CNN_Model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import torch\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "import glob\n",
    "from sklearn.model_selection import train_test_split\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from time import time\n",
    "from torchsummary import summary\n",
    "import math\n",
    "import copy\n",
    "import csv\n",
    "from sklearn.metrics import confusion_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<torch._C.Generator at 0x23519cf0cf0>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lr = 0.001\n",
    "numepoch = 50\n",
    "seed = 0\n",
    "optimizer = torch.optim.Adam\n",
    "loss_function = nn.CrossEntropyLoss()\n",
    "batch_size = 20\n",
    "kernel_size = 5\n",
    "\n",
    "num_kernel1 = 60\n",
    "num_kernel2 = 30\n",
    "first_layer_neurons = 500\n",
    "second_layer_neurons = 150\n",
    "\n",
    "torch.manual_seed(seed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "classes = ('English', 'Mandarin', 'Persian')\n",
    "\n",
    "# Get dataset\n",
    "train_data = torchvision.datasets.ImageFolder(root=\"./train/\", transform=transforms.ToTensor())\n",
    "valid_data = torchvision.datasets.ImageFolder(root=\"./valid/\", transform=transforms.ToTensor())\n",
    "test_data = torchvision.datasets.ImageFolder(root=\"./test/\", transform=transforms.ToTensor())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Net(nn.Module):\n",
    "    \n",
    "    def __init__(self, num_kernel1, num_kernel2, kernel_size, first_layer_neurons, second_layer_neurons):\n",
    "        super(Net, self).__init__()\n",
    "        self.conv1 = nn.Conv2d(3, num_kernel1, kernel_size = kernel_size)\n",
    "        self.conv2 = nn.Conv2d(num_kernel1, num_kernel2, kernel_size = kernel_size)\n",
    "\n",
    "        self.pool = nn.MaxPool2d(2,2)\n",
    "        self.drop_out = nn.Dropout()\n",
    "        \n",
    "        self.img_size = [64,108]\n",
    "        self.img_width = int((int((108 - kernel_size + 1)/2) - kernel_size + 1)/2)\n",
    "        self.img_height = int((int((64 - kernel_size + 1)/2) - kernel_size + 1)/2)\n",
    "        \n",
    "        self.img_size = num_kernel2*self.img_height*self.img_width\n",
    "        \n",
    "        self.fc1 = nn.Linear(self.img_size, first_layer_neurons)\n",
    "        self.fc2 = nn.Linear(first_layer_neurons, second_layer_neurons)\n",
    "        self.fc3 = nn.Linear(second_layer_neurons, 3)\n",
    "        self.fc1_BN = nn.BatchNorm1d(first_layer_neurons)\n",
    "        self.fc2_BN = nn.BatchNorm1d(second_layer_neurons)\n",
    "        \n",
    "    def forward(self, x):\n",
    "        x = self.pool(F.relu(self.conv1(x)))\n",
    "        x = self.pool(F.relu(self.conv2(x)))\n",
    "        \n",
    "        x = x.view(-1,self.img_size)\n",
    "        x = self.drop_out(x)\n",
    "        \n",
    "        x = F.relu(self.fc1_BN(self.fc1(x)))\n",
    "        x = F.relu(self.fc2_BN(self.fc2(x)))\n",
    "        x = self.fc3(x)\n",
    "        x = F.softmax(x,dim=1)\n",
    "        return x\n",
    "\n",
    "#one hot encoding for label vector, used for MSELoss\n",
    "def onehot(labels):\n",
    "    one_hot_labels = np.zeros((len(labels),3))\n",
    "    for i in range(len(labels)):\n",
    "        one_hot_labels[i][labels[i]] = 1\n",
    "    return torch.from_numpy(one_hot_labels)\n",
    "\n",
    "#compute accuracy of the given model on data in loader\n",
    "def accuracy(net, loader):\n",
    "    correct = 0\n",
    "    total = 0\n",
    "    with torch.no_grad():\n",
    "        for data in loader:\n",
    "            images, labels = data\n",
    "            outputs = net(images)\n",
    "            _, predicted = torch.max(outputs.data, 1)\n",
    "            total += labels.size(0)\n",
    "            correct += (predicted == labels).sum().item()\n",
    "    return correct/total"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_loader = torch.utils.data.DataLoader(train_data, batch_size=batch_size, shuffle=True)\n",
    "valid_loader = torch.utils.data.DataLoader(valid_data, batch_size=batch_size)\n",
    "test_loader = torch.utils.data.DataLoader(test_data, batch_size=batch_size)\n",
    "net = Net(num_kernel1, num_kernel2, kernel_size, first_layer_neurons, second_layer_neurons)\n",
    "optimizer = torch.optim.Adam(net.parameters(),lr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "trainLossRec = []\n",
    "validLossRec = []\n",
    "testLossRec = []\n",
    "trainAccRec = []\n",
    "validAccRec = []\n",
    "testAccRec = []\n",
    "nRec = []\n",
    "startTime = time()\n",
    "best_accuV = 0\n",
    "\n",
    "for epoch in range(numepoch):\n",
    "    running_loss = 0.0\n",
    "    numbatch = 0\n",
    "    net.train()\n",
    "    for i, train_data in enumerate(train_loader, 0):\n",
    "        inputs, labels = train_data\n",
    "        optimizer.zero_grad()\n",
    "        predictT = net(inputs.float())\n",
    "        #lossT= loss_function(input=predictT.squeeze(), target=onehot(labels).float()) #MSE\n",
    "        lossT= loss_function(predictT, labels) #CrossEntropy\n",
    "        lossT.backward()\n",
    "        optimizer.step()\n",
    "        running_loss += lossT.item()\n",
    "        numbatch += 1\n",
    "    trainLossRec.append(running_loss/numbatch)\n",
    "    accuT = accuracy(net,train_loader)\n",
    "    trainAccRec.append(accuT)\n",
    "\n",
    "    running_loss = 0.0\n",
    "    numbatch = 0\n",
    "    net.eval()\n",
    "    for i, valid_data in enumerate(valid_loader):\n",
    "        inputs, labels = valid_data\n",
    "        predictV = net(inputs.float())\n",
    "        #lossV = loss_function(input=predictV.squeeze(), target=onehot(labels).float()) #MSE\n",
    "        lossV= loss_function(predictV, labels) #CrossEntropy\n",
    "        running_loss += lossV.item()\n",
    "        numbatch += 1\n",
    "    validLossRec.append(running_loss/numbatch)\n",
    "    accuV = accuracy(net,valid_loader)\n",
    "    validAccRec.append(accuV)\n",
    "\n",
    "    if (accuV > best_accuV):\n",
    "        best_accuV = accuV\n",
    "        best_model_weights = copy.deepcopy(net.state_dict())\n",
    "\n",
    "    nRec.append(epoch)\n",
    "    print(\"Epoch:\",nRec[-1]+1,\"Training Accuracy:\",accuT,\"Validation Accuracy:\",accuV)\n",
    "\n",
    "timeElapsed = time()-startTime\n",
    "print('Training Time:',timeElapsed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# PLOT\n",
    "plt.plot(nRec,trainLossRec, label='Train')\n",
    "plt.plot(nRec,validLossRec, label='Validation')\n",
    "plt.title(\"Loss vs. Epoch\")\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.legend()\n",
    "plt.show()\n",
    "\n",
    "plt.plot(nRec,trainAccRec, label='Train')\n",
    "plt.plot(nRec,validAccRec, label='Validation')\n",
    "plt.title(\"Accuracy vs. Epoch\")\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best CNN Model\n",
      "Training Accuracy:  0.9933333333333333\n",
      "Validation Accuracy:  0.36333333333333334\n",
      "Test Accuracy:  0.41333333333333333\n"
     ]
    }
   ],
   "source": [
    "print(\"Best CNN Model\")\n",
    "#net.load_state_dict(best_model_weights)\n",
    "print(\"Training Accuracy: \", accuracy(net, train_loader))\n",
    "print(\"Validation Accuracy: \", accuracy(net, valid_loader))\n",
