Q1.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import copy\n",
    "import pickle\n",
    "import os\n",
    "import torch\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "from torch.utils.data import Dataset, DataLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE=16\n",
    "NUM_CLASSES=10\n",
    "NUM_EPOCHS=20\n",
    "train_val_split=0.1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data loading functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train: (50000, 32, 32, 3)\n",
      "y_train: (50000,)\n",
      "X_test: (10000, 32, 32, 3)\n",
      "y_test: (10000,)\n"
     ]
    }
   ],
   "source": [
    "def load_CIFAR_batch(filename):\n",
    "    with open(filename, \"rb\") as f:\n",
    "        datadict = pickle.load(f, encoding=\"latin1\")\n",
    "        X = datadict[\"data\"]\n",
    "        Y = datadict[\"labels\"]\n",
    "        X = X.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype(\"int\")\n",
    "        Y = np.array(Y)\n",
    "        return X, Y\n",
    "\n",
    "\n",
    "def load_CIFAR10(ROOT):\n",
    "    \"\"\" load all of cifar batches (1-5 + test)\"\"\"\n",
    "    xs = []\n",
    "    ys = []\n",
    "    for b in range(1, 6):\n",
    "        f = os.path.join(ROOT, \"data_batch_%d\" % (b,))\n",
    "        X, Y = load_CIFAR_batch(f)\n",
    "        xs.append(X)\n",
    "        ys.append(Y)\n",
    "    Xtr = np.concatenate(xs)\n",
    "    Ytr = np.concatenate(ys)\n",
    "    del X, Y\n",
    "    Xte, Yte = load_CIFAR_batch(os.path.join(ROOT, \"test_batch\"))\n",
    "    return {\n",
    "        \"X_train\": Xtr,\n",
    "        \"y_train\": Ytr,\n",
    "        \"X_test\": Xte,\n",
    "        \"y_test\": Yte,\n",
    "    }\n",
    "CIFAR_DIR = r'C:\\Users\\talit\\Documents\\Msc\\Deep_Learning\\HW2\\cifar-10-batches-py'\n",
    "\n",
    "data = load_CIFAR10(CIFAR_DIR)\n",
    "for k, v in list(data.items()):\n",
    "    print(f\"{k}: {v.shape}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## split train data to train and validation data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train: (50000, 32, 32, 3)\n",
      "y_train: (50000,)\n",
      "X_test: (10000, 32, 32, 3)\n",
      "y_test: (10000,)\n",
      "X_train_new: (45000, 32, 32, 3)\n",
      "y_train_new: (45000,)\n",
      "X_val: (5000, 32, 32, 3)\n",
      "y_val: (5000,)\n"
     ]
    }
   ],
   "source": [
    "train_data_size = len(data['X_train'])\n",
    "X_train_new=data['X_train'][:int((1-train_val_split)*train_data_size)]\n",
    "y_train_new=data['y_train'][:int((1-train_val_split)*train_data_size)]\n",
    "X_val=data['X_train'][int((1-train_val_split)*train_data_size):]\n",
    "y_val=data['y_train'][int((1-train_val_split)*train_data_size):]\n",
    "\n",
    "data['X_train_new']=X_train_new\n",
    "data['y_train_new']=y_train_new\n",
    "data['X_val']=X_val\n",
    "data['y_val']=y_val\n",
    "\n",
    "for k, v in list(data.items()):\n",
    "    print(f\"{k}: {v.shape}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 16 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(8, 8))\n",
    "\n",
    "\n",
    "columns = 4\n",
    "rows = 4\n",
    "ids = [0, 200, 1000, 5000, \n",
    "       10000, 10200, 11000, 15000,\n",
    "       20000, 20200, 21000, 25000,\n",
    "       30000, 30200, 31000, 35000]\n",
    "for i in range(1, columns*rows +1):\n",
    "    img = data['X_train'][ids[i-1],:,:,:].squeeze()\n",
    "    fig.add_subplot(rows, columns, i)\n",
    "    plt.imshow(img)\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dataloaders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "class CIFAR10_data(Dataset):\n",
    "    def __init__(self,images,labels,transform=None):\n",
    "        self.images=images\n",
    "        self.labels=labels\n",
    "        self.transform=transform\n",
    "    def __len__(self):\n",
    "        return len(self.images)\n",
    "    def __getitem__(self,index):\n",
    "        label = torch.tensor(self.labels[index],dtype=torch.long)\n",
    "        image = self.images[index].astype('uint8')\n",
    "        if self.transform:\n",
    "            image = self.transform(image)\n",
    "        return image,label\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_transform=transforms.ToTensor()\n",
    "\n",
    "train_dataset = CIFAR10_data(data['X_train_new'],data['y_train_new'],data_transform)\n",
    "val_dataset = CIFAR10_data(data['X_val'],data['y_val'],data_transform)\n",
    "test_dataset = CIFAR10_data(data['X_test'],data['y_test'],data_transform)\n",
    "\n",
    "train_dataloader = DataLoader(train_dataset,batch_size=BATCH_SIZE,shuffle=True)\n",
    "val_dataloader = DataLoader(val_dataset,batch_size=BATCH_SIZE,shuffle=True)\n",
    "test_dataloader = DataLoader(test_dataset,batch_size=BATCH_SIZE,shuffle=True)\n",
    "\n",
    "datatloaders={'train':train_dataloader,'val':val_dataloader,'test':test_dataloader}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "def count_parameters(model): \n",
    "    return sum(p.numel() for p in model.parameters() if p.requires_grad)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fully-Connected model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "class FullyConnectedNet(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.flt = nn.Flatten()\n",
    "        self.fc1 = nn.Linear(32 * 32 * 3, 512)\n",
    "        self.fc2 = nn.Linear(512, 256)\n",
    "        self.fc3 = nn.Linear(256, 128)\n",
    "        self.fc4 = nn.Linear(128, NUM_CLASSES)\n",
    "        self.relu=nn.ReLU(inplace=True)\n",
    "        self.drop =  nn.Dropout(p=0.3)\n",
    "    def forward(self,x):\n",
    "        x=self.flt(x)\n",
    "        x=self.drop(self.relu(self.fc1(x)))\n",
    "        x=self.drop(self.relu(self.fc2(x)))\n",
    "        x=self.drop(self.relu(self.fc3(x)))\n",
    "        x=self.fc4(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "FullyConnectedNet(\n",
      "  (flt): Flatten(start_dim=1, end_dim=-1)\n",
      "  (fc1): Linear(in_features=3072, out_features=512, bias=True)\n",
      "  (fc2): Linear(in_features=512, out_features=256, bias=True)\n",
      "  (fc3): Linear(in_features=256, out_features=128, bias=True)\n",
      "  (fc4): Linear(in_features=128, out_features=10, bias=True)\n",
      "  (relu): ReLU(inplace=True)\n",
      "  (drop): Dropout(p=0.3, inplace=False)\n",
      ")\n",
      "\n",
      "Number of parameters for FC model: 1738890\n",
      "\n",
      "Epoch 0\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.2917, acc: 0.125\n",
      "Batch 100/2813: loss: 2.3052, acc: 0.1175\n",
      "Batch 200/2813: loss: 2.2532, acc: 0.1299\n",
      "Batch 300/2813: loss: 2.2117, acc: 0.1482\n",
      "Batch 400/2813: loss: 2.1849, acc: 0.1567\n",
      "Batch 500/2813: loss: 2.1661, acc: 0.1645\n",
      "Batch 600/2813: loss: 2.1546, acc: 0.1710\n",
      "Batch 700/2813: loss: 2.1415, acc: 0.1781\n",
      "Batch 800/2813: loss: 2.1341, acc: 0.1820\n",
      "Batch 900/2813: loss: 2.1316, acc: 0.1853\n",
      "Batch 1000/2813: loss: 2.1249, acc: 0.1879\n",
      "Batch 1100/2813: loss: 2.1146, acc: 0.1931\n",
      "Batch 1200/2813: loss: 2.1085, acc: 0.1963\n",
      "Batch 1300/2813: loss: 2.1022, acc: 0.2005\n",
      "Batch 1400/2813: loss: 2.0991, acc: 0.2036\n",
      "Batch 1500/2813: loss: 2.0940, acc: 0.2056\n",
      "Batch 1600/2813: loss: 2.0895, acc: 0.2085\n",
      "Batch 1700/2813: loss: 2.0851, acc: 0.2094\n",
      "Batch 1800/2813: loss: 2.0822, acc: 0.2107\n",
      "Batch 1900/2813: loss: 2.0780, acc: 0.2132\n",
      "Batch 2000/2813: loss: 2.0740, acc: 0.2147\n",
      "Batch 2100/2813: loss: 2.0700, acc: 0.2161\n",
      "Batch 2200/2813: loss: 2.0671, acc: 0.2172\n",
      "Batch 2300/2813: loss: 2.0634, acc: 0.2190\n",
      "Batch 2400/2813: loss: 2.0607, acc: 0.2202\n",
      "Batch 2500/2813: loss: 2.0584, acc: 0.2216\n",
      "Batch 2600/2813: loss: 2.0564, acc: 0.2226\n",
      "Batch 2700/2813: loss: 2.0531, acc: 0.2248\n",
      "Batch 2800/2813: loss: 2.0492, acc: 0.2265\n",
      "Epoch 0:  train loss: 2.0488, train acc: 0.2266\n",
      "Epoch 0:  val loss: 1.9155, val acc: 0.2918\n",
      "Saving model, val acc improved from 0.0 to 0.2918\n",
      "\n",
      "Epoch 1\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.0820, acc: 0.375\n",
      "Batch 100/2813: loss: 1.9404, acc: 0.2858\n",
      "Batch 200/2813: loss: 1.9402, acc: 0.2776\n",
      "Batch 300/2813: loss: 1.9475, acc: 0.2763\n",
      "Batch 400/2813: loss: 1.9580, acc: 0.2676\n",
      "Batch 500/2813: loss: 1.9603, acc: 0.2669\n",
      "Batch 600/2813: loss: 1.9567, acc: 0.2677\n",
      "Batch 700/2813: loss: 1.9591, acc: 0.2673\n",
      "Batch 800/2813: loss: 1.9589, acc: 0.2696\n",
      "Batch 900/2813: loss: 1.9542, acc: 0.2703\n",
      "Batch 1000/2813: loss: 1.9568, acc: 0.2691\n",
      "Batch 1100/2813: loss: 1.9541, acc: 0.2688\n",
      "Batch 1200/2813: loss: 1.9549, acc: 0.2695\n",
      "Batch 1300/2813: loss: 1.9555, acc: 0.2688\n",
      "Batch 1400/2813: loss: 1.9579, acc: 0.2692\n",
      "Batch 1500/2813: loss: 1.9576, acc: 0.2691\n",
      "Batch 1600/2813: loss: 1.9578, acc: 0.2695\n",
      "Batch 1700/2813: loss: 1.9586, acc: 0.2703\n",
      "Batch 1800/2813: loss: 1.9591, acc: 0.2700\n",
      "Batch 1900/2813: loss: 1.9574, acc: 0.2702\n",
      "Batch 2000/2813: loss: 1.9564, acc: 0.2712\n",
      "Batch 2100/2813: loss: 1.9546, acc: 0.2718\n",
      "Batch 2200/2813: loss: 1.9545, acc: 0.2713\n",
      "Batch 2300/2813: loss: 1.9516, acc: 0.2729\n",
      "Batch 2400/2813: loss: 1.9509, acc: 0.2733\n",
      "Batch 2500/2813: loss: 1.9501, acc: 0.2739\n",
      "Batch 2600/2813: loss: 1.9490, acc: 0.2742\n",
      "Batch 2700/2813: loss: 1.9482, acc: 0.2745\n",
      "Batch 2800/2813: loss: 1.9465, acc: 0.2755\n",
      "Epoch 1:  train loss: 1.9465, train acc: 0.2756\n",
      "Epoch 1:  val loss: 1.9472, val acc: 0.2818\n",
      "val acc did not improve from 0.2918\n",
      "\n",
      "Epoch 2\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.7391, acc: 0.1875\n",
      "Batch 100/2813: loss: 1.9484, acc: 0.2735\n",
      "Batch 200/2813: loss: 1.9358, acc: 0.2686\n",
      "Batch 300/2813: loss: 1.9362, acc: 0.2738\n",
      "Batch 400/2813: loss: 1.9265, acc: 0.2802\n",
      "Batch 500/2813: loss: 1.9265, acc: 0.2834\n",
      "Batch 600/2813: loss: 1.9213, acc: 0.2886\n",
      "Batch 700/2813: loss: 1.9216, acc: 0.2875\n",
      "Batch 800/2813: loss: 1.9207, acc: 0.2864\n",
      "Batch 900/2813: loss: 1.9256, acc: 0.2866\n",
      "Batch 1000/2813: loss: 1.9234, acc: 0.2891\n",
      "Batch 1100/2813: loss: 1.9245, acc: 0.2875\n",
      "Batch 1200/2813: loss: 1.9236, acc: 0.2877\n",
      "Batch 1300/2813: loss: 1.9235, acc: 0.2878\n",
      "Batch 1400/2813: loss: 1.9211, acc: 0.2880\n",
      "Batch 1500/2813: loss: 1.9205, acc: 0.2880\n",
