \n",
    "        # No-relationtional data       \n",
    "        norel_imgs = gluon.utils.split_and_load(norel_img, ctx)\n",
    "        norel_qsts = gluon.utils.split_and_load(norel_qst, ctx)\n",
    "        norel_labels = gluon.utils.split_and_load(norel_label, ctx)        \n",
    "\n",
    "        with autograd.record():\n",
    "            rel_losses = [loss(model(X,Y, coord),Z) for X, Y, coord, Z in zip(rel_imgs, rel_qsts, coord_tensors, rel_labels )]\n",
    "        for l in rel_losses:\n",
    "            l.backward()\n",
    "            \n",
    "        trainer.step(rel_img.shape[0])\n",
    "        \n",
    "        for l in rel_losses:\n",
    "            cumulative_rel_loss += l.mean().asscalar()\n",
    "        \n",
    "        with autograd.record():\n",
    "            norel_losses = [loss(model(X,Y, coord),Z) for X, Y, coord, Z in zip(norel_imgs, norel_qsts, coord_tensors, norel_labels)]\n",
    "        for l in norel_losses:\n",
    "            l.backward()\n",
    "            \n",
    "        trainer.step(norel_img.shape[0])\n",
    "        \n",
    "        for l in norel_losses:\n",
    "                cumulative_norel_loss += l.mean().asscalar()\n",
    "        \n",
    "    mx.nd.waitall()\n",
    "    acc_rel, acc_norel = evaluate_accuracy(dataloader_test, model, mx.gpu(0))\n",
    "    print(\"[{}], Rel loss: {:.4f}, NoRel loss {:.4f}, Acc Rel {:.4f}, Acc NoRel {:.4f}. {:.4f}sec\".format(\n",
    "        epoch, cumulative_rel_loss/len(dataloader_train), cumulative_norel_loss/len(dataloader_train), \n",
    "        acc_rel.get()[1], acc_norel.get()[1], time.time()-tick))\n",
    "    \n",
    "    rel_loss.append(cumulative_rel_loss/len(dataloader_train))\n",
    "    norel_loss.append(cumulative_norel_loss/len(dataloader_train))\n",
    "    rel_acc.append(acc_rel.get()[1])\n",
    "    norel_acc.append(acc_norel.get()[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T23:57:16.191883Z",
     "start_time": "2018-07-16T23:57:16.189005Z"
    }
   },
   "outputs": [],
   "source": [
    "perf_metric = collections.OrderedDict()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T23:57:17.412193Z",
     "start_time": "2018-07-16T23:57:17.407821Z"
    }
   },
   "outputs": [],
   "source": [
    "perf_metric['epoch'] = np.arange(args['epochs'])+1\n",
    "perf_metric['rel_loss'] = np.array(rel_loss)\n",
    "perf_metric['no_rel_loss'] =np.array(norel_loss)\n",
    "perf_metric['rel_acc'] = np.array(rel_acc)\n",
    "perf_metric['no_rel_acc'] = np.array(norel_acc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T23:57:17.792690Z",
     "start_time": "2018-07-16T23:57:17.789423Z"
    }
   },
   "outputs": [],
   "source": [
    "perf_metric = pd.DataFrame(perf_metric)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T23:57:20.337213Z",
     "start_time": "2018-07-16T23:57:20.089132Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.rcParams.update({'font.size': 12})\n",
    "plt.style.use('seaborn-darkgrid')\n",
    "palette = plt.get_cmap('Set1')\n",
    "plt.figure(figsize=(15,10))\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"loss\")\n",
    "\n",
    "plt.plot(np.arange(args['epochs'])+1,  np.array(rel_acc),label=\"rel_acc\")\n",
    "plt.plot(np.arange(args['epochs'])+1, np.array(norel_acc),label=\"norel_acc\")\n",
    "plt.legend(bbox_to_anchor=(1, 1), loc=2,)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-07-16T23:57:23.876446Z",
     "start_time": "2018-07-16T23:57:23.640015Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.rcParams.update({'font.size': 12})\n",
    "plt.style.use('seaborn-darkgrid')\n",
    "palette = plt.get_cmap('Set1')\n",
    "plt.figure(figsize=(15,10))\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"loss\")\n",
    "\n",
    "plt.plot(np.arange(args['epochs'])+1,  np.array(rel_loss),label=\"rel_acc\")\n",
    "plt.plot(np.arange(args['epochs'])+1, np.array(norel_loss),label=\"norel_loss\")\n",
    "plt.legend(bbox_to_anchor=(1, 1), loc=2,)\n",
    "plt.show()"
   ]
  },
  {
   "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
}