    "print(\"Test Accuracy: \", accuracy(net, test_loader))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#torch.save(net.state_dict(), 'Best_CNN_Model.pt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label:  tensor([1, 1, 0, 2, 1, 0, 2, 1, 0, 0, 1, 0, 0, 0, 2, 1, 1, 1, 1, 2])\n",
      "prediction:  tensor([[1.2677e-04, 9.9925e-01, 6.2247e-04],\n",
      "        [1.1311e-05, 9.9872e-01, 1.2702e-03],\n",
      "        [9.8595e-01, 5.1343e-05, 1.3997e-02],\n",
      "        [4.4059e-03, 3.6951e-06, 9.9559e-01],\n",
      "        [1.4589e-03, 9.1997e-01, 7.8567e-02],\n",
      "        [9.9681e-01, 1.6979e-07, 3.1874e-03],\n",
      "        [2.0086e-06, 1.0620e-04, 9.9989e-01],\n",
      "        [7.9864e-06, 9.9983e-01, 1.5793e-04],\n",
      "        [9.8437e-01, 2.4843e-07, 1.5633e-02],\n",
      "        [9.9993e-01, 3.1626e-06, 7.0347e-05],\n",
      "        [3.0851e-06, 9.9984e-01, 1.6139e-04],\n",
      "        [1.0000e+00, 6.8315e-08, 2.7662e-06],\n",
      "        [9.9948e-01, 3.2924e-07, 5.1785e-04],\n",
      "        [9.9697e-01, 2.1742e-04, 2.8146e-03],\n",
      "        [6.2046e-07, 9.6223e-05, 9.9990e-01],\n",
      "        [1.5825e-05, 9.9998e-01, 2.7043e-07],\n",
      "        [2.6767e-06, 9.9391e-01, 6.0899e-03],\n",
      "        [1.9278e-04, 9.9979e-01, 1.6888e-05],\n",
      "        [7.8916e-04, 9.6822e-01, 3.0988e-02],\n",
      "        [1.3595e-04, 1.7091e-04, 9.9969e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 0, 2, 2, 2, 2, 0, 0, 2, 2, 2, 1, 0, 2, 2, 2, 1, 2, 1])\n",
      "prediction:  tensor([[7.7131e-05, 9.9987e-01, 4.9430e-05],\n",
      "        [4.8052e-05, 2.5677e-04, 9.9970e-01],\n",
      "        [9.9974e-01, 2.4991e-05, 2.3589e-04],\n",
      "        [6.4776e-04, 1.6476e-03, 9.9770e-01],\n",
      "        [4.4289e-02, 2.6747e-02, 9.2896e-01],\n",
      "        [5.8286e-03, 7.9277e-03, 9.8624e-01],\n",
      "        [1.5771e-03, 1.4978e-04, 9.9827e-01],\n",
      "        [1.0000e+00, 8.1938e-08, 7.2819e-09],\n",
      "        [9.9996e-01, 3.5350e-05, 4.0563e-07],\n",
      "        [1.0529e-04, 8.5319e-05, 9.9981e-01],\n",
      "        [6.6946e-04, 6.6995e-05, 9.9926e-01],\n",
      "        [7.1706e-01, 1.2950e-05, 2.8292e-01],\n",
      "        [1.9225e-06, 9.9991e-01, 8.4570e-05],\n",
      "        [9.9944e-01, 1.0822e-04, 4.5597e-04],\n",
      "        [1.7999e-03, 1.2479e-01, 8.7341e-01],\n",
      "        [2.6370e-03, 7.6949e-03, 9.8967e-01],\n",
      "        [9.1193e-04, 5.7473e-04, 9.9851e-01],\n",
      "        [4.6419e-08, 1.0000e+00, 6.0258e-09],\n",
      "        [2.8319e-04, 4.3507e-03, 9.9537e-01],\n",
      "        [3.4759e-05, 9.9996e-01, 1.0054e-06]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 0, 1, 0, 0, 2, 1, 2, 2, 1, 2, 2, 1, 1, 0, 1, 0, 1, 0, 2])\n",
      "prediction:  tensor([[2.6874e-03, 4.0624e-03, 9.9325e-01],\n",
      "        [9.9967e-01, 3.1731e-04, 1.6846e-05],\n",
      "        [1.1460e-06, 9.9983e-01, 1.7183e-04],\n",
      "        [6.4520e-01, 2.4257e-04, 3.5456e-01],\n",
      "        [9.9995e-01, 2.6953e-08, 5.4122e-05],\n",
      "        [8.7012e-05, 1.9713e-04, 9.9972e-01],\n",
      "        [2.6098e-06, 9.9984e-01, 1.5316e-04],\n",
      "        [6.4690e-05, 9.4849e-05, 9.9984e-01],\n",
      "        [1.3097e-03, 5.2491e-06, 9.9869e-01],\n",
      "        [2.2842e-04, 9.9882e-01, 9.4877e-04],\n",
      "        [7.0947e-05, 3.6440e-05, 9.9989e-01],\n",
      "        [9.4280e-06, 8.3718e-03, 9.9162e-01],\n",
      "        [3.9753e-05, 9.9989e-01, 6.6395e-05],\n",
      "        [1.6102e-05, 9.9978e-01, 2.0474e-04],\n",
      "        [9.9997e-01, 1.1471e-05, 1.7251e-05],\n",
      "        [1.1895e-06, 9.9991e-01, 9.2928e-05],\n",
      "        [1.0000e+00, 5.6990e-07, 3.7799e-06],\n",
      "        [3.6875e-06, 9.9985e-01, 1.4680e-04],\n",
      "        [1.0000e+00, 2.1472e-07, 2.0992e-06],\n",
      "        [4.4395e-03, 2.2555e-05, 9.9554e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 2, 0, 0, 0, 2, 2, 2, 2, 0, 2, 1, 2, 2, 2, 0, 2, 2, 1, 1])\n",
      "prediction:  tensor([[3.1024e-03, 7.5013e-05, 9.9682e-01],\n",
      "        [2.1884e-05, 1.8860e-04, 9.9979e-01],\n",
      "        [9.9998e-01, 1.7204e-05, 3.5364e-06],\n",
      "        [9.9999e-01, 1.2173e-05, 3.5065e-08],\n",
      "        [9.6968e-01, 9.0906e-04, 2.9413e-02],\n",
      "        [1.0716e-03, 1.9960e-04, 9.9873e-01],\n",
      "        [9.2244e-05, 4.5231e-04, 9.9946e-01],\n",
      "        [9.5733e-04, 8.4025e-05, 9.9896e-01],\n",
      "        [2.0887e-04, 2.2229e-03, 9.9757e-01],\n",
      "        [9.9998e-01, 1.3807e-05, 1.0659e-05],\n",
      "        [9.4376e-07, 5.0361e-05, 9.9995e-01],\n",
      "        [1.4370e-08, 1.0000e+00, 9.8550e-08],\n",
      "        [1.3591e-02, 1.4683e-02, 9.7173e-01],\n",
      "        [4.2933e-02, 1.5215e-01, 8.0492e-01],\n",
      "        [1.4474e-03, 2.3952e-03, 9.9616e-01],\n",
      "        [9.9999e-01, 6.8397e-06, 6.0698e-08],\n",
      "        [2.3264e-02, 9.8271e-03, 9.6691e-01],\n",
      "        [4.8663e-03, 1.2885e-03, 9.9385e-01],\n",
      "        [2.4259e-06, 1.0000e+00, 4.0865e-07],\n",
      "        [1.1591e-07, 1.0000e+00, 3.5561e-07]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 1, 1, 2, 2, 1, 1, 1, 0, 1, 0, 0, 2, 0, 1, 0, 2, 2, 2, 1])\n",
      "prediction:  tensor([[3.7637e-05, 9.9957e-01, 3.9331e-04],\n",
      "        [6.7955e-05, 9.9645e-01, 3.4809e-03],\n",
      "        [9.7062e-08, 9.9839e-01, 1.6116e-03],\n",
      "        [1.2531e-04, 1.7074e-05, 9.9986e-01],\n",
      "        [2.1562e-05, 6.3078e-06, 9.9997e-01],\n",
      "        [7.0423e-04, 9.9845e-01, 8.4543e-04],\n",
      "        [9.5665e-08, 9.9967e-01, 3.2553e-04],\n",
      "        [8.5735e-05, 9.9971e-01, 2.0131e-04],\n",
      "        [9.9995e-01, 8.8229e-11, 5.2364e-05],\n",
      "        [4.1412e-04, 9.9955e-01, 3.2713e-05],\n",
      "        [9.9999e-01, 2.7117e-06, 4.8549e-06],\n",
      "        [9.9999e-01, 2.2355e-06, 7.1488e-06],\n",
      "        [1.4296e-06, 8.8269e-06, 9.9999e-01],\n",
      "        [9.9999e-01, 3.5160e-06, 9.0305e-06],\n",
      "        [5.1164e-03, 9.9361e-01, 1.2747e-03],\n",
      "        [9.9991e-01, 7.5901e-05, 1.5008e-05],\n",
      "        [2.7752e-05, 7.1609e-05, 9.9990e-01],\n",
      "        [4.7072e-04, 1.5622e-04, 9.9937e-01],\n",
      "        [2.1648e-02, 2.4238e-05, 9.7833e-01],\n",
      "        [1.0827e-03, 9.7724e-01, 2.1676e-02]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 0, 2, 0, 2, 2, 1, 0, 0, 1, 2, 2, 0, 2, 1, 2, 0, 0, 1])\n",
      "prediction:  tensor([[5.4230e-06, 9.9597e-01, 4.0247e-03],\n",
      "        [1.7055e-05, 1.3839e-02, 9.8614e-01],\n",
      "        [9.9731e-01, 2.3967e-03, 2.9810e-04],\n",
      "        [2.0343e-05, 9.7417e-04, 9.9901e-01],\n",
      "        [9.9961e-01, 3.1657e-04, 7.1856e-05],\n",
      "        [8.6951e-03, 5.7572e-06, 9.9130e-01],\n",
      "        [1.3948e-03, 1.0318e-04, 9.9850e-01],\n",
      "        [2.8113e-07, 9.9999e-01, 7.7604e-06],\n",
      "        [1.0000e+00, 3.1904e-08, 3.9528e-06],\n",
      "        [9.9990e-01, 6.5147e-07, 9.7947e-05],\n",
      "        [1.7904e-06, 9.9996e-01, 4.1682e-05],\n",
      "        [7.2719e-06, 3.8096e-05, 9.9995e-01],\n",
      "        [1.5504e-05, 4.6900e-06, 9.9998e-01],\n",
      "        [9.9811e-01, 1.5919e-03, 3.0180e-04],\n",
      "        [2.7931e-02, 5.8616e-04, 9.7148e-01],\n",
      "        [9.7259e-09, 1.0000e+00, 9.7127e-07],\n",
      "        [1.1146e-05, 2.4743e-03, 9.9751e-01],\n",
      "        [9.9983e-01, 3.5983e-05, 1.3189e-04],\n",
      "        [9.9984e-01, 1.5749e-04, 7.3557e-06],\n",