      "Batch 1600/2813: loss: 1.9182, acc: 0.2900\n",
      "Batch 1700/2813: loss: 1.9160, acc: 0.2911\n",
      "Batch 1800/2813: loss: 1.9159, acc: 0.2914\n",
      "Batch 1900/2813: loss: 1.9164, acc: 0.2914\n",
      "Batch 2000/2813: loss: 1.9180, acc: 0.2910\n",
      "Batch 2100/2813: loss: 1.9184, acc: 0.2917\n",
      "Batch 2200/2813: loss: 1.9190, acc: 0.2921\n",
      "Batch 2300/2813: loss: 1.9189, acc: 0.2919\n",
      "Batch 2400/2813: loss: 1.9182, acc: 0.2922\n",
      "Batch 2500/2813: loss: 1.9181, acc: 0.2921\n",
      "Batch 2600/2813: loss: 1.9184, acc: 0.2919\n",
      "Batch 2700/2813: loss: 1.9178, acc: 0.2914\n",
      "Batch 2800/2813: loss: 1.9178, acc: 0.2914\n",
      "Epoch 2:  train loss: 1.9174, train acc: 0.2916\n",
      "Epoch 2:  val loss: 1.8625, val acc: 0.3288\n",
      "Saving model, val acc improved from 0.2918 to 0.3288\n",
      "\n",
      "Epoch 3\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.7102, acc: 0.4375\n",
      "Batch 100/2813: loss: 1.8894, acc: 0.2889\n",
      "Batch 200/2813: loss: 1.8804, acc: 0.3047\n",
      "Batch 300/2813: loss: 1.8889, acc: 0.3079\n",
      "Batch 400/2813: loss: 1.8939, acc: 0.3042\n",
      "Batch 500/2813: loss: 1.8992, acc: 0.3037\n",
      "Batch 600/2813: loss: 1.8923, acc: 0.3030\n",
      "Batch 700/2813: loss: 1.9009, acc: 0.3023\n",
      "Batch 800/2813: loss: 1.9023, acc: 0.3022\n",
      "Batch 900/2813: loss: 1.9012, acc: 0.3021\n",
      "Batch 1000/2813: loss: 1.9021, acc: 0.3021\n",
      "Batch 1100/2813: loss: 1.9022, acc: 0.3021\n",
      "Batch 1200/2813: loss: 1.9004, acc: 0.3015\n",
      "Batch 1300/2813: loss: 1.8994, acc: 0.3021\n",
      "Batch 1400/2813: loss: 1.8973, acc: 0.3010\n",
      "Batch 1500/2813: loss: 1.9007, acc: 0.3006\n",
      "Batch 1600/2813: loss: 1.9027, acc: 0.2999\n",
      "Batch 1700/2813: loss: 1.9021, acc: 0.3008\n",
      "Batch 1800/2813: loss: 1.9015, acc: 0.3012\n",
      "Batch 1900/2813: loss: 1.9013, acc: 0.3008\n",
      "Batch 2000/2813: loss: 1.9014, acc: 0.3003\n",
      "Batch 2100/2813: loss: 1.9021, acc: 0.2989\n",
      "Batch 2200/2813: loss: 1.9021, acc: 0.2991\n",
      "Batch 2300/2813: loss: 1.9016, acc: 0.2992\n",
      "Batch 2400/2813: loss: 1.9009, acc: 0.2999\n",
      "Batch 2500/2813: loss: 1.9028, acc: 0.2995\n",
      "Batch 2600/2813: loss: 1.9024, acc: 0.2997\n",
      "Batch 2700/2813: loss: 1.9034, acc: 0.2994\n",
      "Batch 2800/2813: loss: 1.9019, acc: 0.3000\n",
      "Epoch 3:  train loss: 1.9018, train acc: 0.3000\n",
      "Epoch 3:  val loss: 1.8272, val acc: 0.3434\n",
      "Saving model, val acc improved from 0.3288 to 0.3434\n",
      "\n",
      "Epoch 4\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.8197, acc: 0.3125\n",
      "Batch 100/2813: loss: 1.8558, acc: 0.3211\n",
      "Batch 200/2813: loss: 1.8490, acc: 0.3249\n",
      "Batch 300/2813: loss: 1.8663, acc: 0.3154\n",
      "Batch 400/2813: loss: 1.8763, acc: 0.3084\n",
      "Batch 500/2813: loss: 1.8760, acc: 0.3065\n",
      "Batch 600/2813: loss: 1.8817, acc: 0.3075\n",
      "Batch 700/2813: loss: 1.8855, acc: 0.3077\n",
      "Batch 800/2813: loss: 1.8895, acc: 0.3049\n",
      "Batch 900/2813: loss: 1.8850, acc: 0.3059\n",
      "Batch 1000/2813: loss: 1.8897, acc: 0.3015\n",
      "Batch 1100/2813: loss: 1.8926, acc: 0.2998\n",
      "Batch 1200/2813: loss: 1.8901, acc: 0.3015\n",
      "Batch 1300/2813: loss: 1.8867, acc: 0.3026\n",
      "Batch 1400/2813: loss: 1.8843, acc: 0.3033\n",
      "Batch 1500/2813: loss: 1.8816, acc: 0.3050\n",
      "Batch 1600/2813: loss: 1.8835, acc: 0.3045\n",
      "Batch 1700/2813: loss: 1.8854, acc: 0.3046\n",
      "Batch 1800/2813: loss: 1.8854, acc: 0.3050\n",
      "Batch 1900/2813: loss: 1.8841, acc: 0.3059\n",
      "Batch 2000/2813: loss: 1.8852, acc: 0.3061\n",
      "Batch 2100/2813: loss: 1.8851, acc: 0.3069\n",
      "Batch 2200/2813: loss: 1.8839, acc: 0.3069\n",
      "Batch 2300/2813: loss: 1.8826, acc: 0.3068\n",
      "Batch 2400/2813: loss: 1.8830, acc: 0.3077\n",
      "Batch 2500/2813: loss: 1.8843, acc: 0.3067\n",
      "Batch 2600/2813: loss: 1.8836, acc: 0.3070\n",
      "Batch 2700/2813: loss: 1.8824, acc: 0.3077\n",
      "Batch 2800/2813: loss: 1.8818, acc: 0.3077\n",
      "Epoch 4:  train loss: 1.8822, train acc: 0.3076\n",
      "Epoch 4:  val loss: 1.8242, val acc: 0.35\n",
      "Saving model, val acc improved from 0.3434 to 0.35\n",
      "\n",
      "Epoch 5\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.9055, acc: 0.4375\n",
      "Batch 100/2813: loss: 1.8928, acc: 0.2957\n",
      "Batch 200/2813: loss: 1.8884, acc: 0.2963\n",
      "Batch 300/2813: loss: 1.8905, acc: 0.3010\n",
      "Batch 400/2813: loss: 1.8843, acc: 0.3039\n",
      "Batch 500/2813: loss: 1.8952, acc: 0.3011\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 600/2813: loss: 1.8894, acc: 0.3026\n",
      "Batch 700/2813: loss: 1.8890, acc: 0.3032\n",
      "Batch 800/2813: loss: 1.8862, acc: 0.3036\n",
      "Batch 900/2813: loss: 1.8874, acc: 0.3040\n",
      "Batch 1000/2813: loss: 1.8874, acc: 0.3041\n",
      "Batch 1100/2813: loss: 1.8865, acc: 0.3069\n",
      "Batch 1200/2813: loss: 1.8836, acc: 0.3091\n",
      "Batch 1300/2813: loss: 1.8810, acc: 0.3090\n",
      "Batch 1400/2813: loss: 1.8805, acc: 0.3091\n",
      "Batch 1500/2813: loss: 1.8806, acc: 0.3081\n",
      "Batch 1600/2813: loss: 1.8809, acc: 0.3078\n",
      "Batch 1700/2813: loss: 1.8827, acc: 0.3070\n",
      "Batch 1800/2813: loss: 1.8824, acc: 0.3078\n",
      "Batch 1900/2813: loss: 1.8827, acc: 0.3081\n",
      "Batch 2000/2813: loss: 1.8813, acc: 0.3078\n",
      "Batch 2100/2813: loss: 1.8802, acc: 0.3084\n",
      "Batch 2200/2813: loss: 1.8791, acc: 0.3092\n",
      "Batch 2300/2813: loss: 1.8791, acc: 0.3096\n",
      "Batch 2400/2813: loss: 1.8780, acc: 0.3095\n",
      "Batch 2500/2813: loss: 1.8773, acc: 0.3100\n",
      "Batch 2600/2813: loss: 1.8770, acc: 0.3102\n",
      "Batch 2700/2813: loss: 1.8769, acc: 0.3106\n",
      "Batch 2800/2813: loss: 1.8769, acc: 0.3104\n",
      "Epoch 5:  train loss: 1.8768, train acc: 0.3106\n",
      "Epoch 5:  val loss: 1.8446, val acc: 0.3214\n",
      "val acc did not improve from 0.35\n",
      "\n",
      "Epoch 6\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.8771, acc: 0.4375\n",
      "Batch 100/2813: loss: 1.9063, acc: 0.2908\n",
      "Batch 200/2813: loss: 1.8994, acc: 0.2941\n",
      "Batch 300/2813: loss: 1.8934, acc: 0.3006\n",
      "Batch 400/2813: loss: 1.8849, acc: 0.3014\n",
      "Batch 500/2813: loss: 1.8843, acc: 0.3020\n",
      "Batch 600/2813: loss: 1.8829, acc: 0.3051\n",
      "Batch 700/2813: loss: 1.8873, acc: 0.3046\n",
      "Batch 800/2813: loss: 1.8824, acc: 0.3051\n",
      "Batch 900/2813: loss: 1.8822, acc: 0.3043\n",
      "Batch 1000/2813: loss: 1.8803, acc: 0.3060\n",
      "Batch 1100/2813: loss: 1.8774, acc: 0.3067\n",
      "Batch 1200/2813: loss: 1.8766, acc: 0.3077\n",
      "Batch 1300/2813: loss: 1.8758, acc: 0.3073\n",
      "Batch 1400/2813: loss: 1.8757, acc: 0.3078\n",
      "Batch 1500/2813: loss: 1.8790, acc: 0.3063\n",
      "Batch 1600/2813: loss: 1.8817, acc: 0.3059\n",
      "Batch 1700/2813: loss: 1.8822, acc: 0.3057\n",
      "Batch 1800/2813: loss: 1.8818, acc: 0.3061\n",
      "Batch 1900/2813: loss: 1.8819, acc: 0.3066\n",
      "Batch 2000/2813: loss: 1.8823, acc: 0.3066\n",
      "Batch 2100/2813: loss: 1.8813, acc: 0.3074\n",
      "Batch 2200/2813: loss: 1.8801, acc: 0.3082\n",
      "Batch 2300/2813: loss: 1.8794, acc: 0.3091\n",
      "Batch 2400/2813: loss: 1.8798, acc: 0.3084\n",
      "Batch 2500/2813: loss: 1.8788, acc: 0.3087\n",
      "Batch 2600/2813: loss: 1.8769, acc: 0.3092\n",
      "Batch 2700/2813: loss: 1.8773, acc: 0.3092\n",
      "Batch 2800/2813: loss: 1.8759, acc: 0.3102\n",
      "Epoch 6:  train loss: 1.8760, train acc: 0.3100\n",
      "Epoch 6:  val loss: 1.8082, val acc: 0.3516\n",
      "Saving model, val acc improved from 0.35 to 0.3516\n",
      "\n",
      "Epoch 7\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.8271, acc: 0.4375\n",
      "Batch 100/2813: loss: 1.8479, acc: 0.3131\n",
      "Batch 200/2813: loss: 1.8585, acc: 0.3159\n",
      "Batch 300/2813: loss: 1.8487, acc: 0.3247\n",
      "Batch 400/2813: loss: 1.8545, acc: 0.3246\n",
      "Batch 500/2813: loss: 1.8517, acc: 0.3255\n",
      "Batch 600/2813: loss: 1.8493, acc: 0.3259\n",
      "Batch 700/2813: loss: 1.8528, acc: 0.3234\n",
      "Batch 800/2813: loss: 1.8577, acc: 0.3204\n",
      "Batch 900/2813: loss: 1.8610, acc: 0.3170\n",
      "Batch 1000/2813: loss: 1.8595, acc: 0.3183\n",
      "Batch 1100/2813: loss: 1.8599, acc: 0.3174\n",
      "Batch 1200/2813: loss: 1.8606, acc: 0.3165\n",
      "Batch 1300/2813: loss: 1.8623, acc: 0.3140\n",
      "Batch 1400/2813: loss: 1.8637, acc: 0.3126\n",
      "Batch 1500/2813: loss: 1.8627, acc: 0.3136\n",
      "Batch 1600/2813: loss: 1.8647, acc: 0.3123\n",
      "Batch 1700/2813: loss: 1.8656, acc: 0.3114\n",
      "Batch 1800/2813: loss: 1.8666, acc: 0.3123\n",
      "Batch 1900/2813: loss: 1.8665, acc: 0.3128\n",
      "Batch 2000/2813: loss: 1.8662, acc: 0.3127\n",
      "Batch 2100/2813: loss: 1.8659, acc: 0.3133\n",
      "Batch 2200/2813: loss: 1.8651, acc: 0.3142\n",
      "Batch 2300/2813: loss: 1.8653, acc: 0.3144\n",
      "Batch 2400/2813: loss: 1.8650, acc: 0.3146\n",
      "Batch 2500/2813: loss: 1.8642, acc: 0.3151\n",
      "Batch 2600/2813: loss: 1.8627, acc: 0.3153\n",
      "Batch 2700/2813: loss: 1.8622, acc: 0.3153\n",
      "Batch 2800/2813: loss: 1.8629, acc: 0.3148\n",
      "Epoch 7:  train loss: 1.8633, train acc: 0.3149\n",
      "Epoch 7:  val loss: 1.8481, val acc: 0.3366\n",
      "val acc did not improve from 0.3516\n",
      "\n",
      "Epoch 8\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.0013, acc: 0.25\n",
      "Batch 100/2813: loss: 1.8653, acc: 0.3063\n",
      "Batch 200/2813: loss: 1.8823, acc: 0.3056\n",
      "Batch 300/2813: loss: 1.8704, acc: 0.3089\n",
      "Batch 400/2813: loss: 1.8653, acc: 0.3114\n",
      "Batch 500/2813: loss: 1.8691, acc: 0.3134\n",
      "Batch 600/2813: loss: 1.8640, acc: 0.3158\n",
      "Batch 700/2813: loss: 1.8619, acc: 0.3183\n",
      "Batch 800/2813: loss: 1.8580, acc: 0.3190\n",
      "Batch 900/2813: loss: 1.8572, acc: 0.3188\n",
      "Batch 1000/2813: loss: 1.8514, acc: 0.3186\n",
      "Batch 1100/2813: loss: 1.8502, acc: 0.3200\n",
      "Batch 1200/2813: loss: 1.8512, acc: 0.3200\n",
      "Batch 1300/2813: loss: 1.8531, acc: 0.3201\n",
      "Batch 1400/2813: loss: 1.8561, acc: 0.3199\n",
      "Batch 1500/2813: loss: 1.8563, acc: 0.3202\n",
      "Batch 1600/2813: loss: 1.8553, acc: 0.3205\n",
      "Batch 1700/2813: loss: 1.8539, acc: 0.3218\n",
      "Batch 1800/2813: loss: 1.8546, acc: 0.3212\n",
      "Batch 1900/2813: loss: 1.8535, acc: 0.3217\n",
      "Batch 2000/2813: loss: 1.8530, acc: 0.3217\n",
      "Batch 2100/2813: loss: 1.8533, acc: 0.3214\n",
      "Batch 2200/2813: loss: 1.8532, acc: 0.3218\n",
      "Batch 2300/2813: loss: 1.8545, acc: 0.3218\n",
      "Batch 2400/2813: loss: 1.8551, acc: 0.3219\n",
      "Batch 2500/2813: loss: 1.8552, acc: 0.3219\n",
      "Batch 2600/2813: loss: 1.8549, acc: 0.3225\n",
      "Batch 2700/2813: loss: 1.8556, acc: 0.3223\n",
      "Batch 2800/2813: loss: 1.8551, acc: 0.3222\n",
      "Epoch 8:  train loss: 1.8552, train acc: 0.3221\n",
      "Epoch 8:  val loss: 1.8174, val acc: 0.346\n",
      "val acc did not improve from 0.3516\n",
      "\n",
      "Epoch 9\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.7379, acc: 0.1875\n",
      "Batch 100/2813: loss: 1.8172, acc: 0.3224\n",
      "Batch 200/2813: loss: 1.8290, acc: 0.3277\n",
      "Batch 300/2813: loss: 1.8468, acc: 0.3243\n",
      "Batch 400/2813: loss: 1.8528, acc: 0.3204\n",
      "Batch 500/2813: loss: 1.8580, acc: 0.3191\n",
      "Batch 600/2813: loss: 1.8629, acc: 0.3181\n",
      "Batch 700/2813: loss: 1.8645, acc: 0.3172\n",
      "Batch 800/2813: loss: 1.8582, acc: 0.3191\n",
      "Batch 900/2813: loss: 1.8579, acc: 0.3208\n",
      "Batch 1000/2813: loss: 1.8537, acc: 0.3213\n",
      "Batch 1100/2813: loss: 1.8532, acc: 0.3212\n",
      "Batch 1200/2813: loss: 1.8535, acc: 0.3209\n",
      "Batch 1300/2813: loss: 1.8519, acc: 0.3219\n",
      "Batch 1400/2813: loss: 1.8527, acc: 0.3207\n",
      "Batch 1500/2813: loss: 1.8504, acc: 0.3228\n",
      "Batch 1600/2813: loss: 1.8518, acc: 0.3212\n",
      "Batch 1700/2813: loss: 1.8521, acc: 0.3209\n",
      "Batch 1800/2813: loss: 1.8513, acc: 0.3206\n",
      "Batch 1900/2813: loss: 1.8511, acc: 0.3205\n",
      "Batch 2000/2813: loss: 1.8530, acc: 0.3201\n",
      "Batch 2100/2813: loss: 1.8538, acc: 0.3200\n",
      "Batch 2200/2813: loss: 1.8536, acc: 0.3189\n",
      "Batch 2300/2813: loss: 1.8519, acc: 0.3195\n",
      "Batch 2400/2813: loss: 1.8514, acc: 0.3194\n",
      "Batch 2500/2813: loss: 1.8510, acc: 0.3197\n",
      "Batch 2600/2813: loss: 1.8521, acc: 0.3197\n",
      "Batch 2700/2813: loss: 1.8524, acc: 0.3194\n",
      "Batch 2800/2813: loss: 1.8514, acc: 0.3196\n",
      "Epoch 9:  train loss: 1.8515, train acc: 0.3195\n",
      "Epoch 9:  val loss: 1.7856, val acc: 0.3606\n",
      "Saving model, val acc improved from 0.3516 to 0.3606\n",
      "\n",
      "Epoch 10\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.7129, acc: 0.25\n",
      "Batch 100/2813: loss: 1.8881, acc: 0.3112\n",
      "Batch 200/2813: loss: 1.8703, acc: 0.3125\n",
      "Batch 300/2813: loss: 1.8627, acc: 0.3205\n",
      "Batch 400/2813: loss: 1.8631, acc: 0.3188\n",
      "Batch 500/2813: loss: 1.8612, acc: 0.3174\n",
      "Batch 600/2813: loss: 1.8595, acc: 0.3194\n",
      "Batch 700/2813: loss: 1.8548, acc: 0.3199\n",
      "Batch 800/2813: loss: 1.8585, acc: 0.3174\n",
      "Batch 900/2813: loss: 1.8554, acc: 0.3202\n",
      "Batch 1000/2813: loss: 1.8542, acc: 0.3213\n",
      "Batch 1100/2813: loss: 1.8554, acc: 0.3204\n",
      "Batch 1200/2813: loss: 1.8496, acc: 0.3219\n",
      "Batch 1300/2813: loss: 1.8508, acc: 0.3222\n",
      "Batch 1400/2813: loss: 1.8526, acc: 0.3221\n",
      "Batch 1500/2813: loss: 1.8499, acc: 0.3229\n",
      "Batch 1600/2813: loss: 1.8512, acc: 0.3220\n",
      "Batch 1700/2813: loss: 1.8533, acc: 0.3219\n",
      "Batch 1800/2813: loss: 1.8519, acc: 0.3225\n",
      "Batch 1900/2813: loss: 1.8524, acc: 0.3223\n",
      "Batch 2000/2813: loss: 1.8541, acc: 0.3219\n",
      "Batch 2100/2813: loss: 1.8536, acc: 0.3227\n",
      "Batch 2200/2813: loss: 1.8524, acc: 0.3224\n",
      "Batch 2300/2813: loss: 1.8517, acc: 0.3222\n",
      "Batch 2400/2813: loss: 1.8523, acc: 0.3226\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 2500/2813: loss: 1.8525, acc: 0.3229\n",
      "Batch 2600/2813: loss: 1.8509, acc: 0.3236\n",
      "Batch 2700/2813: loss: 1.8505, acc: 0.3239\n",
      "Batch 2800/2813: loss: 1.8515, acc: 0.3236\n",
      "Epoch 10:  train loss: 1.8512, train acc: 0.3236\n",
      "Epoch 10:  val loss: 1.7561, val acc: 0.3844\n",
      "Saving model, val acc improved from 0.3606 to 0.3844\n",
      "\n",
      "Epoch 11\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.8314, acc: 0.1875\n",
      "Batch 100/2813: loss: 1.8218, acc: 0.3285\n",
      "Batch 200/2813: loss: 1.8503, acc: 0.3317\n",
      "Batch 300/2813: loss: 1.8430, acc: 0.3324\n",
      "Batch 400/2813: loss: 1.8411, acc: 0.3329\n",
      "Batch 500/2813: loss: 1.8505, acc: 0.3257\n",
      "Batch 600/2813: loss: 1.8480, acc: 0.3287\n",
      "Batch 700/2813: loss: 1.8496, acc: 0.3295\n",
      "Batch 800/2813: loss: 1.8468, acc: 0.3284\n",
      "Batch 900/2813: loss: 1.8465, acc: 0.3285\n",
      "Batch 1000/2813: loss: 1.8468, acc: 0.3282\n",
      "Batch 1100/2813: loss: 1.8464, acc: 0.3267\n",
      "Batch 1200/2813: loss: 1.8439, acc: 0.3272\n",
      "Batch 1300/2813: loss: 1.8420, acc: 0.3266\n",
      "Batch 1400/2813: loss: 1.8420, acc: 0.3261\n",
      "Batch 1500/2813: loss: 1.8422, acc: 0.3266\n",
      "Batch 1600/2813: loss: 1.8412, acc: 0.3272\n",
      "Batch 1700/2813: loss: 1.8401, acc: 0.3282\n",
      "Batch 1800/2813: loss: 1.8379, acc: 0.3288\n",
      "Batch 1900/2813: loss: 1.8405, acc: 0.3285\n",
      "Batch 2000/2813: loss: 1.8408, acc: 0.3283\n",
      "Batch 2100/2813: loss: 1.8419, acc: 0.3274\n",
      "Batch 2200/2813: loss: 1.8410, acc: 0.3276\n",
      "Batch 2300/2813: loss: 1.8403, acc: 0.3279\n",
      "Batch 2400/2813: loss: 1.8403, acc: 0.3272\n",
      "Batch 2500/2813: loss: 1.8405, acc: 0.3267\n",
      "Batch 2600/2813: loss: 1.8416, acc: 0.3264\n",
      "Batch 2700/2813: loss: 1.8426, acc: 0.3258\n",
      "Batch 2800/2813: loss: 1.8425, acc: 0.3263\n",
      "Epoch 11:  train loss: 1.8426, train acc: 0.3264\n",
      "Epoch 11:  val loss: 1.7903, val acc: 0.3636\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 12\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.8440, acc: 0.375\n",
      "Batch 100/2813: loss: 1.8482, acc: 0.3353\n",
      "Batch 200/2813: loss: 1.8474, acc: 0.3261\n",
      "Batch 300/2813: loss: 1.8613, acc: 0.3216\n",
      "Batch 400/2813: loss: 1.8521, acc: 0.3241\n",
      "Batch 500/2813: loss: 1.8491, acc: 0.3259\n",
      "Batch 600/2813: loss: 1.8500, acc: 0.3261\n",
      "Batch 700/2813: loss: 1.8442, acc: 0.3287\n",
      "Batch 800/2813: loss: 1.8460, acc: 0.3264\n",
      "Batch 900/2813: loss: 1.8446, acc: 0.3260\n",
      "Batch 1000/2813: loss: 1.8455, acc: 0.3248\n",
      "Batch 1100/2813: loss: 1.8488, acc: 0.3238\n",
      "Batch 1200/2813: loss: 1.8508, acc: 0.3225\n",
      "Batch 1300/2813: loss: 1.8527, acc: 0.3204\n",
      "Batch 1400/2813: loss: 1.8527, acc: 0.3196\n",
      "Batch 1500/2813: loss: 1.8519, acc: 0.3200\n",
      "Batch 1600/2813: loss: 1.8526, acc: 0.3195\n",
      "Batch 1700/2813: loss: 1.8531, acc: 0.3195\n",
      "Batch 1800/2813: loss: 1.8535, acc: 0.3203\n",
      "Batch 1900/2813: loss: 1.8519, acc: 0.3204\n",
      "Batch 2000/2813: loss: 1.8485, acc: 0.3216\n",
      "Batch 2100/2813: loss: 1.8476, acc: 0.3218\n",
      "Batch 2200/2813: loss: 1.8457, acc: 0.3228\n",
      "Batch 2300/2813: loss: 1.8480, acc: 0.3218\n",
      "Batch 2400/2813: loss: 1.8471, acc: 0.3229\n",
      "Batch 2500/2813: loss: 1.8465, acc: 0.3234\n",
      "Batch 2600/2813: loss: 1.8468, acc: 0.3240\n",
      "Batch 2700/2813: loss: 1.8471, acc: 0.3237\n",
      "Batch 2800/2813: loss: 1.8464, acc: 0.3237\n",
      "Epoch 12:  train loss: 1.8467, train acc: 0.3237\n",
      "Epoch 12:  val loss: 1.7837, val acc: 0.3588\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 13\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.2896, acc: 0.3125\n",
      "Batch 100/2813: loss: 1.8373, acc: 0.3366\n",
      "Batch 200/2813: loss: 1.8385, acc: 0.3308\n",
      "Batch 300/2813: loss: 1.8420, acc: 0.3278\n",
      "Batch 400/2813: loss: 1.8315, acc: 0.3330\n",
      "Batch 500/2813: loss: 1.8358, acc: 0.3322\n",
      "Batch 600/2813: loss: 1.8398, acc: 0.3311\n",
      "Batch 700/2813: loss: 1.8379, acc: 0.3319\n",
      "Batch 800/2813: loss: 1.8377, acc: 0.3308\n",
      "Batch 900/2813: loss: 1.8394, acc: 0.3304\n",
      "Batch 1000/2813: loss: 1.8396, acc: 0.3283\n",
      "Batch 1100/2813: loss: 1.8391, acc: 0.3282\n",
      "Batch 1200/2813: loss: 1.8418, acc: 0.3276\n",
      "Batch 1300/2813: loss: 1.8422, acc: 0.3277\n",
      "Batch 1400/2813: loss: 1.8384, acc: 0.3291\n",
      "Batch 1500/2813: loss: 1.8375, acc: 0.3294\n",
      "Batch 1600/2813: loss: 1.8403, acc: 0.3281\n",
      "Batch 1700/2813: loss: 1.8370, acc: 0.3280\n",
      "Batch 1800/2813: loss: 1.8372, acc: 0.3274\n",
      "Batch 1900/2813: loss: 1.8388, acc: 0.3275\n",
      "Batch 2000/2813: loss: 1.8392, acc: 0.3267\n",
      "Batch 2100/2813: loss: 1.8394, acc: 0.3271\n",
      "Batch 2200/2813: loss: 1.8396, acc: 0.3275\n",
      "Batch 2300/2813: loss: 1.8409, acc: 0.3274\n",
      "Batch 2400/2813: loss: 1.8415, acc: 0.3271\n",
      "Batch 2500/2813: loss: 1.8394, acc: 0.3276\n",
      "Batch 2600/2813: loss: 1.8397, acc: 0.3279\n",
      "Batch 2700/2813: loss: 1.8410, acc: 0.3274\n",
      "Batch 2800/2813: loss: 1.8403, acc: 0.3281\n",
      "Epoch 13:  train loss: 1.8401, train acc: 0.3281\n",
      "Epoch 13:  val loss: 1.7873, val acc: 0.3594\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 14\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.7094, acc: 0.3125\n",
      "Batch 100/2813: loss: 1.8480, acc: 0.3205\n",
      "Batch 200/2813: loss: 1.8362, acc: 0.3156\n",
      "Batch 300/2813: loss: 1.8334, acc: 0.3203\n",
      "Batch 400/2813: loss: 1.8295, acc: 0.3196\n",
      "Batch 500/2813: loss: 1.8273, acc: 0.3221\n",
      "Batch 600/2813: loss: 1.8311, acc: 0.3233\n",
      "Batch 700/2813: loss: 1.8296, acc: 0.3240\n",
      "Batch 800/2813: loss: 1.8249, acc: 0.3277\n",
      "Batch 900/2813: loss: 1.8220, acc: 0.3291\n",
      "Batch 1000/2813: loss: 1.8192, acc: 0.3315\n",
      "Batch 1100/2813: loss: 1.8217, acc: 0.3315\n",
      "Batch 1200/2813: loss: 1.8232, acc: 0.3302\n",
      "Batch 1300/2813: loss: 1.8249, acc: 0.3299\n",
      "Batch 1400/2813: loss: 1.8271, acc: 0.3308\n",
      "Batch 1500/2813: loss: 1.8284, acc: 0.3316\n",
      "Batch 1600/2813: loss: 1.8290, acc: 0.3314\n",
      "Batch 1700/2813: loss: 1.8290, acc: 0.3309\n",
      "Batch 1800/2813: loss: 1.8314, acc: 0.3301\n",
      "Batch 1900/2813: loss: 1.8324, acc: 0.3300\n",
      "Batch 2000/2813: loss: 1.8331, acc: 0.3298\n",
      "Batch 2100/2813: loss: 1.8348, acc: 0.3297\n",
      "Batch 2200/2813: loss: 1.8338, acc: 0.3298\n",
      "Batch 2300/2813: loss: 1.8337, acc: 0.3296\n",
      "Batch 2400/2813: loss: 1.8359, acc: 0.3292\n",
      "Batch 2500/2813: loss: 1.8360, acc: 0.3290\n",
      "Batch 2600/2813: loss: 1.8367, acc: 0.3287\n",