      "        [2.7695e-04, 9.9913e-01, 5.9724e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 0, 1, 1, 2, 0, 1, 0, 1, 1, 0, 2, 0, 0, 0, 1, 1, 0, 0, 1])\n",
      "prediction:  tensor([[1.5032e-07, 4.0635e-07, 1.0000e+00],\n",
      "        [9.9999e-01, 4.3222e-06, 1.0221e-05],\n",
      "        [9.0703e-07, 9.9977e-01, 2.2599e-04],\n",
      "        [7.2576e-08, 9.9953e-01, 4.6833e-04],\n",
      "        [1.9950e-05, 9.1214e-06, 9.9997e-01],\n",
      "        [9.8332e-01, 4.0982e-04, 1.6268e-02],\n",
      "        [6.2128e-07, 9.9718e-01, 2.8238e-03],\n",
      "        [6.6957e-01, 9.0800e-05, 3.3034e-01],\n",
      "        [6.4473e-07, 1.0000e+00, 3.2632e-06],\n",
      "        [5.7460e-06, 9.9609e-01, 3.9030e-03],\n",
      "        [9.9922e-01, 2.2758e-05, 7.5747e-04],\n",
      "        [1.7039e-04, 3.9322e-06, 9.9983e-01],\n",
      "        [9.9933e-01, 4.4150e-06, 6.6455e-04],\n",
      "        [9.9850e-01, 6.3377e-04, 8.6357e-04],\n",
      "        [9.9967e-01, 3.3515e-05, 2.9181e-04],\n",
      "        [6.7955e-05, 9.9948e-01, 4.5606e-04],\n",
      "        [2.4853e-04, 9.7546e-01, 2.4291e-02],\n",
      "        [9.9976e-01, 4.4931e-06, 2.3833e-04],\n",
      "        [9.9991e-01, 2.1432e-07, 9.0799e-05],\n",
      "        [2.4705e-06, 9.9846e-01, 1.5381e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 0, 1, 2, 0, 1, 2, 2, 1, 1, 1, 2, 1, 2, 2, 1, 0, 2, 1])\n",
      "prediction:  tensor([[2.7253e-04, 9.9946e-01, 2.6877e-04],\n",
      "        [5.0873e-04, 2.3370e-07, 9.9949e-01],\n",
      "        [1.0000e+00, 8.8815e-09, 3.2731e-06],\n",
      "        [6.3279e-03, 9.9367e-01, 3.8635e-06],\n",
      "        [4.0847e-04, 1.3259e-05, 9.9958e-01],\n",
      "        [9.9997e-01, 5.3988e-08, 2.6574e-05],\n",
      "        [5.8574e-04, 9.9880e-01, 6.1794e-04],\n",
      "        [1.4244e-04, 2.0845e-03, 9.9777e-01],\n",
      "        [8.2349e-05, 1.1101e-02, 9.8882e-01],\n",
      "        [8.7881e-03, 9.9054e-01, 6.6814e-04],\n",
      "        [3.8635e-07, 9.9954e-01, 4.6009e-04],\n",
      "        [1.1473e-05, 9.9848e-01, 1.5043e-03],\n",
      "        [6.4130e-05, 2.7071e-05, 9.9991e-01],\n",
      "        [3.3681e-03, 9.8964e-01, 6.9895e-03],\n",
      "        [1.3404e-03, 2.1074e-07, 9.9866e-01],\n",
      "        [1.0447e-03, 1.6094e-03, 9.9735e-01],\n",
      "        [9.9978e-04, 9.9900e-01, 4.7459e-06],\n",
      "        [9.9998e-01, 1.2071e-07, 1.5091e-05],\n",
      "        [4.7499e-03, 6.6517e-03, 9.8860e-01],\n",
      "        [1.2432e-03, 9.9854e-01, 2.1784e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 0, 0, 1, 2, 1, 2, 1, 2, 1, 1, 2, 1, 0, 2, 0, 1, 1, 2, 1])\n",
      "prediction:  tensor([[9.9997e-01, 7.2043e-08, 2.6392e-05],\n",
      "        [9.9980e-01, 1.3442e-08, 1.9568e-04],\n",
      "        [1.0000e+00, 1.4590e-07, 5.3091e-07],\n",
      "        [7.4785e-05, 9.9782e-01, 2.1014e-03],\n",
      "        [2.2591e-04, 7.9112e-05, 9.9970e-01],\n",
      "        [3.7342e-02, 9.6231e-01, 3.4288e-04],\n",
      "        [6.6155e-03, 1.4360e-06, 9.9338e-01],\n",
      "        [3.9546e-07, 9.9964e-01, 3.6215e-04],\n",
      "        [1.7168e-04, 4.7999e-05, 9.9978e-01],\n",
      "        [4.5233e-06, 9.9998e-01, 1.2043e-05],\n",
      "        [2.8919e-03, 9.6019e-01, 3.6916e-02],\n",
      "        [1.3217e-03, 2.1030e-03, 9.9658e-01],\n",
      "        [3.8805e-06, 9.8201e-01, 1.7990e-02],\n",
      "        [9.9999e-01, 4.3743e-06, 9.0932e-07],\n",
      "        [2.2194e-03, 7.6033e-05, 9.9770e-01],\n",
      "        [9.9756e-01, 7.1675e-05, 2.3649e-03],\n",
      "        [1.8573e-07, 9.9997e-01, 3.4179e-05],\n",
      "        [5.0087e-05, 9.9947e-01, 4.7520e-04],\n",
      "        [2.5623e-05, 1.2077e-05, 9.9996e-01],\n",
      "        [9.7493e-05, 9.5740e-01, 4.2504e-02]], grad_fn=<SoftmaxBackward>)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label:  tensor([0, 2, 1, 0, 1, 1, 1, 2, 0, 2, 1, 0, 0, 1, 1, 2, 2, 0, 0, 2])\n",
      "prediction:  tensor([[1.0000e+00, 1.7058e-06, 7.8249e-07],\n",
      "        [1.3491e-05, 3.4340e-05, 9.9995e-01],\n",
      "        [3.5878e-04, 9.8564e-01, 1.4005e-02],\n",
      "        [9.9995e-01, 8.8998e-06, 3.9834e-05],\n",
      "        [3.0583e-05, 9.9991e-01, 6.1059e-05],\n",
      "        [1.5678e-04, 9.9947e-01, 3.6935e-04],\n",
      "        [1.9414e-07, 9.9990e-01, 1.0417e-04],\n",
      "        [4.7135e-04, 3.8465e-05, 9.9949e-01],\n",
      "        [9.9794e-01, 1.1675e-03, 8.9181e-04],\n",
      "        [1.1909e-06, 2.5450e-05, 9.9997e-01],\n",
      "        [8.4643e-05, 9.9991e-01, 9.8273e-06],\n",
      "        [1.0000e+00, 1.0280e-07, 1.4056e-06],\n",
      "        [9.9702e-01, 6.7413e-07, 2.9792e-03],\n",
      "        [1.7169e-08, 9.9999e-01, 1.1223e-05],\n",
      "        [7.3078e-06, 9.9731e-01, 2.6789e-03],\n",
      "        [4.9648e-04, 6.5686e-04, 9.9885e-01],\n",
      "        [5.9663e-05, 6.2815e-06, 9.9993e-01],\n",
      "        [9.8740e-01, 4.8238e-04, 1.2121e-02],\n",
      "        [9.9184e-01, 5.9895e-06, 8.1574e-03],\n",
      "        [1.4613e-03, 5.4304e-03, 9.9311e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 0, 0, 0, 2, 1, 1, 0, 0, 1, 0, 0, 1, 0, 2, 1, 0, 2, 2, 1])\n",
      "prediction:  tensor([[7.2889e-05, 9.9652e-01, 3.4074e-03],\n",
      "        [9.9959e-01, 5.3255e-05, 3.5934e-04],\n",
      "        [9.9909e-01, 1.6624e-07, 9.0872e-04],\n",
      "        [9.9996e-01, 8.1179e-06, 3.5655e-05],\n",
      "        [1.2510e-05, 9.3448e-05, 9.9989e-01],\n",
      "        [1.1056e-06, 9.9360e-01, 6.3961e-03],\n",
      "        [7.7621e-05, 9.9809e-01, 1.8318e-03],\n",
      "        [9.9936e-01, 1.8117e-04, 4.6067e-04],\n",
      "        [9.9775e-01, 2.2057e-05, 2.2240e-03],\n",
      "        [1.3046e-08, 9.9982e-01, 1.7600e-04],\n",
      "        [9.9998e-01, 2.0457e-05, 3.0603e-06],\n",
      "        [9.9736e-01, 8.3679e-05, 2.5559e-03],\n",
      "        [1.8590e-03, 6.9148e-01, 3.0666e-01],\n",
      "        [9.9884e-01, 4.0608e-06, 1.1588e-03],\n",
      "        [2.7062e-07, 1.2114e-07, 1.0000e+00],\n",
      "        [7.4493e-10, 1.0000e+00, 3.3776e-06],\n",
      "        [9.9913e-01, 2.3396e-04, 6.3963e-04],\n",
      "        [2.7799e-04, 3.9764e-05, 9.9968e-01],\n",
      "        [5.6621e-03, 7.4010e-03, 9.8694e-01],\n",
      "        [5.5363e-09, 9.9843e-01, 1.5721e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 0, 1, 2, 0, 2, 2, 2, 0, 0, 0, 2, 0, 1, 2, 2, 0, 1, 0, 1])\n",
      "prediction:  tensor([[9.9996e-01, 2.5499e-05, 9.7010e-06],\n",
      "        [9.7599e-01, 2.4002e-02, 1.2650e-05],\n",
      "        [6.2396e-06, 9.9997e-01, 1.9052e-05],\n",
      "        [9.4413e-06, 5.2518e-06, 9.9999e-01],\n",
      "        [9.8949e-01, 6.2304e-04, 9.8846e-03],\n",
      "        [8.9830e-06, 8.5314e-04, 9.9914e-01],\n",
      "        [7.0444e-04, 6.1848e-03, 9.9311e-01],\n",
      "        [1.0464e-04, 1.0319e-04, 9.9979e-01],\n",
      "        [9.9998e-01, 3.9152e-06, 1.4259e-05],\n",
      "        [9.9952e-01, 4.5597e-04, 1.9740e-05],\n",
      "        [9.9489e-01, 1.7289e-03, 3.3819e-03],\n",
      "        [1.5663e-06, 6.7338e-05, 9.9993e-01],\n",
      "        [9.9882e-01, 8.1564e-04, 3.6291e-04],\n",
      "        [1.2384e-04, 9.9676e-01, 3.1167e-03],\n",
      "        [8.0752e-06, 3.1416e-03, 9.9685e-01],\n",
      "        [2.1185e-04, 6.3329e-05, 9.9972e-01],\n",
      "        [9.9998e-01, 2.3363e-06, 1.3785e-05],\n",
      "        [2.9794e-09, 9.9998e-01, 1.7435e-05],\n",