      "Batch 2700/2813: loss: 1.8370, acc: 0.3288\n",
      "Batch 2800/2813: loss: 1.8365, acc: 0.3286\n",
      "Epoch 14:  train loss: 1.8370, train acc: 0.3284\n",
      "Epoch 14:  val loss: 1.7803, val acc: 0.3592\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 15\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.5802, acc: 0.375\n",
      "Batch 100/2813: loss: 1.8286, acc: 0.3211\n",
      "Batch 200/2813: loss: 1.8267, acc: 0.3171\n",
      "Batch 300/2813: loss: 1.8303, acc: 0.3195\n",
      "Batch 400/2813: loss: 1.8259, acc: 0.3224\n",
      "Batch 500/2813: loss: 1.8294, acc: 0.3247\n",
      "Batch 600/2813: loss: 1.8322, acc: 0.3251\n",
      "Batch 700/2813: loss: 1.8224, acc: 0.3292\n",
      "Batch 800/2813: loss: 1.8253, acc: 0.3309\n",
      "Batch 900/2813: loss: 1.8233, acc: 0.3303\n",
      "Batch 1000/2813: loss: 1.8277, acc: 0.3289\n",
      "Batch 1100/2813: loss: 1.8275, acc: 0.3291\n",
      "Batch 1200/2813: loss: 1.8286, acc: 0.3300\n",
      "Batch 1300/2813: loss: 1.8287, acc: 0.3311\n",
      "Batch 1400/2813: loss: 1.8274, acc: 0.3315\n",
      "Batch 1500/2813: loss: 1.8262, acc: 0.3316\n",
      "Batch 1600/2813: loss: 1.8277, acc: 0.3319\n",
      "Batch 1700/2813: loss: 1.8284, acc: 0.3308\n",
      "Batch 1800/2813: loss: 1.8286, acc: 0.3298\n",
      "Batch 1900/2813: loss: 1.8282, acc: 0.3296\n",
      "Batch 2000/2813: loss: 1.8309, acc: 0.3283\n",
      "Batch 2100/2813: loss: 1.8309, acc: 0.3285\n",
      "Batch 2200/2813: loss: 1.8302, acc: 0.3288\n",
      "Batch 2300/2813: loss: 1.8299, acc: 0.3288\n",
      "Batch 2400/2813: loss: 1.8314, acc: 0.3292\n",
      "Batch 2500/2813: loss: 1.8324, acc: 0.3288\n",
      "Batch 2600/2813: loss: 1.8320, acc: 0.3292\n",
      "Batch 2700/2813: loss: 1.8338, acc: 0.3292\n",
      "Batch 2800/2813: loss: 1.8337, acc: 0.3293\n",
      "Epoch 15:  train loss: 1.8338, train acc: 0.3293\n",
      "Epoch 15:  val loss: 1.7909, val acc: 0.3692\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 16\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.2701, acc: 0.1875\n",
      "Batch 100/2813: loss: 1.8769, acc: 0.3143\n",
      "Batch 200/2813: loss: 1.8541, acc: 0.3180\n",
      "Batch 300/2813: loss: 1.8422, acc: 0.3286\n",
      "Batch 400/2813: loss: 1.8369, acc: 0.3307\n",
      "Batch 500/2813: loss: 1.8306, acc: 0.3330\n",
      "Batch 600/2813: loss: 1.8269, acc: 0.3342\n",
      "Batch 700/2813: loss: 1.8265, acc: 0.3351\n",
      "Batch 800/2813: loss: 1.8245, acc: 0.3343\n",
      "Batch 900/2813: loss: 1.8281, acc: 0.3344\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 1000/2813: loss: 1.8269, acc: 0.3341\n",
      "Batch 1100/2813: loss: 1.8261, acc: 0.3341\n",
      "Batch 1200/2813: loss: 1.8285, acc: 0.3333\n",
      "Batch 1300/2813: loss: 1.8286, acc: 0.3324\n",
      "Batch 1400/2813: loss: 1.8288, acc: 0.3322\n",
      "Batch 1500/2813: loss: 1.8260, acc: 0.3327\n",
      "Batch 1600/2813: loss: 1.8276, acc: 0.3320\n",
      "Batch 1700/2813: loss: 1.8276, acc: 0.3326\n",
      "Batch 1800/2813: loss: 1.8270, acc: 0.3334\n",
      "Batch 1900/2813: loss: 1.8280, acc: 0.3323\n",
      "Batch 2000/2813: loss: 1.8297, acc: 0.3323\n",
      "Batch 2100/2813: loss: 1.8297, acc: 0.3321\n",
      "Batch 2200/2813: loss: 1.8301, acc: 0.3320\n",
      "Batch 2300/2813: loss: 1.8319, acc: 0.3316\n",
      "Batch 2400/2813: loss: 1.8319, acc: 0.3316\n",
      "Batch 2500/2813: loss: 1.8314, acc: 0.3318\n",
      "Batch 2600/2813: loss: 1.8306, acc: 0.3322\n",
      "Batch 2700/2813: loss: 1.8316, acc: 0.3314\n",
      "Batch 2800/2813: loss: 1.8316, acc: 0.3313\n",
      "Epoch 16:  train loss: 1.8316, train acc: 0.3312\n",
      "Epoch 16:  val loss: 1.7746, val acc: 0.3656\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 17\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.6396, acc: 0.4375\n",
      "Batch 100/2813: loss: 1.8450, acc: 0.3217\n",
      "Batch 200/2813: loss: 1.8596, acc: 0.3171\n",
      "Batch 300/2813: loss: 1.8566, acc: 0.3220\n",
      "Batch 400/2813: loss: 1.8494, acc: 0.3246\n",
      "Batch 500/2813: loss: 1.8570, acc: 0.3216\n",
      "Batch 600/2813: loss: 1.8523, acc: 0.3240\n",
      "Batch 700/2813: loss: 1.8544, acc: 0.3226\n",
      "Batch 800/2813: loss: 1.8514, acc: 0.3229\n",
      "Batch 900/2813: loss: 1.8454, acc: 0.3235\n",
      "Batch 1000/2813: loss: 1.8427, acc: 0.3241\n",
      "Batch 1100/2813: loss: 1.8402, acc: 0.3247\n",
      "Batch 1200/2813: loss: 1.8348, acc: 0.3267\n",
      "Batch 1300/2813: loss: 1.8346, acc: 0.3266\n",
      "Batch 1400/2813: loss: 1.8358, acc: 0.3263\n",
      "Batch 1500/2813: loss: 1.8380, acc: 0.3256\n",
      "Batch 1600/2813: loss: 1.8378, acc: 0.3256\n",
      "Batch 1700/2813: loss: 1.8366, acc: 0.3260\n",
      "Batch 1800/2813: loss: 1.8385, acc: 0.3246\n",
      "Batch 1900/2813: loss: 1.8365, acc: 0.3260\n",
      "Batch 2000/2813: loss: 1.8385, acc: 0.3263\n",
      "Batch 2100/2813: loss: 1.8384, acc: 0.3263\n",
      "Batch 2200/2813: loss: 1.8387, acc: 0.3267\n",
      "Batch 2300/2813: loss: 1.8376, acc: 0.3277\n",
      "Batch 2400/2813: loss: 1.8377, acc: 0.3278\n",
      "Batch 2500/2813: loss: 1.8374, acc: 0.3284\n",
      "Batch 2600/2813: loss: 1.8359, acc: 0.3296\n",
      "Batch 2700/2813: loss: 1.8358, acc: 0.3301\n",
      "Batch 2800/2813: loss: 1.8343, acc: 0.3303\n",
      "Epoch 17:  train loss: 1.8345, train acc: 0.3304\n",
      "Epoch 17:  val loss: 1.7493, val acc: 0.3824\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 18\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.5662, acc: 0.3125\n",
      "Batch 100/2813: loss: 1.8192, acc: 0.3409\n",
      "Batch 200/2813: loss: 1.8390, acc: 0.3252\n",
      "Batch 300/2813: loss: 1.8298, acc: 0.3251\n",
      "Batch 400/2813: loss: 1.8332, acc: 0.3298\n",
      "Batch 500/2813: loss: 1.8304, acc: 0.3328\n",
      "Batch 600/2813: loss: 1.8293, acc: 0.3348\n",
      "Batch 700/2813: loss: 1.8297, acc: 0.3342\n",
      "Batch 800/2813: loss: 1.8293, acc: 0.3352\n",
      "Batch 900/2813: loss: 1.8291, acc: 0.3352\n",
      "Batch 1000/2813: loss: 1.8315, acc: 0.3357\n",
      "Batch 1100/2813: loss: 1.8297, acc: 0.3366\n",
      "Batch 1200/2813: loss: 1.8271, acc: 0.3376\n",
      "Batch 1300/2813: loss: 1.8269, acc: 0.3370\n",
      "Batch 1400/2813: loss: 1.8293, acc: 0.3351\n",
      "Batch 1500/2813: loss: 1.8301, acc: 0.3329\n",
      "Batch 1600/2813: loss: 1.8284, acc: 0.3342\n",
      "Batch 1700/2813: loss: 1.8257, acc: 0.3340\n",
      "Batch 1800/2813: loss: 1.8259, acc: 0.3343\n",
      "Batch 1900/2813: loss: 1.8241, acc: 0.3342\n",
      "Batch 2000/2813: loss: 1.8225, acc: 0.3356\n",
      "Batch 2100/2813: loss: 1.8236, acc: 0.3348\n",
      "Batch 2200/2813: loss: 1.8250, acc: 0.3340\n",
      "Batch 2300/2813: loss: 1.8252, acc: 0.3336\n",
      "Batch 2400/2813: loss: 1.8253, acc: 0.3334\n",
      "Batch 2500/2813: loss: 1.8250, acc: 0.3334\n",
      "Batch 2600/2813: loss: 1.8255, acc: 0.3330\n",
      "Batch 2700/2813: loss: 1.8236, acc: 0.3333\n",
      "Batch 2800/2813: loss: 1.8240, acc: 0.3329\n",
      "Epoch 18:  train loss: 1.8245, train acc: 0.3328\n",
      "Epoch 18:  val loss: 1.7717, val acc: 0.3766\n",
      "val acc did not improve from 0.3844\n",
      "\n",
      "Epoch 19\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.8277, acc: 0.5\n",
      "Batch 100/2813: loss: 1.7967, acc: 0.3310\n",
      "Batch 200/2813: loss: 1.8193, acc: 0.3302\n",
      "Batch 300/2813: loss: 1.8239, acc: 0.3264\n",
      "Batch 400/2813: loss: 1.8364, acc: 0.3234\n",
      "Batch 500/2813: loss: 1.8306, acc: 0.3248\n",
      "Batch 600/2813: loss: 1.8307, acc: 0.3256\n",
      "Batch 700/2813: loss: 1.8222, acc: 0.3316\n",
      "Batch 800/2813: loss: 1.8213, acc: 0.3328\n",
      "Batch 900/2813: loss: 1.8207, acc: 0.3322\n",
      "Batch 1000/2813: loss: 1.8214, acc: 0.3320\n",
      "Batch 1100/2813: loss: 1.8218, acc: 0.3327\n",
      "Batch 1200/2813: loss: 1.8192, acc: 0.3342\n",
      "Batch 1300/2813: loss: 1.8199, acc: 0.3357\n",
      "Batch 1400/2813: loss: 1.8209, acc: 0.3345\n",
      "Batch 1500/2813: loss: 1.8183, acc: 0.3352\n",
      "Batch 1600/2813: loss: 1.8202, acc: 0.3340\n",
      "Batch 1700/2813: loss: 1.8214, acc: 0.3348\n",
      "Batch 1800/2813: loss: 1.8204, acc: 0.3349\n",
      "Batch 1900/2813: loss: 1.8220, acc: 0.3350\n",
      "Batch 2000/2813: loss: 1.8244, acc: 0.3328\n",
      "Batch 2100/2813: loss: 1.8237, acc: 0.3325\n",
      "Batch 2200/2813: loss: 1.8224, acc: 0.3320\n",
      "Batch 2300/2813: loss: 1.8235, acc: 0.3315\n",
      "Batch 2400/2813: loss: 1.8236, acc: 0.3314\n",
      "Batch 2500/2813: loss: 1.8249, acc: 0.3313\n",
      "Batch 2600/2813: loss: 1.8250, acc: 0.3312\n",
      "Batch 2700/2813: loss: 1.8249, acc: 0.3313\n",
      "Batch 2800/2813: loss: 1.8266, acc: 0.3309\n",
      "Epoch 19:  train loss: 1.8268, train acc: 0.3310\n",
      "Epoch 19:  val loss: 1.7609, val acc: 0.373\n",
      "val acc did not improve from 0.3844\n"
     ]
    }
   ],
   "source": [
    "fc_model=FullyConnectedNet()\n",
    "print(fc_model)\n",
    "print()\n",
    "print('Number of parameters for FC model: '+str(count_parameters(fc_model)))\n",
    "\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.Adam(fc_model.parameters())\n",
    "best_fc_val_acc=0.0\n",
    "best_fc_model=copy.deepcopy(fc_model)\n",
    "dest_path_best_model='FC_best_model.pth'\n",
    "\n",
    "history={'train_acc':[],'train_loss':[],'val_acc':[],'val_loss':[]}\n",
    "num_batches=len(train_dataloader)\n",
    "for epoch in range(NUM_EPOCHS):  \n",
    "    print()\n",
    "    print('Epoch '+str(epoch))\n",
    "    print('-'*100)\n",
    "    for phase in ['train','val']:\n",
    "        if phase=='train':\n",
    "            fc_model.train()\n",
    "        else:\n",
    "            fc_model.eval()\n",
    "        running_loss = 0.0\n",
    "        running_acc = 0.0\n",
    "        samples=0\n",
    "        for i, data_btch in enumerate(datatloaders[phase]):\n",
    "            inputs, labels = data_btch\n",
    "            optimizer.zero_grad()\n",
    "            outputs = fc_model(inputs)\n",
    "            loss = criterion(outputs, labels)\n",
    "            \n",
    "            if phase=='train':\n",
    "                loss.backward()\n",
    "                optimizer.step()\n",
    "\n",
    "            samples += labels.size(0)\n",
    "            running_loss += loss.item()*labels.size(0)\n",
    "            _, predicted = torch.max(outputs, 1)\n",
    "            running_acc += (predicted == labels).sum().item()\n",
    "            if phase=='train':\n",
    "                if i%100==0:\n",
    "                    print('Batch {}/{}: loss: {}, acc: {}'.format(str(i),str(num_batches),str(running_loss/samples)[:6],str(running_acc/samples)[:6]))\n",
    "        epoch_loss = float(running_loss)/samples\n",