      "        [9.8718e-01, 7.3254e-04, 1.2087e-02],\n",
      "        [1.1107e-06, 9.9905e-01, 9.4411e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 0, 2, 0, 1, 2, 1, 1, 1, 0, 2, 2, 0, 1, 2, 1, 0, 2, 0, 2])\n",
      "prediction:  tensor([[9.9916e-01, 3.6921e-04, 4.6674e-04],\n",
      "        [9.9987e-01, 8.9588e-05, 4.0620e-05],\n",
      "        [4.1917e-06, 1.3780e-02, 9.8622e-01],\n",
      "        [9.9998e-01, 1.0137e-05, 1.1070e-05],\n",
      "        [9.7291e-06, 9.9999e-01, 1.7080e-06],\n",
      "        [2.4982e-05, 1.5831e-06, 9.9997e-01],\n",
      "        [6.8162e-06, 9.8930e-01, 1.0691e-02],\n",
      "        [1.2992e-06, 9.9982e-01, 1.7895e-04],\n",
      "        [4.0427e-08, 9.9994e-01, 5.9216e-05],\n",
      "        [9.9998e-01, 4.8646e-07, 2.0696e-05],\n",
      "        [5.0038e-04, 1.4122e-03, 9.9809e-01],\n",
      "        [2.9051e-01, 8.0282e-07, 7.0949e-01],\n",
      "        [9.9997e-01, 5.5202e-06, 2.0659e-05],\n",
      "        [2.4276e-06, 9.9550e-01, 4.4990e-03],\n",
      "        [1.7346e-04, 3.1182e-05, 9.9980e-01],\n",
      "        [4.9988e-06, 9.9951e-01, 4.8681e-04],\n",
      "        [9.9991e-01, 1.2544e-05, 8.1547e-05],\n",
      "        [2.1813e-04, 4.6721e-05, 9.9974e-01],\n",
      "        [9.9492e-01, 4.3694e-03, 7.0867e-04],\n",
      "        [4.5563e-04, 1.4479e-03, 9.9810e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 0, 0, 0, 1, 1, 2, 1, 1, 2, 0, 0, 1, 1, 0, 2, 1, 2, 1])\n",
      "prediction:  tensor([[3.8250e-05, 9.9972e-01, 2.4548e-04],\n",
      "        [1.6135e-03, 5.6554e-04, 9.9782e-01],\n",
      "        [9.9599e-01, 3.2525e-11, 4.0097e-03],\n",
      "        [9.9969e-01, 8.1929e-11, 3.0518e-04],\n",
      "        [9.9994e-01, 4.0461e-05, 1.7265e-05],\n",
      "        [1.9443e-05, 9.8744e-01, 1.2542e-02],\n",
      "        [4.2266e-07, 9.9974e-01, 2.5903e-04],\n",
      "        [5.6476e-04, 1.3934e-03, 9.9804e-01],\n",
      "        [3.8035e-04, 9.8543e-01, 1.4194e-02],\n",
      "        [2.6794e-04, 9.8323e-01, 1.6501e-02],\n",
      "        [2.7376e-05, 5.6812e-05, 9.9992e-01],\n",
      "        [1.0000e+00, 5.6882e-07, 2.8381e-06],\n",
      "        [9.9983e-01, 1.0201e-04, 7.0455e-05],\n",
      "        [1.0398e-05, 9.9993e-01, 5.9858e-05],\n",
      "        [1.0522e-05, 9.9997e-01, 2.1990e-05],\n",
      "        [9.9986e-01, 3.9020e-05, 9.9711e-05],\n",
      "        [3.5192e-04, 1.4544e-02, 9.8510e-01],\n",
      "        [2.1533e-06, 9.9393e-01, 6.0642e-03],\n",
      "        [2.6531e-04, 1.6370e-05, 9.9972e-01],\n",
      "        [2.5714e-04, 9.9129e-01, 8.4499e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 1, 2, 2, 2, 2, 2, 1, 0, 2, 2, 2, 2, 1, 2, 0, 1, 1, 2])\n",
      "prediction:  tensor([[1.0000e+00, 5.2224e-09, 1.1963e-08],\n",
      "        [1.4620e-04, 9.9944e-01, 4.1651e-04],\n",
      "        [1.4410e-06, 9.9962e-01, 3.7398e-04],\n",
      "        [9.8937e-01, 3.2329e-04, 1.0309e-02],\n",
      "        [1.3523e-03, 2.0352e-03, 9.9661e-01],\n",
      "        [4.0175e-02, 1.6371e-05, 9.5981e-01],\n",
      "        [7.9194e-04, 5.1617e-03, 9.9405e-01],\n",
      "        [1.0831e-01, 7.7272e-04, 8.9092e-01],\n",
      "        [3.1515e-07, 1.0000e+00, 1.6077e-06],\n",
      "        [1.0000e+00, 9.3465e-08, 1.3477e-06],\n",
      "        [1.2497e-05, 1.8940e-04, 9.9980e-01],\n",
      "        [9.4242e-04, 4.8266e-03, 9.9423e-01],\n",
      "        [1.6370e-04, 1.4549e-04, 9.9969e-01],\n",
      "        [6.2694e-02, 2.4751e-03, 9.3483e-01],\n",
      "        [4.5786e-04, 9.9605e-01, 3.4960e-03],\n",
      "        [1.1287e-04, 2.4303e-05, 9.9986e-01],\n",
      "        [9.9990e-01, 9.5048e-05, 1.6661e-07],\n",
      "        [2.3367e-02, 8.9736e-01, 7.9273e-02],\n",
      "        [1.3017e-07, 9.9999e-01, 6.2197e-06],\n",
      "        [7.0606e-03, 7.0495e-03, 9.8589e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 1, 2, 1, 0, 0, 0, 0, 2, 2, 1, 1, 2, 1, 0, 2, 1, 0, 1, 0])\n",
      "prediction:  tensor([[5.3512e-07, 9.9983e-01, 1.7247e-04],\n",
      "        [3.1184e-08, 9.9963e-01, 3.6690e-04],\n",
      "        [4.3426e-07, 3.8902e-06, 1.0000e+00],\n",
      "        [6.6524e-06, 9.9976e-01, 2.3187e-04],\n",
      "        [9.9983e-01, 1.6884e-07, 1.7190e-04],\n",
      "        [9.9998e-01, 1.2047e-05, 8.7956e-06],\n",
      "        [9.9989e-01, 8.5077e-05, 2.5769e-05],\n",
      "        [9.9887e-01, 9.3034e-07, 1.1332e-03],\n",
      "        [1.1267e-03, 2.9894e-05, 9.9884e-01],\n",
      "        [2.3273e-04, 4.4800e-05, 9.9972e-01],\n",
      "        [1.9544e-05, 9.9947e-01, 5.1148e-04],\n",
      "        [1.0878e-04, 9.8968e-01, 1.0210e-02],\n",
      "        [3.5829e-04, 2.7395e-05, 9.9961e-01],\n",
      "        [1.8544e-05, 9.9831e-01, 1.6674e-03],\n",
      "        [9.9544e-01, 1.0174e-03, 3.5423e-03],\n",
      "        [9.1656e-05, 3.5362e-05, 9.9987e-01],\n",
      "        [1.4765e-06, 1.0000e+00, 2.3539e-07],\n",
      "        [9.9997e-01, 1.8146e-07, 2.6535e-05],\n",
      "        [2.0919e-04, 9.0450e-01, 9.5292e-02],\n",
      "        [9.9973e-01, 3.5375e-08, 2.6561e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 2, 0, 1, 0, 2, 1, 1, 0, 2, 1, 2, 2, 0, 2, 2, 2, 2, 1, 0])\n",
      "prediction:  tensor([[1.1883e-04, 4.4642e-05, 9.9984e-01],\n",
      "        [1.6043e-03, 4.2277e-04, 9.9797e-01],\n",
      "        [9.9995e-01, 1.4998e-05, 3.0340e-05],\n",
      "        [7.3278e-05, 9.9990e-01, 2.5242e-05],\n",
      "        [9.9930e-01, 1.6007e-04, 5.4116e-04],\n",
      "        [2.0309e-04, 6.6340e-04, 9.9913e-01],\n",
      "        [1.8374e-08, 9.9996e-01, 4.3041e-05],\n",
      "        [3.6348e-05, 9.9986e-01, 9.9916e-05],\n",
      "        [1.0000e+00, 4.1549e-06, 1.9108e-07],\n",
      "        [1.0108e-03, 3.0397e-03, 9.9595e-01],\n",
      "        [6.1223e-05, 9.9994e-01, 5.4720e-07],\n",
      "        [2.8821e-02, 2.4497e-05, 9.7115e-01],\n",
      "        [5.0150e-05, 1.1216e-04, 9.9984e-01],\n",
      "        [1.0000e+00, 8.9907e-08, 2.5155e-06],\n",
      "        [6.9020e-05, 2.8703e-03, 9.9706e-01],\n",
      "        [5.1269e-04, 6.1466e-03, 9.9334e-01],\n",
      "        [4.1087e-03, 2.3304e-04, 9.9566e-01],\n",
      "        [5.3368e-04, 5.4622e-04, 9.9892e-01],\n",
      "        [1.4805e-08, 9.9999e-01, 1.0796e-05],\n",
      "        [9.9999e-01, 8.3492e-06, 7.9074e-07]], grad_fn=<SoftmaxBackward>)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label:  tensor([0, 1, 1, 1, 2, 1, 0, 2, 0, 2, 1, 0, 2, 2, 1, 2, 1, 0, 2, 2])\n",
      "prediction:  tensor([[9.9999e-01, 8.2953e-06, 6.3765e-06],\n",
      "        [5.4994e-06, 9.9972e-01, 2.7256e-04],\n",
      "        [5.7279e-05, 9.9356e-01, 6.3849e-03],\n",
      "        [1.1295e-06, 9.9996e-01, 3.9116e-05],\n",
      "        [6.3667e-04, 6.7566e-04, 9.9869e-01],\n",
      "        [1.5025e-04, 9.9910e-01, 7.4844e-04],\n",
      "        [1.0000e+00, 1.2632e-07, 1.8793e-06],\n",
      "        [3.2067e-04, 2.6679e-05, 9.9965e-01],\n",
      "        [9.9998e-01, 1.4445e-05, 6.2689e-06],\n",
      "        [1.0950e-04, 7.4809e-06, 9.9988e-01],\n",
      "        [7.4887e-08, 9.9998e-01, 1.9070e-05],\n",
      "        [9.9680e-01, 1.2051e-05, 3.1881e-03],\n",
      "        [6.0390e-04, 8.0834e-05, 9.9932e-01],\n",
      "        [6.9261e-06, 9.0455e-06, 9.9998e-01],\n",
      "        [1.5721e-06, 1.0000e+00, 2.5477e-06],\n",
      "        [6.6670e-03, 9.5256e-04, 9.9238e-01],\n",