    "        epoch_acc = float(running_acc)/samples\n",
    "        history[phase+'_acc'].append(epoch_acc)\n",
    "        history[phase+'_loss'].append(epoch_loss)\n",
    "        print('Epoch {}:  {} loss: {}, {} acc: {}'.format(str(epoch),phase,str(epoch_loss)[:6],phase,str(epoch_acc)[:6]))\n",
    "        if phase=='val':\n",
    "            if epoch_acc>best_fc_val_acc:\n",
    "                print('Saving model, val acc improved from {} to {}'.format(str(best_fc_val_acc)[:6],str(epoch_acc)[:6]))\n",
    "                best_fc_val_acc = epoch_acc\n",
    "                best_fc_model = copy.deepcopy(fc_model)\n",
    "                torch.save(fc_model.state_dict(),dest_path_best_model)\n",
    "            else:\n",
    "                print('val acc did not improve from {}'.format(str(best_fc_val_acc)[:6]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history['train_acc'],c='red')\n",
    "plt.plot(history['val_acc'],c='blue')\n",
    "plt.legend([\"train\", \"validation\"], loc =\"lower right\")\n",
    "plt.title('FC accuracy')\n",
    "plt.show()\n",
    "plt.clf()\n",
    "\n",
    "plt.plot(history['train_loss'],c='red')\n",
    "plt.plot(history['val_loss'],c='blue')\n",
    "plt.legend([\"train\", \"validation\"], loc =\"lower right\")\n",
    "plt.title('FC loss')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CNN model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "class CNNNet(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.conv1 = nn.Conv2d(3, 64, 3,padding='same')\n",
    "        self.conv2 = nn.Conv2d(64, 64, 3,padding='same')\n",
    "        \n",
    "        self.conv3 = nn.Conv2d(64, 128, 3,padding='same')\n",
    "        self.conv4 = nn.Conv2d(128, 128, 3,padding='same')\n",
    "        \n",
    "        self.conv5 = nn.Conv2d(128, 256, 3,padding='same')\n",
    "        self.conv6 = nn.Conv2d(256, 256, 3,padding='same')\n",
    "        \n",
    "        self.fc1 = nn.Linear(4096, 128)\n",
    "        self.fc2= nn.Linear(128, NUM_CLASSES)\n",
    "        \n",
    "        self.pool = nn.MaxPool2d(2, 2)\n",
    "        self.flt = nn.Flatten()\n",
    "        self.relu= nn.ReLU(inplace=True)\n",
    "        self.drop1 =  nn.Dropout(p=0.2)\n",
    "        self.drop2 =  nn.Dropout(p=0.3)\n",
    "\n",
    "    def forward(self, x):\n",
    "        x=self.relu(self.conv1(x))\n",
    "        x=self.relu(self.conv2(x))\n",
    "        x = self.pool(x)\n",
    "        x = self.drop1(x)\n",
    "        \n",
    "        x=self.relu(self.conv3(x))\n",
    "        x=self.relu(self.conv4(x))\n",
    "        x = self.pool(x)\n",
    "        x = self.drop1(x)\n",
    "        \n",
    "        x=self.relu(self.conv5(x))\n",
    "        x=self.relu(self.conv6(x))\n",
    "        x = self.pool(x)\n",
    "        x = self.drop1(x)\n",
    "\n",
    "        x = self.flt(x)\n",
    "        x = self.drop2(self.relu(self.fc1(x)))\n",
    "        x = self.fc2(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Number of parameters for CNN model: 1671114\n",
      "\n",
      "Epoch 0\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.2942, acc: 0.125\n",
      "Batch 100/2813: loss: 2.3044, acc: 0.1002\n",
      "Batch 200/2813: loss: 2.3004, acc: 0.1072\n",
      "Batch 300/2813: loss: 2.2644, acc: 0.1293\n",
      "Batch 400/2813: loss: 2.2241, acc: 0.1500\n",
      "Batch 500/2813: loss: 2.1874, acc: 0.1672\n",
      "Batch 600/2813: loss: 2.1594, acc: 0.1804\n",
      "Batch 700/2813: loss: 2.1310, acc: 0.1924\n",
      "Batch 800/2813: loss: 2.1026, acc: 0.2040\n",
      "Batch 900/2813: loss: 2.0764, acc: 0.2158\n",
      "Batch 1000/2813: loss: 2.0505, acc: 0.2258\n",
      "Batch 1100/2813: loss: 2.0289, acc: 0.2345\n",
      "Batch 1200/2813: loss: 2.0089, acc: 0.2417\n",
      "Batch 1300/2813: loss: 1.9909, acc: 0.2484\n",
      "Batch 1400/2813: loss: 1.9692, acc: 0.2575\n",
      "Batch 1500/2813: loss: 1.9524, acc: 0.2635\n",
      "Batch 1600/2813: loss: 1.9377, acc: 0.2695\n",
      "Batch 1700/2813: loss: 1.9236, acc: 0.2760\n",
      "Batch 1800/2813: loss: 1.9094, acc: 0.2812\n",
      "Batch 1900/2813: loss: 1.8950, acc: 0.2867\n",
      "Batch 2000/2813: loss: 1.8810, acc: 0.2924\n",
      "Batch 2100/2813: loss: 1.8680, acc: 0.2976\n",
      "Batch 2200/2813: loss: 1.8556, acc: 0.3022\n",
      "Batch 2300/2813: loss: 1.8431, acc: 0.3070\n",
      "Batch 2400/2813: loss: 1.8316, acc: 0.3120\n",
      "Batch 2500/2813: loss: 1.8213, acc: 0.3164\n",
      "Batch 2600/2813: loss: 1.8097, acc: 0.3215\n",
      "Batch 2700/2813: loss: 1.7986, acc: 0.3262\n",
      "Batch 2800/2813: loss: 1.7886, acc: 0.3303\n",
      "Epoch 0:  train loss: 1.7874, train acc: 0.3306\n",
      "Epoch 0:  val loss: 1.3589, val acc: 0.4924\n",
      "Saving model, val acc improved from 0.0 to 0.4924\n",
      "\n",
      "Epoch 1\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.9205, acc: 0.25\n",
      "Batch 100/2813: loss: 1.5171, acc: 0.4474\n",
      "Batch 200/2813: loss: 1.4919, acc: 0.4502\n",
      "Batch 300/2813: loss: 1.4826, acc: 0.4580\n",
      "Batch 400/2813: loss: 1.4796, acc: 0.4554\n",
      "Batch 500/2813: loss: 1.4783, acc: 0.4535\n",
      "Batch 600/2813: loss: 1.4759, acc: 0.4565\n",
      "Batch 700/2813: loss: 1.4746, acc: 0.4583\n",
      "Batch 800/2813: loss: 1.4712, acc: 0.4605\n",
      "Batch 900/2813: loss: 1.4676, acc: 0.4642\n",
      "Batch 1000/2813: loss: 1.4694, acc: 0.4652\n",
      "Batch 1100/2813: loss: 1.4695, acc: 0.4651\n",
      "Batch 1200/2813: loss: 1.4621, acc: 0.4672\n",
      "Batch 1300/2813: loss: 1.4560, acc: 0.4694\n",
      "Batch 1400/2813: loss: 1.4523, acc: 0.4697\n",
      "Batch 1500/2813: loss: 1.4449, acc: 0.4714\n",
      "Batch 1600/2813: loss: 1.4418, acc: 0.4724\n",
      "Batch 1700/2813: loss: 1.4377, acc: 0.4746\n",
      "Batch 1800/2813: loss: 1.4347, acc: 0.4760\n",
      "Batch 1900/2813: loss: 1.4307, acc: 0.4771\n",
      "Batch 2000/2813: loss: 1.4282, acc: 0.4785\n",
      "Batch 2100/2813: loss: 1.4256, acc: 0.4798\n",
      "Batch 2200/2813: loss: 1.4238, acc: 0.4806\n",
      "Batch 2300/2813: loss: 1.4198, acc: 0.4825\n",
      "Batch 2400/2813: loss: 1.4159, acc: 0.4834\n",
      "Batch 2500/2813: loss: 1.4139, acc: 0.4849\n",
      "Batch 2600/2813: loss: 1.4123, acc: 0.4856\n",
      "Batch 2700/2813: loss: 1.4084, acc: 0.4869\n",
      "Batch 2800/2813: loss: 1.4053, acc: 0.4883\n",
      "Epoch 1:  train loss: 1.4047, train acc: 0.4888\n",
      "Epoch 1:  val loss: 1.1553, val acc: 0.5844\n",
      "Saving model, val acc improved from 0.4924 to 0.5844\n",
      "\n",
      "Epoch 2\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.1361, acc: 0.625\n",
      "Batch 100/2813: loss: 1.2610, acc: 0.5402\n",
      "Batch 200/2813: loss: 1.2748, acc: 0.5416\n",
      "Batch 300/2813: loss: 1.2825, acc: 0.5392\n",
      "Batch 400/2813: loss: 1.2805, acc: 0.5384\n",
      "Batch 500/2813: loss: 1.2772, acc: 0.5346\n",
      "Batch 600/2813: loss: 1.2704, acc: 0.5379\n",
      "Batch 700/2813: loss: 1.2710, acc: 0.5399\n",
      "Batch 800/2813: loss: 1.2681, acc: 0.5422\n",
      "Batch 900/2813: loss: 1.2683, acc: 0.5423\n",
      "Batch 1000/2813: loss: 1.2705, acc: 0.5420\n",
      "Batch 1100/2813: loss: 1.2652, acc: 0.5434\n",
      "Batch 1200/2813: loss: 1.2640, acc: 0.5458\n",
      "Batch 1300/2813: loss: 1.2610, acc: 0.5474\n",
      "Batch 1400/2813: loss: 1.2587, acc: 0.5486\n",
      "Batch 1500/2813: loss: 1.2584, acc: 0.5491\n",
      "Batch 1600/2813: loss: 1.2595, acc: 0.5487\n",
      "Batch 1700/2813: loss: 1.2591, acc: 0.5492\n",
      "Batch 1800/2813: loss: 1.2567, acc: 0.5504\n",
      "Batch 1900/2813: loss: 1.2516, acc: 0.5520\n",
      "Batch 2000/2813: loss: 1.2503, acc: 0.5532\n",
      "Batch 2100/2813: loss: 1.2490, acc: 0.5532\n",
      "Batch 2200/2813: loss: 1.2516, acc: 0.5525\n",
      "Batch 2300/2813: loss: 1.2501, acc: 0.5528\n",
      "Batch 2400/2813: loss: 1.2498, acc: 0.5533\n",
      "Batch 2500/2813: loss: 1.2485, acc: 0.5538\n",
      "Batch 2600/2813: loss: 1.2462, acc: 0.5544\n",
      "Batch 2700/2813: loss: 1.2450, acc: 0.5555\n",
      "Batch 2800/2813: loss: 1.2419, acc: 0.5566\n",
      "Epoch 2:  train loss: 1.2413, train acc: 0.5567\n",
      "Epoch 2:  val loss: 0.9992, val acc: 0.6388\n",
      "Saving model, val acc improved from 0.5844 to 0.6388\n",
      "\n",
      "Epoch 3\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.2824, acc: 0.375\n",
      "Batch 100/2813: loss: 1.1494, acc: 0.5761\n",
      "Batch 200/2813: loss: 1.1468, acc: 0.5814\n",
      "Batch 300/2813: loss: 1.1452, acc: 0.5840\n",
      "Batch 400/2813: loss: 1.1360, acc: 0.5871\n",
      "Batch 500/2813: loss: 1.1267, acc: 0.5953\n",
      "Batch 600/2813: loss: 1.1323, acc: 0.5953\n",
      "Batch 700/2813: loss: 1.1287, acc: 0.5963\n",
      "Batch 800/2813: loss: 1.1362, acc: 0.5959\n",
      "Batch 900/2813: loss: 1.1356, acc: 0.5964\n",
      "Batch 1000/2813: loss: 1.1372, acc: 0.5952\n",
      "Batch 1100/2813: loss: 1.1402, acc: 0.5936\n",
      "Batch 1200/2813: loss: 1.1343, acc: 0.5961\n",
      "Batch 1300/2813: loss: 1.1322, acc: 0.5973\n",
      "Batch 1400/2813: loss: 1.1291, acc: 0.5987\n",
      "Batch 1500/2813: loss: 1.1296, acc: 0.5992\n",
      "Batch 1600/2813: loss: 1.1260, acc: 0.6005\n",
      "Batch 1700/2813: loss: 1.1252, acc: 0.6010\n",
      "Batch 1800/2813: loss: 1.1270, acc: 0.6007\n",
      "Batch 1900/2813: loss: 1.1245, acc: 0.6015\n",
      "Batch 2000/2813: loss: 1.1225, acc: 0.6026\n",
      "Batch 2100/2813: loss: 1.1217, acc: 0.6031\n",
      "Batch 2200/2813: loss: 1.1232, acc: 0.6027\n",
      "Batch 2300/2813: loss: 1.1235, acc: 0.6025\n",
      "Batch 2400/2813: loss: 1.1230, acc: 0.6030\n",
      "Batch 2500/2813: loss: 1.1211, acc: 0.6037\n",
      "Batch 2600/2813: loss: 1.1210, acc: 0.6043\n",
      "Batch 2700/2813: loss: 1.1200, acc: 0.6041\n",
      "Batch 2800/2813: loss: 1.1181, acc: 0.6051\n",
      "Epoch 3:  train loss: 1.1177, train acc: 0.6053\n",
      "Epoch 3:  val loss: 0.9362, val acc: 0.6698\n",
      "Saving model, val acc improved from 0.6388 to 0.6698\n",
      "\n",
      "Epoch 4\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.3754, acc: 0.375\n",
      "Batch 100/2813: loss: 0.9887, acc: 0.6633\n",
      "Batch 200/2813: loss: 1.0083, acc: 0.6498\n",
      "Batch 300/2813: loss: 1.0125, acc: 0.6461\n",
      "Batch 400/2813: loss: 1.0259, acc: 0.6413\n",
      "Batch 500/2813: loss: 1.0286, acc: 0.6447\n",
      "Batch 600/2813: loss: 1.0327, acc: 0.6405\n",
      "Batch 700/2813: loss: 1.0350, acc: 0.6422\n",
      "Batch 800/2813: loss: 1.0303, acc: 0.6420\n",
      "Batch 900/2813: loss: 1.0361, acc: 0.6408\n",
      "Batch 1000/2813: loss: 1.0383, acc: 0.6383\n",
      "Batch 1100/2813: loss: 1.0423, acc: 0.6364\n",
      "Batch 1200/2813: loss: 1.0390, acc: 0.6363\n",