      "        [6.1060e-03, 9.9368e-01, 2.1182e-04],\n",
      "        [9.9998e-01, 1.4878e-05, 8.8953e-07],\n",
      "        [1.4534e-03, 3.1934e-06, 9.9854e-01],\n",
      "        [3.0845e-04, 4.3437e-04, 9.9926e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 1, 0, 2, 1, 1, 1, 0, 0, 2, 1, 2, 2, 2, 0, 2, 2, 2, 0, 1])\n",
      "prediction:  tensor([[3.5768e-07, 9.9998e-01, 1.6593e-05],\n",
      "        [6.2391e-05, 9.9596e-01, 3.9765e-03],\n",
      "        [9.9996e-01, 3.4387e-05, 8.4867e-06],\n",
      "        [6.6812e-03, 9.7420e-03, 9.8358e-01],\n",
      "        [1.0727e-04, 9.9987e-01, 2.0196e-05],\n",
      "        [4.2777e-05, 9.9994e-01, 1.6830e-05],\n",
      "        [9.0871e-04, 9.9893e-01, 1.5737e-04],\n",
      "        [1.0000e+00, 4.7027e-06, 6.0429e-08],\n",
      "        [9.9937e-01, 5.8584e-06, 6.2331e-04],\n",
      "        [3.3717e-07, 5.6798e-06, 9.9999e-01],\n",
      "        [2.0226e-05, 9.9167e-01, 8.3058e-03],\n",
      "        [5.3230e-03, 1.2733e-06, 9.9468e-01],\n",
      "        [8.3425e-04, 8.1300e-06, 9.9916e-01],\n",
      "        [3.9755e-05, 1.7323e-04, 9.9979e-01],\n",
      "        [9.9998e-01, 1.8319e-06, 2.0523e-05],\n",
      "        [8.0307e-05, 8.6843e-05, 9.9983e-01],\n",
      "        [4.1080e-05, 3.3542e-05, 9.9993e-01],\n",
      "        [2.7591e-04, 5.5170e-05, 9.9967e-01],\n",
      "        [9.9998e-01, 2.3316e-06, 1.6554e-05],\n",
      "        [4.8584e-06, 9.9999e-01, 4.8716e-06]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 2, 0, 2, 2, 1, 1, 0, 0, 2, 0, 1, 2, 1, 0, 1, 2, 2, 2, 0])\n",
      "prediction:  tensor([[9.9987e-01, 8.0880e-06, 1.2121e-04],\n",
      "        [1.3871e-05, 1.2129e-03, 9.9877e-01],\n",
      "        [9.9685e-01, 8.2374e-07, 3.1504e-03],\n",
      "        [2.6343e-05, 2.5273e-04, 9.9972e-01],\n",
      "        [1.1118e-03, 1.6909e-04, 9.9872e-01],\n",
      "        [1.8246e-06, 9.9940e-01, 5.9962e-04],\n",
      "        [2.3706e-08, 1.0000e+00, 7.1726e-07],\n",
      "        [9.9985e-01, 2.7790e-07, 1.4749e-04],\n",
      "        [9.9999e-01, 6.3166e-06, 5.6373e-06],\n",
      "        [1.1420e-06, 8.8099e-05, 9.9991e-01],\n",
      "        [9.9958e-01, 4.1624e-04, 4.3318e-07],\n",
      "        [2.3382e-04, 9.9887e-01, 8.9240e-04],\n",
      "        [5.5237e-03, 8.5930e-04, 9.9362e-01],\n",
      "        [5.2576e-07, 9.9995e-01, 4.7761e-05],\n",
      "        [9.9961e-01, 1.3858e-04, 2.5156e-04],\n",
      "        [4.4055e-07, 9.9999e-01, 5.4929e-06],\n",
      "        [2.0168e-04, 3.7776e-03, 9.9602e-01],\n",
      "        [3.1357e-04, 1.2863e-05, 9.9967e-01],\n",
      "        [8.1979e-04, 5.4156e-05, 9.9913e-01],\n",
      "        [9.9952e-01, 3.4700e-04, 1.3539e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 2, 0, 0, 2, 2, 0, 2, 1, 2, 2, 0, 2, 1, 2, 2, 0, 2, 1, 1])\n",
      "prediction:  tensor([[9.9977e-01, 2.5779e-05, 1.9990e-04],\n",
      "        [1.1990e-04, 7.7416e-04, 9.9911e-01],\n",
      "        [9.9998e-01, 4.4510e-06, 2.0165e-05],\n",
      "        [9.9999e-01, 3.5155e-06, 4.6988e-06],\n",
      "        [3.1886e-06, 7.0073e-02, 9.2992e-01],\n",
      "        [5.7477e-04, 5.6099e-04, 9.9886e-01],\n",
      "        [1.0000e+00, 2.5440e-06, 3.6643e-08],\n",
      "        [1.2478e-02, 5.9099e-05, 9.8746e-01],\n",
      "        [2.1561e-06, 9.9985e-01, 1.4803e-04],\n",
      "        [1.5344e-02, 1.7861e-04, 9.8448e-01],\n",
      "        [1.0062e-03, 3.5542e-02, 9.6345e-01],\n",
      "        [9.9478e-01, 4.6859e-03, 5.3178e-04],\n",
      "        [2.5000e-03, 1.6435e-03, 9.9586e-01],\n",
      "        [3.5537e-06, 9.9996e-01, 4.0581e-05],\n",
      "        [9.8861e-03, 1.0888e-03, 9.8903e-01],\n",
      "        [2.0839e-04, 7.6861e-03, 9.9211e-01],\n",
      "        [9.9969e-01, 1.3788e-04, 1.6720e-04],\n",
      "        [2.0612e-02, 2.5443e-03, 9.7684e-01],\n",
      "        [5.4441e-09, 9.9989e-01, 1.1405e-04],\n",
      "        [9.0992e-07, 9.9982e-01, 1.7734e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 2, 2, 1, 1, 1, 2, 2, 0, 0, 1, 1, 0, 0, 0, 2, 2, 1, 2])\n",
      "prediction:  tensor([[9.9997e-01, 2.7346e-05, 1.8054e-06],\n",
      "        [1.8381e-05, 9.9978e-01, 2.0580e-04],\n",
      "        [6.0155e-04, 1.2130e-04, 9.9928e-01],\n",
      "        [3.6849e-03, 9.4267e-05, 9.9622e-01],\n",
      "        [1.3707e-05, 9.9937e-01, 6.1383e-04],\n",
      "        [8.2904e-06, 9.9994e-01, 4.9067e-05],\n",
      "        [4.0421e-08, 9.9989e-01, 1.1184e-04],\n",
      "        [3.7291e-05, 3.4149e-05, 9.9993e-01],\n",
      "        [2.9408e-04, 5.5776e-05, 9.9965e-01],\n",
      "        [9.9994e-01, 2.5433e-05, 3.2802e-05],\n",
      "        [9.9982e-01, 1.7111e-04, 8.2289e-06],\n",
      "        [1.3633e-06, 9.9933e-01, 6.7335e-04],\n",
      "        [4.1529e-05, 9.9927e-01, 6.9023e-04],\n",
      "        [9.9961e-01, 1.1915e-08, 3.9034e-04],\n",
      "        [9.9993e-01, 4.5418e-05, 2.6455e-05],\n",
      "        [9.9999e-01, 4.7472e-08, 1.2810e-05],\n",
      "        [2.1714e-02, 1.1839e-03, 9.7710e-01],\n",
      "        [3.1633e-03, 7.4552e-04, 9.9609e-01],\n",
      "        [9.9018e-06, 9.9954e-01, 4.4523e-04],\n",
      "        [2.9523e-08, 2.5160e-06, 1.0000e+00]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 2, 2, 0, 2, 1, 2, 1, 1, 2, 1, 1, 2, 0, 1, 0, 2, 2, 1, 1])\n",
      "prediction:  tensor([[9.9985e-01, 2.6772e-06, 1.5195e-04],\n",
      "        [3.1415e-01, 2.4991e-05, 6.8582e-01],\n",
      "        [3.3654e-02, 9.8496e-03, 9.5650e-01],\n",
      "        [9.9974e-01, 1.1611e-08, 2.6338e-04],\n",
      "        [7.3445e-03, 7.8113e-05, 9.9258e-01],\n",
      "        [2.8253e-07, 9.9997e-01, 3.3616e-05],\n",
      "        [1.3993e-03, 1.3552e-04, 9.9847e-01],\n",
      "        [1.8541e-06, 1.0000e+00, 6.3699e-09],\n",
      "        [8.2541e-04, 2.8087e-01, 7.1831e-01],\n",
      "        [1.0521e-03, 6.5959e-05, 9.9888e-01],\n",
      "        [8.9269e-04, 9.9891e-01, 2.0180e-04],\n",
      "        [3.0428e-04, 9.9908e-01, 6.1909e-04],\n",
      "        [7.6709e-04, 1.1407e-04, 9.9912e-01],\n",
      "        [9.9998e-01, 4.7328e-06, 2.0144e-05],\n",
      "        [1.4665e-05, 9.9782e-01, 2.1680e-03],\n",
      "        [1.0000e+00, 6.8839e-08, 1.2912e-06],\n",
      "        [3.1931e-04, 1.2339e-02, 9.8734e-01],\n",
      "        [4.5561e-07, 2.0485e-05, 9.9998e-01],\n",
      "        [1.8845e-04, 9.9981e-01, 3.5405e-07],\n",
      "        [9.6704e-04, 5.2664e-01, 4.7239e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 2, 0, 1, 2, 2, 0, 1, 0, 2, 0, 2, 2, 2, 0, 0, 1, 1, 2, 2])\n",
      "prediction:  tensor([[3.7890e-05, 6.9976e-06, 9.9996e-01],\n",
      "        [1.9911e-04, 6.3571e-03, 9.9344e-01],\n",
      "        [9.9995e-01, 3.4680e-05, 1.7331e-05],\n",
      "        [8.2829e-09, 1.0000e+00, 1.4637e-07],\n",
      "        [3.2215e-05, 3.3522e-04, 9.9963e-01],\n",
      "        [4.0776e-04, 3.4743e-04, 9.9924e-01],\n",
      "        [9.9999e-01, 7.5846e-06, 2.9657e-06],\n",
      "        [9.9926e-07, 9.9984e-01, 1.5775e-04],\n",
      "        [9.9888e-01, 1.1062e-03, 8.8242e-06],\n",
      "        [5.4073e-02, 7.7369e-04, 9.4515e-01],\n",
      "        [9.9988e-01, 2.0839e-05, 9.7568e-05],\n",
      "        [3.8695e-02, 1.0423e-05, 9.6129e-01],\n",
      "        [1.9490e-03, 7.4969e-03, 9.9055e-01],\n",
      "        [3.1857e-04, 4.5911e-04, 9.9922e-01],\n",
      "        [9.9945e-01, 5.5403e-04, 8.2677e-07],\n",
      "        [9.9987e-01, 4.2625e-05, 8.2852e-05],\n",
      "        [3.8292e-06, 9.9854e-01, 1.4541e-03],\n",
      "        [9.5078e-08, 9.9996e-01, 4.1625e-05],\n",
      "        [5.0516e-04, 3.1112e-04, 9.9918e-01],\n",
      "        [8.4048e-03, 4.6035e-03, 9.8699e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 0, 1, 1, 0, 0, 0, 1, 0, 2, 0, 1, 1, 0, 2, 2, 1, 0, 0, 0])\n",
      "prediction:  tensor([[1.9561e-06, 9.9973e-01, 2.6985e-04],\n",
      "        [9.2488e-01, 3.6437e-04, 7.4760e-02],\n",
      "        [1.3694e-08, 9.9988e-01, 1.1968e-04],\n",
      "        [9.7249e-05, 9.9926e-01, 6.3799e-04],\n",
      "        [9.9933e-01, 4.3442e-06, 6.6245e-04],\n",
      "        [9.9912e-01, 3.9335e-05, 8.4013e-04],\n",
      "        [8.7004e-01, 9.9417e-04, 1.2897e-01],\n",
      "        [1.6938e-07, 9.9965e-01, 3.5130e-04],\n",
      "        [9.9772e-01, 1.9850e-05, 2.2582e-03],\n",
      "        [1.7036e-05, 1.7079e-04, 9.9981e-01],\n",
      "        [9.9983e-01, 1.4614e-04, 2.7489e-05],\n",
      "        [7.2310e-06, 9.9630e-01, 3.6881e-03],\n",
      "        [1.9725e-06, 9.9349e-01, 6.5129e-03],\n",
      "        [1.0000e+00, 2.6360e-06, 1.2923e-07],\n",
      "        [1.2383e-04, 7.5280e-04, 9.9912e-01],\n",
      "        [2.8254e-05, 1.2713e-05, 9.9996e-01],\n",
      "        [1.7781e-06, 9.9689e-01, 3.1047e-03],\n",
      "        [9.5238e-01, 1.5592e-06, 4.7623e-02],\n",
      "        [9.9842e-01, 1.8978e-04, 1.3885e-03],\n",
      "        [9.7736e-01, 8.2477e-06, 2.2630e-02]], grad_fn=<SoftmaxBackward>)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label:  tensor([1, 1, 2, 0, 0, 1, 0, 2, 1, 0, 1, 0, 0, 1, 2, 0, 0, 1, 1, 0])\n",
      "prediction:  tensor([[1.9557e-08, 9.9974e-01, 2.5693e-04],\n",
      "        [2.4095e-05, 9.9926e-01, 7.1482e-04],\n",
      "        [5.2307e-03, 6.2093e-08, 9.9477e-01],\n",
      "        [9.9999e-01, 4.4180e-07, 7.4048e-06],\n",
      "        [9.7396e-01, 3.2086e-03, 2.2829e-02],\n",
      "        [8.3228e-07, 9.9996e-01, 4.3956e-05],\n",
      "        [9.5396e-01, 4.3401e-02, 2.6351e-03],\n",
      "        [2.2665e-04, 8.6378e-05, 9.9969e-01],\n",
      "        [7.6522e-06, 9.8704e-01, 1.2948e-02],\n",
      "        [9.9970e-01, 1.8343e-05, 2.8397e-04],\n",
      "        [1.5136e-05, 9.5096e-01, 4.9025e-02],\n",
      "        [9.9998e-01, 1.1160e-05, 7.5241e-06],\n",
      "        [7.9160e-01, 6.9243e-06, 2.0839e-01],\n",
      "        [1.6560e-06, 9.9979e-01, 2.0833e-04],\n",
      "        [5.1650e-06, 1.8173e-06, 9.9999e-01],\n",
      "        [9.9980e-01, 3.0737e-05, 1.6576e-04],\n",
      "        [9.9976e-01, 7.7477e-07, 2.3540e-04],\n",
      "        [8.8214e-09, 9.7079e-01, 2.9211e-02],\n",
      "        [2.9484e-05, 9.9448e-01, 5.4924e-03],\n",
      "        [9.9749e-01, 6.1325e-06, 2.5053e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 1, 1, 2, 1, 1, 1, 0, 1, 2, 2, 2, 1, 2, 2, 0, 0, 0, 0, 1])\n",
      "prediction:  tensor([[6.7601e-04, 1.4625e-04, 9.9918e-01],\n",
      "        [8.1697e-06, 9.8314e-01, 1.6856e-02],\n",
      "        [8.7913e-04, 9.8452e-01, 1.4601e-02],\n",
      "        [1.2728e-04, 3.8960e-04, 9.9948e-01],\n",
      "        [3.9950e-06, 9.9999e-01, 3.5018e-06],\n",
      "        [3.4778e-03, 9.9643e-01, 9.0530e-05],\n",
      "        [1.0923e-06, 9.9987e-01, 1.3137e-04],\n",
      "        [9.9957e-01, 2.4825e-07, 4.2502e-04],\n",
      "        [3.5761e-04, 9.9962e-01, 2.5444e-05],\n",
      "        [4.4313e-06, 1.8065e-05, 9.9998e-01],\n",
      "        [2.2027e-03, 2.3470e-04, 9.9756e-01],\n",
      "        [1.3516e-05, 2.4694e-04, 9.9974e-01],\n",
      "        [1.2901e-06, 9.9960e-01, 3.9415e-04],\n",
      "        [1.7134e-03, 9.1865e-07, 9.9829e-01],\n",
      "        [1.1560e-03, 8.2592e-04, 9.9802e-01],\n",
      "        [9.9999e-01, 1.4123e-07, 1.1353e-05],\n",
      "        [9.9841e-01, 4.1215e-07, 1.5915e-03],\n",
      "        [1.0000e+00, 4.6858e-08, 2.7045e-07],\n",
      "        [9.9999e-01, 1.2708e-06, 4.8868e-06],\n",
      "        [1.3967e-06, 9.9997e-01, 2.8762e-05]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 1, 2, 0, 1, 2, 2, 0, 0, 1, 0, 2, 2, 2, 1, 1, 0, 2, 2])\n",
      "prediction:  tensor([[9.9951e-01, 3.7313e-04, 1.2006e-04],\n",
      "        [2.8052e-02, 9.5935e-01, 1.2595e-02],\n",
      "        [1.6759e-05, 9.9997e-01, 1.0601e-05],\n",
      "        [1.7820e-05, 2.4236e-06, 9.9998e-01],\n",
      "        [9.9980e-01, 1.3406e-05, 1.8828e-04],\n",
      "        [8.9271e-07, 9.8732e-01, 1.2679e-02],\n",
      "        [6.6925e-08, 1.4916e-04, 9.9985e-01],\n",
      "        [1.0759e-01, 3.1808e-04, 8.9209e-01],\n",
      "        [9.9925e-01, 7.2542e-04, 2.7197e-05],\n",
      "        [9.9945e-01, 5.1469e-04, 3.5408e-05],\n",
      "        [7.0529e-09, 1.0000e+00, 4.8712e-08],\n",
      "        [9.9998e-01, 7.4683e-06, 8.9243e-06],\n",
      "        [1.5844e-02, 3.7283e-05, 9.8412e-01],\n",
      "        [6.6928e-05, 1.2458e-05, 9.9992e-01],\n",
      "        [3.5262e-04, 8.9158e-04, 9.9876e-01],\n",
      "        [6.2969e-07, 9.9890e-01, 1.0992e-03],\n",
      "        [6.8182e-06, 9.9993e-01, 6.2763e-05],\n",
      "        [9.9998e-01, 6.0538e-06, 1.8103e-05],\n",
      "        [2.5992e-03, 1.1820e-04, 9.9728e-01],\n",
      "        [4.9513e-03, 3.2153e-05, 9.9502e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 0, 2, 0, 0, 1, 2, 0, 2, 2, 1, 2, 0, 1, 1, 0, 0, 1, 2])\n",
      "prediction:  tensor([[9.8698e-01, 6.7714e-06, 1.3017e-02],\n",
      "        [2.0283e-06, 9.9853e-01, 1.4631e-03],\n",
      "        [9.9466e-01, 1.1264e-03, 4.2131e-03],\n",
      "        [1.1886e-04, 5.3192e-04, 9.9935e-01],\n",
      "        [9.9309e-01, 3.4397e-05, 6.8806e-03],\n",
      "        [9.9996e-01, 6.2822e-06, 3.2777e-05],\n",
      "        [1.0512e-06, 9.9998e-01, 1.4281e-05],\n",
      "        [7.3048e-06, 2.0736e-03, 9.9792e-01],\n",
      "        [9.9999e-01, 8.5930e-06, 4.7954e-06],\n",
      "        [4.2182e-05, 6.8048e-05, 9.9989e-01],\n",
      "        [1.6192e-04, 5.1603e-02, 9.4824e-01],\n",
      "        [3.9623e-07, 9.9998e-01, 2.0020e-05],\n",
      "        [1.6640e-06, 3.4270e-06, 9.9999e-01],\n",
      "        [9.9999e-01, 5.9262e-08, 1.2736e-05],\n",
      "        [1.3417e-03, 9.9863e-01, 2.6593e-05],\n",
      "        [4.1454e-06, 9.9995e-01, 4.6863e-05],\n",
      "        [9.9555e-01, 1.3654e-06, 4.4480e-03],\n",
      "        [9.3446e-01, 4.5795e-04, 6.5080e-02],\n",
      "        [9.7204e-06, 9.9990e-01, 8.5452e-05],\n",
      "        [2.7211e-02, 2.3701e-04, 9.7255e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 1, 1, 2, 2, 0, 1, 1, 0, 0, 2, 2, 0, 0, 0, 0, 0, 1, 1])\n",
      "prediction:  tensor([[5.7435e-06, 9.9978e-01, 2.1153e-04],\n",
      "        [4.0994e-06, 7.9529e-04, 9.9920e-01],\n",
      "        [2.1458e-08, 1.0000e+00, 1.4861e-06],\n",
      "        [1.1638e-03, 4.9261e-01, 5.0623e-01],\n",
      "        [1.9542e-05, 6.4444e-05, 9.9992e-01],\n",
      "        [2.5435e-05, 1.8715e-04, 9.9979e-01],\n",
      "        [9.9973e-01, 2.3859e-04, 3.3053e-05],\n",
      "        [6.5941e-06, 9.9997e-01, 2.5677e-05],\n",
      "        [7.3250e-07, 9.9988e-01, 1.1852e-04],\n",
      "        [9.9771e-01, 2.2424e-05, 2.2689e-03],\n",
      "        [9.9922e-01, 5.0462e-04, 2.7787e-04],\n",
      "        [1.9352e-05, 3.1670e-06, 9.9998e-01],\n",
      "        [2.6560e-05, 5.3974e-06, 9.9997e-01],\n",
      "        [9.9817e-01, 1.8316e-04, 1.6429e-03],\n",
      "        [9.9981e-01, 1.3869e-04, 5.1423e-05],\n",