      "Batch 1300/2813: loss: 1.0392, acc: 0.6367\n",
      "Batch 1400/2813: loss: 1.0398, acc: 0.6364\n",
      "Batch 1500/2813: loss: 1.0383, acc: 0.6374\n",
      "Batch 1600/2813: loss: 1.0368, acc: 0.6383\n",
      "Batch 1700/2813: loss: 1.0387, acc: 0.6387\n",
      "Batch 1800/2813: loss: 1.0372, acc: 0.6385\n",
      "Batch 1900/2813: loss: 1.0350, acc: 0.6395\n",
      "Batch 2000/2813: loss: 1.0331, acc: 0.6402\n",
      "Batch 2100/2813: loss: 1.0309, acc: 0.6410\n",
      "Batch 2200/2813: loss: 1.0308, acc: 0.6408\n",
      "Batch 2300/2813: loss: 1.0291, acc: 0.6414\n",
      "Batch 2400/2813: loss: 1.0286, acc: 0.6417\n",
      "Batch 2500/2813: loss: 1.0282, acc: 0.6418\n",
      "Batch 2600/2813: loss: 1.0281, acc: 0.6417\n",
      "Batch 2700/2813: loss: 1.0277, acc: 0.6422\n",
      "Batch 2800/2813: loss: 1.0266, acc: 0.6424\n",
      "Epoch 4:  train loss: 1.0264, train acc: 0.6424\n",
      "Epoch 4:  val loss: 0.8580, val acc: 0.7132\n",
      "Saving model, val acc improved from 0.6698 to 0.7132\n",
      "\n",
      "Epoch 5\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.1738, acc: 0.5625\n",
      "Batch 100/2813: loss: 0.9804, acc: 0.6602\n",
      "Batch 200/2813: loss: 0.9732, acc: 0.6592\n",
      "Batch 300/2813: loss: 0.9720, acc: 0.6636\n",
      "Batch 400/2813: loss: 0.9794, acc: 0.6577\n",
      "Batch 500/2813: loss: 0.9793, acc: 0.6560\n",
      "Batch 600/2813: loss: 0.9711, acc: 0.6572\n",
      "Batch 700/2813: loss: 0.9641, acc: 0.6614\n",
      "Batch 800/2813: loss: 0.9609, acc: 0.6629\n",
      "Batch 900/2813: loss: 0.9586, acc: 0.6642\n",
      "Batch 1000/2813: loss: 0.9665, acc: 0.6624\n",
      "Batch 1100/2813: loss: 0.9662, acc: 0.6613\n",
      "Batch 1200/2813: loss: 0.9728, acc: 0.6599\n",
      "Batch 1300/2813: loss: 0.9760, acc: 0.6587\n",
      "Batch 1400/2813: loss: 0.9745, acc: 0.6591\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 1500/2813: loss: 0.9737, acc: 0.6588\n",
      "Batch 1600/2813: loss: 0.9709, acc: 0.6597\n",
      "Batch 1700/2813: loss: 0.9702, acc: 0.6596\n",
      "Batch 1800/2813: loss: 0.9725, acc: 0.6590\n",
      "Batch 1900/2813: loss: 0.9720, acc: 0.6590\n",
      "Batch 2000/2813: loss: 0.9697, acc: 0.6598\n",
      "Batch 2100/2813: loss: 0.9686, acc: 0.6606\n",
      "Batch 2200/2813: loss: 0.9695, acc: 0.6604\n",
      "Batch 2300/2813: loss: 0.9680, acc: 0.6604\n",
      "Batch 2400/2813: loss: 0.9674, acc: 0.6613\n",
      "Batch 2500/2813: loss: 0.9680, acc: 0.6617\n",
      "Batch 2600/2813: loss: 0.9681, acc: 0.6618\n",
      "Batch 2700/2813: loss: 0.9685, acc: 0.6617\n",
      "Batch 2800/2813: loss: 0.9682, acc: 0.6623\n",
      "Epoch 5:  train loss: 0.9680, train acc: 0.6623\n",
      "Epoch 5:  val loss: 0.9133, val acc: 0.6916\n",
      "val acc did not improve from 0.7132\n",
      "\n",
      "Epoch 6\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.0060, acc: 0.625\n",
      "Batch 100/2813: loss: 0.9578, acc: 0.6683\n",
      "Batch 200/2813: loss: 0.9476, acc: 0.6713\n",
      "Batch 300/2813: loss: 0.9488, acc: 0.6710\n",
      "Batch 400/2813: loss: 0.9362, acc: 0.6740\n",
      "Batch 500/2813: loss: 0.9308, acc: 0.6763\n",
      "Batch 600/2813: loss: 0.9313, acc: 0.6777\n",
      "Batch 700/2813: loss: 0.9367, acc: 0.6772\n",
      "Batch 800/2813: loss: 0.9282, acc: 0.6806\n",
      "Batch 900/2813: loss: 0.9298, acc: 0.6798\n",
      "Batch 1000/2813: loss: 0.9277, acc: 0.6796\n",
      "Batch 1100/2813: loss: 0.9267, acc: 0.6805\n",
      "Batch 1200/2813: loss: 0.9256, acc: 0.6801\n",
      "Batch 1300/2813: loss: 0.9244, acc: 0.6810\n",
      "Batch 1400/2813: loss: 0.9218, acc: 0.6818\n",
      "Batch 1500/2813: loss: 0.9218, acc: 0.6817\n",
      "Batch 1600/2813: loss: 0.9203, acc: 0.6821\n",
      "Batch 1700/2813: loss: 0.9192, acc: 0.6819\n",
      "Batch 1800/2813: loss: 0.9188, acc: 0.6820\n",
      "Batch 1900/2813: loss: 0.9196, acc: 0.6819\n",
      "Batch 2000/2813: loss: 0.9192, acc: 0.6817\n",
      "Batch 2100/2813: loss: 0.9194, acc: 0.6820\n",
      "Batch 2200/2813: loss: 0.9203, acc: 0.6809\n",
      "Batch 2300/2813: loss: 0.9219, acc: 0.6807\n",
      "Batch 2400/2813: loss: 0.9222, acc: 0.6811\n",
      "Batch 2500/2813: loss: 0.9226, acc: 0.6811\n",
      "Batch 2600/2813: loss: 0.9215, acc: 0.6811\n",
      "Batch 2700/2813: loss: 0.9222, acc: 0.6810\n",
      "Batch 2800/2813: loss: 0.9214, acc: 0.6812\n",
      "Epoch 6:  train loss: 0.9218, train acc: 0.6812\n",
      "Epoch 6:  val loss: 0.8216, val acc: 0.727\n",
      "Saving model, val acc improved from 0.7132 to 0.727\n",
      "\n",
      "Epoch 7\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.5523, acc: 0.75\n",
      "Batch 100/2813: loss: 0.8483, acc: 0.7023\n",
      "Batch 200/2813: loss: 0.8666, acc: 0.6996\n",
      "Batch 300/2813: loss: 0.8757, acc: 0.6978\n",
      "Batch 400/2813: loss: 0.8637, acc: 0.7037\n",
      "Batch 500/2813: loss: 0.8635, acc: 0.7020\n",
      "Batch 600/2813: loss: 0.8712, acc: 0.6987\n",
      "Batch 700/2813: loss: 0.8778, acc: 0.6954\n",
      "Batch 800/2813: loss: 0.8714, acc: 0.6957\n",
      "Batch 900/2813: loss: 0.8723, acc: 0.6955\n",
      "Batch 1000/2813: loss: 0.8740, acc: 0.6944\n",
      "Batch 1100/2813: loss: 0.8774, acc: 0.6938\n",
      "Batch 1200/2813: loss: 0.8772, acc: 0.6944\n",
      "Batch 1300/2813: loss: 0.8807, acc: 0.6934\n",
      "Batch 1400/2813: loss: 0.8802, acc: 0.6930\n",
      "Batch 1500/2813: loss: 0.8803, acc: 0.6936\n",
      "Batch 1600/2813: loss: 0.8818, acc: 0.6927\n",
      "Batch 1700/2813: loss: 0.8802, acc: 0.6930\n",
      "Batch 1800/2813: loss: 0.8813, acc: 0.6929\n",
      "Batch 1900/2813: loss: 0.8807, acc: 0.6931\n",
      "Batch 2000/2813: loss: 0.8805, acc: 0.6935\n",
      "Batch 2100/2813: loss: 0.8828, acc: 0.6933\n",
      "Batch 2200/2813: loss: 0.8831, acc: 0.6944\n",
      "Batch 2300/2813: loss: 0.8819, acc: 0.6949\n",
      "Batch 2400/2813: loss: 0.8840, acc: 0.6942\n",
      "Batch 2500/2813: loss: 0.8860, acc: 0.6934\n",
      "Batch 2600/2813: loss: 0.8860, acc: 0.6939\n",
      "Batch 2700/2813: loss: 0.8876, acc: 0.6940\n",
      "Batch 2800/2813: loss: 0.8888, acc: 0.6936\n",
      "Epoch 7:  train loss: 0.8887, train acc: 0.6936\n",
      "Epoch 7:  val loss: 0.7868, val acc: 0.7284\n",
      "Saving model, val acc improved from 0.727 to 0.7284\n",
      "\n",
      "Epoch 8\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.0673, acc: 0.625\n",
      "Batch 100/2813: loss: 0.8693, acc: 0.7011\n",
      "Batch 200/2813: loss: 0.8431, acc: 0.7086\n",
      "Batch 300/2813: loss: 0.8289, acc: 0.7093\n",
      "Batch 400/2813: loss: 0.8401, acc: 0.7068\n",
      "Batch 500/2813: loss: 0.8378, acc: 0.7067\n",
      "Batch 600/2813: loss: 0.8415, acc: 0.7062\n",
      "Batch 700/2813: loss: 0.8426, acc: 0.7056\n",
      "Batch 800/2813: loss: 0.8370, acc: 0.7084\n",
      "Batch 900/2813: loss: 0.8336, acc: 0.7107\n",
      "Batch 1000/2813: loss: 0.8310, acc: 0.7104\n",
      "Batch 1100/2813: loss: 0.8284, acc: 0.7113\n",
      "Batch 1200/2813: loss: 0.8304, acc: 0.7116\n",
      "Batch 1300/2813: loss: 0.8329, acc: 0.7106\n",
      "Batch 1400/2813: loss: 0.8347, acc: 0.7086\n",
      "Batch 1500/2813: loss: 0.8383, acc: 0.7080\n",
      "Batch 1600/2813: loss: 0.8375, acc: 0.7082\n",
      "Batch 1700/2813: loss: 0.8389, acc: 0.7082\n",
      "Batch 1800/2813: loss: 0.8400, acc: 0.7084\n",
      "Batch 1900/2813: loss: 0.8393, acc: 0.7081\n",
      "Batch 2000/2813: loss: 0.8419, acc: 0.7066\n",
      "Batch 2100/2813: loss: 0.8414, acc: 0.7067\n",
      "Batch 2200/2813: loss: 0.8433, acc: 0.7057\n",
      "Batch 2300/2813: loss: 0.8414, acc: 0.7068\n",
      "Batch 2400/2813: loss: 0.8395, acc: 0.7079\n",
      "Batch 2500/2813: loss: 0.8408, acc: 0.7071\n",
      "Batch 2600/2813: loss: 0.8416, acc: 0.7072\n",
      "Batch 2700/2813: loss: 0.8429, acc: 0.7070\n",
      "Batch 2800/2813: loss: 0.8432, acc: 0.7070\n",
      "Epoch 8:  train loss: 0.8431, train acc: 0.7069\n",
      "Epoch 8:  val loss: 0.7582, val acc: 0.7414\n",
      "Saving model, val acc improved from 0.7284 to 0.7414\n",
      "\n",
      "Epoch 9\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 1.0244, acc: 0.6875\n",
      "Batch 100/2813: loss: 0.7643, acc: 0.7314\n",
      "Batch 200/2813: loss: 0.7661, acc: 0.7341\n",
      "Batch 300/2813: loss: 0.7917, acc: 0.7261\n",
      "Batch 400/2813: loss: 0.7757, acc: 0.7314\n",
      "Batch 500/2813: loss: 0.7874, acc: 0.7255\n",
      "Batch 600/2813: loss: 0.7933, acc: 0.7246\n",
      "Batch 700/2813: loss: 0.7901, acc: 0.7257\n",
      "Batch 800/2813: loss: 0.7937, acc: 0.7247\n",
      "Batch 900/2813: loss: 0.7907, acc: 0.7255\n",
      "Batch 1000/2813: loss: 0.7923, acc: 0.7242\n",
      "Batch 1100/2813: loss: 0.7977, acc: 0.7218\n",
      "Batch 1200/2813: loss: 0.8020, acc: 0.7212\n",
      "Batch 1300/2813: loss: 0.8002, acc: 0.7219\n",
      "Batch 1400/2813: loss: 0.8027, acc: 0.7216\n",
      "Batch 1500/2813: loss: 0.8034, acc: 0.7217\n",
      "Batch 1600/2813: loss: 0.8045, acc: 0.7214\n",
      "Batch 1700/2813: loss: 0.8053, acc: 0.7217\n",
      "Batch 1800/2813: loss: 0.8088, acc: 0.7199\n",
      "Batch 1900/2813: loss: 0.8090, acc: 0.7194\n",
      "Batch 2000/2813: loss: 0.8111, acc: 0.7182\n",
      "Batch 2100/2813: loss: 0.8110, acc: 0.7181\n",
      "Batch 2200/2813: loss: 0.8098, acc: 0.7183\n",
      "Batch 2300/2813: loss: 0.8112, acc: 0.7186\n",
      "Batch 2400/2813: loss: 0.8120, acc: 0.7186\n",
      "Batch 2500/2813: loss: 0.8138, acc: 0.7180\n",
      "Batch 2600/2813: loss: 0.8164, acc: 0.7171\n",
      "Batch 2700/2813: loss: 0.8155, acc: 0.7173\n",
      "Batch 2800/2813: loss: 0.8183, acc: 0.7169\n",
      "Epoch 9:  train loss: 0.8184, train acc: 0.7169\n",
      "Epoch 9:  val loss: 0.7327, val acc: 0.752\n",
      "Saving model, val acc improved from 0.7414 to 0.752\n",
      "\n",
      "Epoch 10\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.7952, acc: 0.625\n",
      "Batch 100/2813: loss: 0.7465, acc: 0.7419\n",
      "Batch 200/2813: loss: 0.7758, acc: 0.7350\n",
      "Batch 300/2813: loss: 0.7728, acc: 0.7313\n",
      "Batch 400/2813: loss: 0.7687, acc: 0.7339\n",
      "Batch 500/2813: loss: 0.7841, acc: 0.7314\n",
      "Batch 600/2813: loss: 0.7747, acc: 0.7336\n",
      "Batch 700/2813: loss: 0.7734, acc: 0.7353\n",
      "Batch 800/2813: loss: 0.7715, acc: 0.7366\n",
      "Batch 900/2813: loss: 0.7730, acc: 0.7349\n",
      "Batch 1000/2813: loss: 0.7788, acc: 0.7326\n",
      "Batch 1100/2813: loss: 0.7847, acc: 0.7316\n",
      "Batch 1200/2813: loss: 0.7869, acc: 0.7305\n",
      "Batch 1300/2813: loss: 0.7862, acc: 0.7312\n",
      "Batch 1400/2813: loss: 0.7874, acc: 0.7309\n",
      "Batch 1500/2813: loss: 0.7895, acc: 0.7305\n",
      "Batch 1600/2813: loss: 0.7889, acc: 0.7299\n",
      "Batch 1700/2813: loss: 0.7882, acc: 0.7301\n",
      "Batch 1800/2813: loss: 0.7899, acc: 0.7292\n",
      "Batch 1900/2813: loss: 0.7920, acc: 0.7281\n",
      "Batch 2000/2813: loss: 0.7936, acc: 0.7278\n",
      "Batch 2100/2813: loss: 0.7948, acc: 0.7275\n",
      "Batch 2200/2813: loss: 0.7941, acc: 0.7278\n",
      "Batch 2300/2813: loss: 0.7943, acc: 0.7278\n",
      "Batch 2400/2813: loss: 0.7947, acc: 0.7281\n",
      "Batch 2500/2813: loss: 0.7956, acc: 0.7276\n",
      "Batch 2600/2813: loss: 0.7945, acc: 0.7277\n",
      "Batch 2700/2813: loss: 0.7968, acc: 0.7269\n",
      "Batch 2800/2813: loss: 0.8013, acc: 0.7257\n",
      "Epoch 10:  train loss: 0.8015, train acc: 0.7256\n",
      "Epoch 10:  val loss: 0.7297, val acc: 0.7522\n",
      "Saving model, val acc improved from 0.752 to 0.7522\n",
      "\n",
      "Epoch 11\n",
      "----------------------------------------------------------------------------------------------------\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 0/2813: loss: 0.6418, acc: 0.75\n",
      "Batch 100/2813: loss: 0.7334, acc: 0.7444\n",
      "Batch 200/2813: loss: 0.7394, acc: 0.7450\n",
      "Batch 300/2813: loss: 0.7445, acc: 0.7425\n",
      "Batch 400/2813: loss: 0.7529, acc: 0.7408\n",
      "Batch 500/2813: loss: 0.7590, acc: 0.7400\n",
      "Batch 600/2813: loss: 0.7577, acc: 0.7413\n",
      "Batch 700/2813: loss: 0.7630, acc: 0.7407\n",
      "Batch 800/2813: loss: 0.7708, acc: 0.7371\n",
      "Batch 900/2813: loss: 0.7699, acc: 0.7365\n",
      "Batch 1000/2813: loss: 0.7716, acc: 0.7357\n",
      "Batch 1100/2813: loss: 0.7743, acc: 0.7338\n",
      "Batch 1200/2813: loss: 0.7764, acc: 0.7328\n",
      "Batch 1300/2813: loss: 0.7744, acc: 0.7328\n",
      "Batch 1400/2813: loss: 0.7724, acc: 0.7337\n",
      "Batch 1500/2813: loss: 0.7718, acc: 0.7337\n",
      "Batch 1600/2813: loss: 0.7722, acc: 0.7339\n",
      "Batch 1700/2813: loss: 0.7729, acc: 0.7336\n",
      "Batch 1800/2813: loss: 0.7745, acc: 0.7328\n",
      "Batch 1900/2813: loss: 0.7735, acc: 0.7331\n",
      "Batch 2000/2813: loss: 0.7749, acc: 0.7322\n",
      "Batch 2100/2813: loss: 0.7766, acc: 0.7314\n",
      "Batch 2200/2813: loss: 0.7787, acc: 0.7308\n",
      "Batch 2300/2813: loss: 0.7791, acc: 0.7301\n",
      "Batch 2400/2813: loss: 0.7783, acc: 0.7311\n",
      "Batch 2500/2813: loss: 0.7778, acc: 0.7311\n",
      "Batch 2600/2813: loss: 0.7790, acc: 0.7308\n",
      "Batch 2700/2813: loss: 0.7778, acc: 0.7312\n",
      "Batch 2800/2813: loss: 0.7788, acc: 0.7309\n",
      "Epoch 11:  train loss: 0.7792, train acc: 0.7308\n",
      "Epoch 11:  val loss: 0.7226, val acc: 0.7582\n",
      "Saving model, val acc improved from 0.7522 to 0.7582\n",
      "\n",
      "Epoch 12\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.7143, acc: 0.75\n",
      "Batch 100/2813: loss: 0.7974, acc: 0.7246\n",
      "Batch 200/2813: loss: 0.7609, acc: 0.7300\n",
      "Batch 300/2813: loss: 0.7664, acc: 0.7308\n",
      "Batch 400/2813: loss: 0.7680, acc: 0.7314\n",
      "Batch 500/2813: loss: 0.7694, acc: 0.7315\n",
      "Batch 600/2813: loss: 0.7670, acc: 0.7327\n",
      "Batch 700/2813: loss: 0.7653, acc: 0.7335\n",
      "Batch 800/2813: loss: 0.7563, acc: 0.7359\n",
      "Batch 900/2813: loss: 0.7622, acc: 0.7341\n",
      "Batch 1000/2813: loss: 0.7625, acc: 0.7348\n",
      "Batch 1100/2813: loss: 0.7622, acc: 0.7350\n",
      "Batch 1200/2813: loss: 0.7597, acc: 0.7362\n",
      "Batch 1300/2813: loss: 0.7589, acc: 0.7365\n",
      "Batch 1400/2813: loss: 0.7589, acc: 0.7368\n",
      "Batch 1500/2813: loss: 0.7617, acc: 0.7359\n",
      "Batch 1600/2813: loss: 0.7624, acc: 0.7357\n",
      "Batch 1700/2813: loss: 0.7614, acc: 0.7359\n",
      "Batch 1800/2813: loss: 0.7614, acc: 0.7363\n",
      "Batch 1900/2813: loss: 0.7600, acc: 0.7371\n",
      "Batch 2000/2813: loss: 0.7606, acc: 0.7361\n",
      "Batch 2100/2813: loss: 0.7601, acc: 0.7363\n",
      "Batch 2200/2813: loss: 0.7594, acc: 0.7363\n",
      "Batch 2300/2813: loss: 0.7592, acc: 0.7359\n",
      "Batch 2400/2813: loss: 0.7585, acc: 0.7360\n",
      "Batch 2500/2813: loss: 0.7586, acc: 0.7360\n",
      "Batch 2600/2813: loss: 0.7595, acc: 0.7356\n",
      "Batch 2700/2813: loss: 0.7603, acc: 0.7351\n",
      "Batch 2800/2813: loss: 0.7611, acc: 0.7354\n",
      "Epoch 12:  train loss: 0.7615, train acc: 0.7354\n",
      "Epoch 12:  val loss: 0.7193, val acc: 0.7564\n",
      "val acc did not improve from 0.7582\n",
      "\n",
      "Epoch 13\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.5464, acc: 0.8125\n",
      "Batch 100/2813: loss: 0.7371, acc: 0.7469\n",
      "Batch 200/2813: loss: 0.7412, acc: 0.7409\n",
      "Batch 300/2813: loss: 0.7235, acc: 0.7514\n",
      "Batch 400/2813: loss: 0.7202, acc: 0.7524\n",
      "Batch 500/2813: loss: 0.7186, acc: 0.7531\n",
      "Batch 600/2813: loss: 0.7273, acc: 0.7483\n",
      "Batch 700/2813: loss: 0.7312, acc: 0.7477\n",
      "Batch 800/2813: loss: 0.7291, acc: 0.7481\n",
      "Batch 900/2813: loss: 0.7331, acc: 0.7467\n",
      "Batch 1000/2813: loss: 0.7356, acc: 0.7463\n",
      "Batch 1100/2813: loss: 0.7373, acc: 0.7460\n",
      "Batch 1200/2813: loss: 0.7422, acc: 0.7443\n",
      "Batch 1300/2813: loss: 0.7449, acc: 0.7430\n",
      "Batch 1400/2813: loss: 0.7463, acc: 0.7425\n",
      "Batch 1500/2813: loss: 0.7501, acc: 0.7411\n",
      "Batch 1600/2813: loss: 0.7518, acc: 0.7404\n",
      "Batch 1700/2813: loss: 0.7516, acc: 0.7397\n",
      "Batch 1800/2813: loss: 0.7527, acc: 0.7393\n",
      "Batch 1900/2813: loss: 0.7501, acc: 0.7404\n",
      "Batch 2000/2813: loss: 0.7479, acc: 0.7408\n",
      "Batch 2100/2813: loss: 0.7460, acc: 0.7414\n",
      "Batch 2200/2813: loss: 0.7458, acc: 0.7418\n",
      "Batch 2300/2813: loss: 0.7478, acc: 0.7409\n",
      "Batch 2400/2813: loss: 0.7477, acc: 0.7407\n",
      "Batch 2500/2813: loss: 0.7468, acc: 0.7405\n",
      "Batch 2600/2813: loss: 0.7465, acc: 0.7404\n",
      "Batch 2700/2813: loss: 0.7471, acc: 0.7403\n",
      "Batch 2800/2813: loss: 0.7472, acc: 0.7405\n",
      "Epoch 13:  train loss: 0.7467, train acc: 0.7406\n",
      "Epoch 13:  val loss: 0.7220, val acc: 0.7634\n",
      "Saving model, val acc improved from 0.7582 to 0.7634\n",
      "\n",
      "Epoch 14\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.0269, acc: 0.625\n",
      "Batch 100/2813: loss: 0.7393, acc: 0.7407\n",
      "Batch 200/2813: loss: 0.7274, acc: 0.7462\n",
      "Batch 300/2813: loss: 0.7254, acc: 0.7483\n",
      "Batch 400/2813: loss: 0.7214, acc: 0.7520\n",
      "Batch 500/2813: loss: 0.7113, acc: 0.7549\n",
      "Batch 600/2813: loss: 0.7138, acc: 0.7550\n",
      "Batch 700/2813: loss: 0.7175, acc: 0.7539\n",
      "Batch 800/2813: loss: 0.7176, acc: 0.7525\n",
      "Batch 900/2813: loss: 0.7192, acc: 0.7499\n",
      "Batch 1000/2813: loss: 0.7216, acc: 0.7495\n",
      "Batch 1100/2813: loss: 0.7203, acc: 0.7499\n",
      "Batch 1200/2813: loss: 0.7192, acc: 0.7505\n",
      "Batch 1300/2813: loss: 0.7180, acc: 0.7508\n",
      "Batch 1400/2813: loss: 0.7202, acc: 0.75\n",
      "Batch 1500/2813: loss: 0.7241, acc: 0.7492\n",
      "Batch 1600/2813: loss: 0.7265, acc: 0.7484\n",
      "Batch 1700/2813: loss: 0.7256, acc: 0.7488\n",
      "Batch 1800/2813: loss: 0.7251, acc: 0.7495\n",
      "Batch 1900/2813: loss: 0.7260, acc: 0.7495\n",
      "Batch 2000/2813: loss: 0.7294, acc: 0.7481\n",
      "Batch 2100/2813: loss: 0.7286, acc: 0.7483\n",
      "Batch 2200/2813: loss: 0.7297, acc: 0.7478\n",
      "Batch 2300/2813: loss: 0.7301, acc: 0.7478\n",
      "Batch 2400/2813: loss: 0.7293, acc: 0.7482\n",
      "Batch 2500/2813: loss: 0.7287, acc: 0.7486\n",
      "Batch 2600/2813: loss: 0.7290, acc: 0.7487\n",
      "Batch 2700/2813: loss: 0.7320, acc: 0.7478\n",
      "Batch 2800/2813: loss: 0.7323, acc: 0.7470\n",
      "Epoch 14:  train loss: 0.7326, train acc: 0.7469\n",
      "Epoch 14:  val loss: 0.7828, val acc: 0.7394\n",
      "val acc did not improve from 0.7634\n",
      "\n",
      "Epoch 15\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 2.5078, acc: 0.5\n",
      "Batch 100/2813: loss: 0.7158, acc: 0.7617\n",
      "Batch 200/2813: loss: 0.7120, acc: 0.7559\n",
      "Batch 300/2813: loss: 0.7041, acc: 0.7545\n",
      "Batch 400/2813: loss: 0.7038, acc: 0.7540\n",
      "Batch 500/2813: loss: 0.7025, acc: 0.7556\n",
      "Batch 600/2813: loss: 0.7029, acc: 0.7547\n",
      "Batch 700/2813: loss: 0.7001, acc: 0.7573\n",
      "Batch 800/2813: loss: 0.7025, acc: 0.7571\n",
      "Batch 900/2813: loss: 0.7036, acc: 0.7562\n",
      "Batch 1000/2813: loss: 0.6999, acc: 0.7582\n",
      "Batch 1100/2813: loss: 0.6986, acc: 0.7591\n",
      "Batch 1200/2813: loss: 0.7020, acc: 0.7579\n",
      "Batch 1300/2813: loss: 0.7039, acc: 0.7572\n",
      "Batch 1400/2813: loss: 0.7029, acc: 0.7580\n",
      "Batch 1500/2813: loss: 0.7016, acc: 0.7583\n",
      "Batch 1600/2813: loss: 0.7045, acc: 0.7571\n",
      "Batch 1700/2813: loss: 0.7072, acc: 0.7560\n",
      "Batch 1800/2813: loss: 0.7067, acc: 0.7562\n",
      "Batch 1900/2813: loss: 0.7076, acc: 0.7553\n",
      "Batch 2000/2813: loss: 0.7097, acc: 0.7555\n",
      "Batch 2100/2813: loss: 0.7088, acc: 0.7562\n",
      "Batch 2200/2813: loss: 0.7093, acc: 0.7558\n",
      "Batch 2300/2813: loss: 0.7081, acc: 0.7564\n",
      "Batch 2400/2813: loss: 0.7074, acc: 0.7569\n",
      "Batch 2500/2813: loss: 0.7108, acc: 0.7560\n",
      "Batch 2600/2813: loss: 0.7141, acc: 0.7546\n",
      "Batch 2700/2813: loss: 0.7146, acc: 0.7549\n",
      "Batch 2800/2813: loss: 0.7136, acc: 0.7551\n",
      "Epoch 15:  train loss: 0.7138, train acc: 0.7552\n",
      "Epoch 15:  val loss: 0.7133, val acc: 0.7644\n",
      "Saving model, val acc improved from 0.7634 to 0.7644\n",
      "\n",
      "Epoch 16\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.4832, acc: 0.75\n",
      "Batch 100/2813: loss: 0.6571, acc: 0.7747\n",
      "Batch 200/2813: loss: 0.6951, acc: 0.7627\n",
      "Batch 300/2813: loss: 0.6956, acc: 0.7589\n",
      "Batch 400/2813: loss: 0.7001, acc: 0.7567\n",
      "Batch 500/2813: loss: 0.7015, acc: 0.7564\n",
      "Batch 600/2813: loss: 0.7032, acc: 0.7553\n",
      "Batch 700/2813: loss: 0.7041, acc: 0.7551\n",
      "Batch 800/2813: loss: 0.7085, acc: 0.7537\n",
      "Batch 900/2813: loss: 0.7086, acc: 0.7551\n",
      "Batch 1000/2813: loss: 0.7064, acc: 0.7558\n",
      "Batch 1100/2813: loss: 0.7034, acc: 0.7559\n",
      "Batch 1200/2813: loss: 0.7003, acc: 0.7574\n",
      "Batch 1300/2813: loss: 0.7004, acc: 0.7581\n",
      "Batch 1400/2813: loss: 0.7000, acc: 0.7585\n",
      "Batch 1500/2813: loss: 0.7032, acc: 0.7570\n",
      "Batch 1600/2813: loss: 0.7085, acc: 0.7555\n",
      "Batch 1700/2813: loss: 0.7081, acc: 0.7554\n",
      "Batch 1800/2813: loss: 0.7067, acc: 0.7562\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch 1900/2813: loss: 0.7053, acc: 0.7564\n",