      "        [9.9991e-01, 6.1900e-09, 9.2390e-05],\n",
      "        [9.9994e-01, 3.1228e-05, 2.9477e-05],\n",
      "        [9.9985e-01, 7.6735e-08, 1.4702e-04],\n",
      "        [2.5024e-04, 9.0617e-01, 9.3577e-02],\n",
      "        [2.9295e-06, 9.9912e-01, 8.7279e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 0, 2, 2, 2, 0, 1, 0, 0, 2, 1, 1, 1, 0, 1, 2, 2, 2, 0, 1])\n",
      "prediction:  tensor([[9.9869e-01, 1.6969e-05, 1.2917e-03],\n",
      "        [9.9997e-01, 1.6948e-05, 1.5961e-05],\n",
      "        [1.3330e-03, 4.1726e-04, 9.9825e-01],\n",
      "        [2.1583e-04, 2.3560e-03, 9.9743e-01],\n",
      "        [5.1001e-05, 1.4503e-04, 9.9980e-01],\n",
      "        [9.9995e-01, 1.2644e-05, 3.9366e-05],\n",
      "        [9.6465e-07, 9.9965e-01, 3.4422e-04],\n",
      "        [9.9992e-01, 2.5610e-07, 8.3080e-05],\n",
      "        [9.9575e-01, 7.5865e-05, 4.1751e-03],\n",
      "        [5.0671e-05, 1.7584e-04, 9.9977e-01],\n",
      "        [1.1019e-05, 9.9932e-01, 6.7000e-04],\n",
      "        [1.6990e-06, 9.9951e-01, 4.8747e-04],\n",
      "        [7.6449e-05, 9.9989e-01, 3.5963e-05],\n",
      "        [9.9998e-01, 2.6250e-06, 1.4028e-05],\n",
      "        [5.1546e-06, 9.9986e-01, 1.3149e-04],\n",
      "        [2.5423e-06, 2.3483e-05, 9.9997e-01],\n",
      "        [1.0045e-02, 2.6233e-04, 9.8969e-01],\n",
      "        [3.6289e-05, 6.5522e-05, 9.9990e-01],\n",
      "        [9.9960e-01, 8.3066e-06, 3.9125e-04],\n",
      "        [6.8284e-08, 1.0000e+00, 5.7084e-08]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 2, 1, 1, 1, 0, 2, 0, 1, 1, 0, 2, 0, 2, 0, 1, 1, 0, 2, 0])\n",
      "prediction:  tensor([[6.0391e-08, 1.5208e-06, 1.0000e+00],\n",
      "        [7.7647e-05, 1.1088e-04, 9.9981e-01],\n",
      "        [2.2054e-06, 9.9667e-01, 3.3248e-03],\n",
      "        [6.9905e-08, 9.9995e-01, 4.4904e-05],\n",
      "        [4.4406e-05, 9.9992e-01, 3.8427e-05],\n",
      "        [9.9998e-01, 1.3467e-05, 1.0935e-05],\n",
      "        [1.2601e-03, 1.1894e-03, 9.9755e-01],\n",
      "        [9.9995e-01, 4.5701e-05, 2.1046e-06],\n",
      "        [4.6713e-04, 9.7233e-01, 2.7199e-02],\n",
      "        [3.5708e-07, 9.9114e-01, 8.8575e-03],\n",
      "        [9.9982e-01, 1.0256e-06, 1.8036e-04],\n",
      "        [4.0260e-04, 4.8978e-06, 9.9959e-01],\n",
      "        [9.9995e-01, 4.2694e-05, 7.1834e-06],\n",
      "        [7.6076e-03, 4.9131e-04, 9.9190e-01],\n",
      "        [9.9566e-01, 1.1110e-05, 4.3297e-03],\n",
      "        [1.9035e-07, 9.9990e-01, 9.9469e-05],\n",
      "        [1.3136e-04, 9.9987e-01, 2.0236e-06],\n",
      "        [9.9570e-01, 5.9643e-06, 4.2984e-03],\n",
      "        [5.5869e-03, 5.3708e-05, 9.9436e-01],\n",
      "        [9.9943e-01, 1.3149e-05, 5.5422e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 1, 0, 0, 1, 0, 1, 2, 0, 1, 2, 0, 0, 0, 1, 1, 1, 2, 0])\n",
      "prediction:  tensor([[1.8585e-05, 9.9997e-01, 1.0561e-05],\n",
      "        [1.8473e-03, 9.0585e-04, 9.9725e-01],\n",
      "        [8.3099e-05, 9.9784e-01, 2.0766e-03],\n",
      "        [9.9998e-01, 3.2451e-07, 1.9641e-05],\n",
      "        [9.9884e-01, 1.5834e-06, 1.1578e-03],\n",
      "        [1.1857e-05, 9.2183e-01, 7.8159e-02],\n",
      "        [9.9993e-01, 4.0018e-06, 6.6418e-05],\n",
      "        [2.7044e-08, 9.9881e-01, 1.1892e-03],\n",
      "        [6.5235e-05, 1.0005e-04, 9.9983e-01],\n",
      "        [9.8103e-01, 1.4028e-04, 1.8825e-02],\n",
      "        [9.3886e-04, 9.9883e-01, 2.2761e-04],\n",
      "        [3.5156e-07, 6.1673e-07, 1.0000e+00],\n",
      "        [9.9977e-01, 1.3283e-04, 9.5311e-05],\n",
      "        [9.9913e-01, 2.3048e-05, 8.4604e-04],\n",
      "        [9.7578e-01, 4.4347e-06, 2.4215e-02],\n",
      "        [3.4727e-10, 9.9820e-01, 1.7976e-03],\n",
      "        [2.1185e-06, 9.9985e-01, 1.5082e-04],\n",
      "        [4.1032e-04, 9.9602e-01, 3.5659e-03],\n",
      "        [1.3332e-03, 1.9018e-05, 9.9865e-01],\n",
      "        [9.9991e-01, 2.1494e-05, 6.4377e-05]], grad_fn=<SoftmaxBackward>)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label:  tensor([2, 1, 1, 0, 0, 0, 1, 1, 2, 0, 0, 1, 0, 0, 0, 2, 0, 2, 1, 0])\n",
      "prediction:  tensor([[1.9790e-05, 7.0596e-05, 9.9991e-01],\n",
      "        [8.7078e-06, 9.9799e-01, 2.0053e-03],\n",
      "        [9.4417e-08, 9.9999e-01, 1.2808e-05],\n",
      "        [9.8272e-01, 9.9335e-03, 7.3513e-03],\n",
      "        [9.9876e-01, 2.8731e-06, 1.2417e-03],\n",
      "        [9.9866e-01, 9.1649e-05, 1.2461e-03],\n",
      "        [1.1177e-05, 9.9983e-01, 1.5663e-04],\n",
      "        [9.7834e-07, 9.9966e-01, 3.4020e-04],\n",
      "        [1.0510e-05, 2.4141e-05, 9.9997e-01],\n",
      "        [9.9426e-01, 1.0974e-04, 5.6275e-03],\n",
      "        [9.9906e-01, 4.8491e-04, 4.5221e-04],\n",
      "        [2.5150e-06, 9.9702e-01, 2.9799e-03],\n",
      "        [9.8113e-01, 6.9196e-03, 1.1953e-02],\n",
      "        [9.9958e-01, 1.7408e-06, 4.1381e-04],\n",
      "        [9.9999e-01, 6.2669e-06, 5.1545e-06],\n",
      "        [3.4274e-05, 1.6272e-05, 9.9995e-01],\n",
      "        [9.9176e-01, 1.3834e-04, 8.1001e-03],\n",
      "        [2.3790e-04, 1.9196e-04, 9.9957e-01],\n",
      "        [1.1094e-10, 9.9936e-01, 6.4191e-04],\n",
      "        [9.9917e-01, 1.9363e-06, 8.2863e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 1, 0, 2, 2, 1, 1, 1, 0, 1, 0, 0, 1, 2, 2, 2, 0, 2, 2, 0])\n",
      "prediction:  tensor([[5.3021e-04, 1.5976e-03, 9.9787e-01],\n",
      "        [4.2531e-07, 1.0000e+00, 5.5298e-07],\n",
      "        [9.9996e-01, 3.9157e-05, 7.0223e-08],\n",
      "        [1.4090e-05, 6.7906e-04, 9.9931e-01],\n",
      "        [2.7233e-03, 3.6671e-04, 9.9691e-01],\n",
      "        [6.9051e-06, 9.9993e-01, 6.3056e-05],\n",
      "        [3.8686e-05, 9.9969e-01, 2.7010e-04],\n",
      "        [2.8170e-03, 9.5884e-01, 3.8348e-02],\n",
      "        [9.9826e-01, 1.6944e-06, 1.7406e-03],\n",
      "        [4.2482e-05, 9.9428e-01, 5.6769e-03],\n",
      "        [9.9997e-01, 1.0021e-06, 2.6515e-05],\n",
      "        [9.9997e-01, 1.3700e-05, 2.1235e-05],\n",
      "        [3.0311e-07, 9.9927e-01, 7.3273e-04],\n",
      "        [3.2628e-05, 3.5707e-03, 9.9640e-01],\n",
      "        [6.6500e-03, 2.3767e-05, 9.9333e-01],\n",
      "        [1.0164e-02, 4.6420e-01, 5.2564e-01],\n",
      "        [9.9999e-01, 5.5396e-08, 9.8981e-06],\n",
      "        [1.4209e-04, 8.1321e-05, 9.9978e-01],\n",
      "        [6.2508e-05, 1.2146e-02, 9.8779e-01],\n",
      "        [9.9805e-01, 2.8278e-07, 1.9519e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 2, 2, 1, 2, 1, 1, 1, 0, 2, 2, 1, 1, 0, 0, 0, 0, 2, 0, 1])\n",
      "prediction:  tensor([[9.9953e-01, 5.1645e-06, 4.6962e-04],\n",
      "        [2.8698e-04, 2.8303e-05, 9.9968e-01],\n",
      "        [1.6907e-04, 1.7793e-03, 9.9805e-01],\n",
      "        [4.5471e-08, 9.9913e-01, 8.7260e-04],\n",
      "        [8.3191e-05, 7.3896e-04, 9.9918e-01],\n",
      "        [1.4855e-06, 8.0964e-01, 1.9036e-01],\n",
      "        [9.4238e-05, 9.9925e-01, 6.5885e-04],\n",
      "        [2.0802e-04, 9.9976e-01, 2.7333e-05],\n",
      "        [9.9861e-01, 2.7094e-04, 1.1175e-03],\n",
      "        [6.5675e-05, 2.7259e-06, 9.9993e-01],\n",
      "        [1.1950e-03, 9.0083e-05, 9.9871e-01],\n",
      "        [1.9721e-06, 9.9991e-01, 8.4982e-05],\n",
      "        [9.5810e-06, 9.9993e-01, 5.6386e-05],\n",
      "        [9.9994e-01, 5.0081e-05, 7.9586e-06],\n",