      "Batch 2000/2813: loss: 0.7045, acc: 0.7571\n",
      "Batch 2100/2813: loss: 0.7052, acc: 0.7569\n",
      "Batch 2200/2813: loss: 0.7051, acc: 0.7574\n",
      "Batch 2300/2813: loss: 0.7070, acc: 0.7571\n",
      "Batch 2400/2813: loss: 0.7052, acc: 0.7580\n",
      "Batch 2500/2813: loss: 0.7064, acc: 0.7576\n",
      "Batch 2600/2813: loss: 0.7077, acc: 0.7577\n",
      "Batch 2700/2813: loss: 0.7077, acc: 0.7579\n",
      "Batch 2800/2813: loss: 0.7072, acc: 0.7582\n",
      "Epoch 16:  train loss: 0.7073, train acc: 0.7582\n",
      "Epoch 16:  val loss: 0.7098, val acc: 0.7656\n",
      "Saving model, val acc improved from 0.7644 to 0.7656\n",
      "\n",
      "Epoch 17\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.7116, acc: 0.8125\n",
      "Batch 100/2813: loss: 0.6623, acc: 0.7827\n",
      "Batch 200/2813: loss: 0.6691, acc: 0.7764\n",
      "Batch 300/2813: loss: 0.6609, acc: 0.7745\n",
      "Batch 400/2813: loss: 0.6623, acc: 0.7741\n",
      "Batch 500/2813: loss: 0.6732, acc: 0.7660\n",
      "Batch 600/2813: loss: 0.6750, acc: 0.7662\n",
      "Batch 700/2813: loss: 0.6754, acc: 0.7662\n",
      "Batch 800/2813: loss: 0.6727, acc: 0.7677\n",
      "Batch 900/2813: loss: 0.6725, acc: 0.7673\n",
      "Batch 1000/2813: loss: 0.6763, acc: 0.7658\n",
      "Batch 1100/2813: loss: 0.6768, acc: 0.7651\n",
      "Batch 1200/2813: loss: 0.6800, acc: 0.7644\n",
      "Batch 1300/2813: loss: 0.6807, acc: 0.7640\n",
      "Batch 1400/2813: loss: 0.6832, acc: 0.7633\n",
      "Batch 1500/2813: loss: 0.6844, acc: 0.7627\n",
      "Batch 1600/2813: loss: 0.6831, acc: 0.7629\n",
      "Batch 1700/2813: loss: 0.6842, acc: 0.7630\n",
      "Batch 1800/2813: loss: 0.6873, acc: 0.7621\n",
      "Batch 1900/2813: loss: 0.6877, acc: 0.7623\n",
      "Batch 2000/2813: loss: 0.6882, acc: 0.7623\n",
      "Batch 2100/2813: loss: 0.6905, acc: 0.7621\n",
      "Batch 2200/2813: loss: 0.6922, acc: 0.7614\n",
      "Batch 2300/2813: loss: 0.6938, acc: 0.7605\n",
      "Batch 2400/2813: loss: 0.6961, acc: 0.7603\n",
      "Batch 2500/2813: loss: 0.6958, acc: 0.7611\n",
      "Batch 2600/2813: loss: 0.6951, acc: 0.7614\n",
      "Batch 2700/2813: loss: 0.6963, acc: 0.7616\n",
      "Batch 2800/2813: loss: 0.6999, acc: 0.7600\n",
      "Epoch 17:  train loss: 0.7000, train acc: 0.7602\n",
      "Epoch 17:  val loss: 0.7014, val acc: 0.762\n",
      "val acc did not improve from 0.7656\n",
      "\n",
      "Epoch 18\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.8367, acc: 0.6875\n",
      "Batch 100/2813: loss: 0.6858, acc: 0.7704\n",
      "Batch 200/2813: loss: 0.6805, acc: 0.7667\n",
      "Batch 300/2813: loss: 0.6650, acc: 0.7697\n",
      "Batch 400/2813: loss: 0.6687, acc: 0.7693\n",
      "Batch 500/2813: loss: 0.6791, acc: 0.7678\n",
      "Batch 600/2813: loss: 0.6815, acc: 0.7668\n",
      "Batch 700/2813: loss: 0.6836, acc: 0.7672\n",
      "Batch 800/2813: loss: 0.6830, acc: 0.7666\n",
      "Batch 900/2813: loss: 0.6826, acc: 0.7672\n",
      "Batch 1000/2813: loss: 0.6803, acc: 0.7672\n",
      "Batch 1100/2813: loss: 0.6785, acc: 0.7672\n",
      "Batch 1200/2813: loss: 0.6803, acc: 0.7669\n",
      "Batch 1300/2813: loss: 0.6781, acc: 0.7670\n",
      "Batch 1400/2813: loss: 0.6801, acc: 0.7662\n",
      "Batch 1500/2813: loss: 0.6797, acc: 0.7659\n",
      "Batch 1600/2813: loss: 0.6786, acc: 0.7659\n",
      "Batch 1700/2813: loss: 0.6828, acc: 0.7641\n",
      "Batch 1800/2813: loss: 0.6840, acc: 0.7643\n",
      "Batch 1900/2813: loss: 0.6820, acc: 0.7649\n",
      "Batch 2000/2813: loss: 0.6875, acc: 0.7632\n",
      "Batch 2100/2813: loss: 0.6859, acc: 0.7640\n",
      "Batch 2200/2813: loss: 0.6887, acc: 0.7628\n",
      "Batch 2300/2813: loss: 0.6886, acc: 0.7638\n",
      "Batch 2400/2813: loss: 0.6880, acc: 0.7644\n",
      "Batch 2500/2813: loss: 0.6868, acc: 0.7645\n",
      "Batch 2600/2813: loss: 0.6864, acc: 0.7648\n",
      "Batch 2700/2813: loss: 0.6870, acc: 0.7649\n",
      "Batch 2800/2813: loss: 0.6881, acc: 0.7646\n",
      "Epoch 18:  train loss: 0.6876, train acc: 0.7648\n",
      "Epoch 18:  val loss: 0.8054, val acc: 0.744\n",
      "val acc did not improve from 0.7656\n",
      "\n",
      "Epoch 19\n",
      "----------------------------------------------------------------------------------------------------\n",
      "Batch 0/2813: loss: 0.5612, acc: 0.75\n",
      "Batch 100/2813: loss: 0.6883, acc: 0.7568\n",
      "Batch 200/2813: loss: 0.6574, acc: 0.7699\n",
      "Batch 300/2813: loss: 0.6553, acc: 0.7747\n",
      "Batch 400/2813: loss: 0.6548, acc: 0.7736\n",
      "Batch 500/2813: loss: 0.6564, acc: 0.7761\n",
      "Batch 600/2813: loss: 0.6565, acc: 0.7767\n",
      "Batch 700/2813: loss: 0.6482, acc: 0.7787\n",
      "Batch 800/2813: loss: 0.6554, acc: 0.7754\n",
      "Batch 900/2813: loss: 0.6573, acc: 0.7749\n",
      "Batch 1000/2813: loss: 0.6543, acc: 0.7749\n",
      "Batch 1100/2813: loss: 0.6551, acc: 0.7740\n",
      "Batch 1200/2813: loss: 0.6588, acc: 0.7728\n",
      "Batch 1300/2813: loss: 0.6585, acc: 0.7720\n",
      "Batch 1400/2813: loss: 0.6593, acc: 0.7725\n",
      "Batch 1500/2813: loss: 0.6598, acc: 0.7729\n",
      "Batch 1600/2813: loss: 0.6622, acc: 0.7720\n",
      "Batch 1700/2813: loss: 0.6625, acc: 0.7719\n",
      "Batch 1800/2813: loss: 0.6664, acc: 0.7713\n",
      "Batch 1900/2813: loss: 0.6698, acc: 0.7700\n",
      "Batch 2000/2813: loss: 0.6700, acc: 0.7698\n",
      "Batch 2100/2813: loss: 0.6727, acc: 0.7694\n",
      "Batch 2200/2813: loss: 0.6747, acc: 0.7697\n",
      "Batch 2300/2813: loss: 0.6772, acc: 0.7682\n",
      "Batch 2400/2813: loss: 0.6778, acc: 0.7677\n",
      "Batch 2500/2813: loss: 0.6829, acc: 0.7660\n",
      "Batch 2600/2813: loss: 0.6823, acc: 0.7662\n",
      "Batch 2700/2813: loss: 0.6819, acc: 0.7661\n",
      "Batch 2800/2813: loss: 0.6826, acc: 0.7659\n",
      "Epoch 19:  train loss: 0.6822, train acc: 0.7660\n",
      "Epoch 19:  val loss: 0.7030, val acc: 0.7766\n",
      "Saving model, val acc improved from 0.7656 to 0.7766\n"
     ]
    }
   ],
   "source": [
    "cnn_model = CNNNet()\n",
    "print()\n",
    "print('Number of parameters for CNN model: '+str(count_parameters(cnn_model)))\n",
    "\n",
    "cnn_criterion = nn.CrossEntropyLoss()\n",
    "cnn_optimizer = optim.Adam(cnn_model.parameters())\n",
    "\n",
    "best_cnn_val_acc=0.0\n",
    "best_cnn_model=copy.deepcopy(cnn_model)\n",
    "dest_path_best_model='CNN_best_model.pth'\n",
    "\n",
    "history_cnn={'train_acc':[],'train_loss':[],'val_acc':[],'val_loss':[]}\n",
    "\n",
    "num_batches=len(train_dataloader)\n",
    "for epoch in range(NUM_EPOCHS):  \n",
    "    print()\n",
    "    print('Epoch '+str(epoch))\n",
    "    print('-'*100)\n",
    "    for phase in ['train','val']:\n",
    "        if phase=='train':\n",
    "            cnn_model.train()\n",
    "        else:\n",
    "            cnn_model.eval()\n",
    "        running_loss = 0.0\n",
    "        running_acc = 0.0\n",
    "        samples=0\n",
    "        for i, data_btch in enumerate(datatloaders[phase]):\n",
    "            inputs, labels = data_btch\n",
    "            cnn_optimizer.zero_grad()\n",
    "            outputs = cnn_model(inputs)\n",
    "            cnn_loss = cnn_criterion(outputs, labels)\n",
    "            \n",
    "            if phase=='train':\n",
    "                cnn_loss.backward()\n",
    "                cnn_optimizer.step()\n",
    "\n",
    "            samples += labels.size(0)\n",
    "            running_loss += cnn_loss.item()*labels.size(0)\n",
    "            _, predicted = torch.max(outputs, 1)\n",
    "            running_acc += (predicted == labels).sum().item()\n",
    "            if phase=='train':\n",
    "                if i%100==0:\n",
    "                    print('Batch {}/{}: loss: {}, acc: {}'.format(str(i),str(num_batches),str(running_loss/samples)[:6],str(running_acc/samples)[:6]))\n",
    "        epoch_loss = float(running_loss)/samples\n",
    "        epoch_acc = float(running_acc)/samples\n",
    "        history_cnn[phase+'_acc'].append(epoch_acc)\n",
    "        history_cnn[phase+'_loss'].append(epoch_loss)\n",
    "        print('Epoch {}:  {} loss: {}, {} acc: {}'.format(str(epoch),phase,str(epoch_loss)[:6],phase,str(epoch_acc)[:6]))\n",
    "        if phase=='val':\n",
    "            if epoch_acc>best_cnn_val_acc:\n",
    "                print('Saving model, val acc improved from {} to {}'.format(str(best_cnn_val_acc)[:6],str(epoch_acc)[:6]))\n",
    "                best_cnn_val_acc = epoch_acc\n",
    "                best_cnn_model = copy.deepcopy(cnn_model)\n",
    "                torch.save(cnn_model.state_dict(),dest_path_best_model)\n",
    "            else:\n",
    "                print('val acc did not improve from {}'.format(str(best_cnn_val_acc)[:6]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history_cnn['train_acc'],c='red')\n",
    "plt.plot(history_cnn['val_acc'],c='blue')\n",
    "plt.legend([\"train\", \"validation\"], loc =\"lower right\")\n",
    "plt.title('CNN accuracy')\n",
    "plt.show()\n",
    "plt.clf()\n",
    "\n",
    "plt.plot(history_cnn['train_loss'],c='red')\n",
    "plt.plot(history_cnn['val_loss'],c='blue')\n",
    "plt.legend([\"train\", \"validation\"], loc =\"lower right\")\n",
    "plt.title('CNN loss')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing and comparing both models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "FC model test accuracy:  0.5146\n",
      "CNN model test accuracy:  0.61224\n"
     ]
    }
   ],
   "source": [
    "##FC model\n",
    "fc_model = FullyConnectedNet()\n",
    "fc_model.load_state_dict(torch.load('FC_best_model.pth'))\n",
    "fc_model.eval()\n",
    "with torch.no_grad():\n",
    "    for i, data_btch in enumerate(datatloaders['test']):\n",
    "            inputs, labels = data_btch\n",
    "            outputs = fc_model(inputs)\n",
    "            _, predicted = torch.max(outputs, 1)\n",
    "            samples += labels.size(0)\n",
    "            running_acc += (predicted == labels).sum().item()\n",
    "    acc = running_acc/samples\n",
    "    print('FC model test accuracy: ',acc)\n",
    "            \n",
    "        \n",
    "        \n",
    "cnn_model = CNNNet()\n",
    "cnn_model.load_state_dict(torch.load('CNN_best_model.pth'))\n",
    "cnn_model.eval()\n",
    "with torch.no_grad():\n",
    "    for i, data_btch in enumerate(datatloaders['test']):\n",
    "            inputs, labels = data_btch\n",
    "            outputs = cnn_model(inputs)\n",
    "            _, predicted = torch.max(outputs, 1)\n",
    "            samples += labels.size(0)\n",
    "            running_acc += (predicted == labels).sum().item()\n",
    "    acc = running_acc/samples\n",
    "    print('CNN model test accuracy: ',acc)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}