      "        [9.9972e-01, 1.5009e-08, 2.7522e-04],\n",
      "        [9.9989e-01, 2.4226e-06, 1.0798e-04],\n",
      "        [1.0000e+00, 3.6514e-06, 3.0098e-07],\n",
      "        [2.0064e-03, 4.5734e-06, 9.9799e-01],\n",
      "        [9.7431e-01, 1.8411e-02, 7.2764e-03],\n",
      "        [7.3521e-07, 9.9964e-01, 3.5544e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 0, 1, 1, 2, 2, 1, 2, 1, 0, 1, 0, 0, 1, 2, 0, 1, 0, 0])\n",
      "prediction:  tensor([[9.9994e-01, 4.4847e-06, 5.1967e-05],\n",
      "        [8.4893e-06, 9.9901e-01, 9.8562e-04],\n",
      "        [9.9006e-01, 1.0348e-05, 9.9297e-03],\n",
      "        [5.2747e-05, 9.9381e-01, 6.1377e-03],\n",
      "        [8.5927e-04, 7.8809e-01, 2.1105e-01],\n",
      "        [1.6577e-07, 5.7429e-06, 9.9999e-01],\n",
      "        [2.5979e-05, 1.0900e-04, 9.9987e-01],\n",
      "        [5.1419e-07, 9.9999e-01, 5.1748e-06],\n",
      "        [1.7768e-03, 1.1953e-05, 9.9821e-01],\n",
      "        [1.9468e-07, 9.9999e-01, 4.7543e-06],\n",
      "        [9.8277e-01, 2.1346e-05, 1.7209e-02],\n",
      "        [1.3330e-08, 9.9933e-01, 6.6857e-04],\n",
      "        [9.8914e-01, 7.9318e-05, 1.0785e-02],\n",
      "        [9.9996e-01, 1.1929e-05, 2.8942e-05],\n",
      "        [2.3146e-04, 9.9841e-01, 1.3633e-03],\n",
      "        [7.5146e-06, 9.4358e-06, 9.9998e-01],\n",
      "        [1.0000e+00, 1.1105e-07, 1.3372e-07],\n",
      "        [1.0084e-02, 5.4620e-01, 4.4371e-01],\n",
      "        [9.6548e-01, 3.8077e-03, 3.0712e-02],\n",
      "        [9.9999e-01, 1.9465e-06, 4.8387e-06]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 0, 0, 1, 2, 0, 2, 0, 0, 0, 1, 0, 0, 2, 0, 1, 2, 2, 0])\n",
      "prediction:  tensor([[9.9792e-01, 9.6085e-04, 1.1177e-03],\n",
      "        [8.6608e-13, 9.9998e-01, 2.4314e-05],\n",
      "        [9.8299e-01, 5.0187e-08, 1.7013e-02],\n",
      "        [9.9203e-01, 7.7341e-03, 2.3963e-04],\n",
      "        [3.8367e-10, 9.9971e-01, 2.9008e-04],\n",
      "        [2.0631e-04, 1.9295e-04, 9.9960e-01],\n",
      "        [9.9940e-01, 1.8294e-04, 4.1953e-04],\n",
      "        [2.4383e-05, 2.1600e-03, 9.9782e-01],\n",
      "        [9.9955e-01, 1.4734e-05, 4.3763e-04],\n",
      "        [9.9901e-01, 6.2743e-04, 3.6146e-04],\n",
      "        [9.8012e-01, 1.7245e-02, 2.6323e-03],\n",
      "        [1.0089e-04, 9.9693e-01, 2.9680e-03],\n",
      "        [9.9663e-01, 1.0266e-04, 3.2637e-03],\n",
      "        [9.9885e-01, 4.4042e-04, 7.1001e-04],\n",
      "        [3.0302e-03, 1.7997e-03, 9.9517e-01],\n",
      "        [9.9908e-01, 2.6933e-04, 6.4986e-04],\n",
      "        [2.8465e-06, 8.8511e-01, 1.1489e-01],\n",
      "        [4.0717e-05, 1.9689e-02, 9.8027e-01],\n",
      "        [5.2325e-06, 5.1378e-05, 9.9994e-01],\n",
      "        [9.9952e-01, 1.5300e-04, 3.3136e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([1, 2, 2, 1, 0, 0, 0, 1, 1, 2, 1, 0, 0, 2, 1, 1, 2, 2, 1, 0])\n",
      "prediction:  tensor([[5.8068e-06, 9.9979e-01, 2.0788e-04],\n",
      "        [7.1747e-03, 1.2520e-05, 9.9281e-01],\n",
      "        [6.2253e-04, 1.6091e-03, 9.9777e-01],\n",
      "        [3.1581e-04, 9.9847e-01, 1.2119e-03],\n",
      "        [9.9981e-01, 1.9089e-07, 1.8591e-04],\n",
      "        [9.9921e-01, 6.1942e-04, 1.7090e-04],\n",
      "        [9.9994e-01, 2.9840e-06, 5.8777e-05],\n",
      "        [1.1409e-04, 9.9984e-01, 4.8935e-05],\n",
      "        [1.0432e-06, 9.9979e-01, 2.0472e-04],\n",
      "        [2.1876e-02, 2.0849e-05, 9.7810e-01],\n",
      "        [3.8715e-05, 9.9923e-01, 7.2869e-04],\n",
      "        [9.9998e-01, 7.4669e-07, 1.8668e-05],\n",
      "        [9.9998e-01, 5.2690e-06, 1.4385e-05],\n",
      "        [6.6839e-05, 4.1570e-06, 9.9993e-01],\n",
      "        [5.9088e-07, 9.9033e-01, 9.6643e-03],\n",
      "        [6.0685e-08, 9.9956e-01, 4.3763e-04],\n",
      "        [9.0372e-05, 1.9922e-04, 9.9971e-01],\n",
      "        [7.4696e-06, 1.0799e-04, 9.9988e-01],\n",
      "        [5.5175e-06, 9.9988e-01, 1.1899e-04],\n",
      "        [9.9998e-01, 1.1314e-08, 2.2932e-05]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 2, 1, 1, 1, 0, 1, 1, 1, 0, 0, 1, 2, 1, 0, 0, 0, 1, 2, 1])\n",
      "prediction:  tensor([[1.0563e-02, 6.7808e-08, 9.8944e-01],\n",
      "        [6.2407e-05, 7.7608e-04, 9.9916e-01],\n",
      "        [3.5140e-07, 9.9985e-01, 1.5385e-04],\n",
      "        [1.2353e-04, 9.8826e-01, 1.1619e-02],\n",
      "        [1.2313e-06, 9.9922e-01, 7.7449e-04],\n",
      "        [1.0000e+00, 1.2197e-09, 2.9807e-06],\n",
      "        [2.4166e-04, 9.9960e-01, 1.5979e-04],\n",
      "        [8.2238e-05, 9.8694e-01, 1.2975e-02],\n",
      "        [1.3463e-05, 9.9866e-01, 1.3257e-03],\n",
      "        [9.9997e-01, 2.2596e-05, 1.0416e-05],\n",
      "        [9.9939e-01, 1.3001e-04, 4.8065e-04],\n",
      "        [4.8661e-03, 9.2373e-01, 7.1399e-02],\n",
      "        [3.1091e-07, 5.2329e-06, 9.9999e-01],\n",
      "        [2.8531e-06, 9.9710e-01, 2.8958e-03],\n",
      "        [9.9918e-01, 3.9985e-07, 8.1488e-04],\n",
      "        [9.9969e-01, 6.0355e-08, 3.1419e-04],\n",
      "        [9.9998e-01, 2.5303e-08, 2.1932e-05],\n",
      "        [7.8609e-05, 9.9975e-01, 1.6976e-04],\n",
      "        [9.0946e-05, 1.0476e-05, 9.9990e-01],\n",
      "        [1.3041e-04, 9.9352e-01, 6.3510e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 2, 2, 2, 0, 0, 2, 1, 0, 1, 0, 0, 1, 2, 2, 2, 2, 2, 2])\n",
      "prediction:  tensor([[9.9988e-01, 1.1197e-04, 6.1455e-06],\n",
      "        [1.2170e-07, 1.0000e+00, 5.1032e-07],\n",
      "        [1.0966e-04, 4.8131e-04, 9.9941e-01],\n",
      "        [3.7390e-04, 1.0205e-03, 9.9861e-01],\n",
      "        [3.8058e-02, 2.1493e-04, 9.6173e-01],\n",
      "        [9.9975e-01, 1.9128e-05, 2.3080e-04],\n",
      "        [9.9989e-01, 1.0333e-04, 4.4546e-06],\n",
      "        [7.6396e-04, 3.3180e-04, 9.9890e-01],\n",
      "        [3.8834e-05, 9.9981e-01, 1.5591e-04],\n",
      "        [9.9999e-01, 3.1378e-06, 3.9239e-06],\n",
      "        [1.5277e-07, 9.9997e-01, 2.8624e-05],\n",
      "        [9.9994e-01, 3.0267e-05, 3.3771e-05],\n",
      "        [9.9842e-01, 1.5785e-03, 3.7956e-06],\n",
      "        [1.0333e-05, 9.9985e-01, 1.4291e-04],\n",
      "        [6.1744e-04, 2.1016e-04, 9.9917e-01],\n",
      "        [1.8138e-05, 1.0528e-04, 9.9988e-01],\n",
      "        [4.9218e-04, 5.7853e-05, 9.9945e-01],\n",
      "        [1.5407e-05, 2.3082e-04, 9.9975e-01],\n",
      "        [1.3645e-02, 2.1654e-03, 9.8419e-01],\n",
      "        [2.9433e-05, 1.2757e-01, 8.7240e-01]], grad_fn=<SoftmaxBackward>)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label:  tensor([0, 1, 0, 0, 0, 0, 1, 2, 0, 0, 2, 1, 1, 2, 2, 0, 2, 1, 2, 1])\n",
      "prediction:  tensor([[9.8391e-01, 1.1647e-02, 4.4401e-03],\n",
      "        [1.5361e-06, 1.0000e+00, 1.0908e-06],\n",
      "        [9.9328e-01, 7.9898e-05, 6.6443e-03],\n",
      "        [9.9998e-01, 1.2078e-07, 1.8648e-05],\n",
      "        [9.9836e-01, 1.5051e-03, 1.3448e-04],\n",
      "        [9.9920e-01, 1.9791e-05, 7.8333e-04],\n",
      "        [4.2749e-07, 9.9723e-01, 2.7744e-03],\n",
      "        [2.5385e-06, 1.1007e-04, 9.9989e-01],\n",
      "        [9.9938e-01, 6.2110e-04, 3.2187e-06],\n",
      "        [9.9992e-01, 5.7227e-09, 7.7893e-05],\n",
      "        [7.5090e-06, 4.0017e-04, 9.9959e-01],\n",
      "        [3.5133e-08, 9.9990e-01, 9.5197e-05],\n",
      "        [1.1970e-03, 9.0873e-01, 9.0071e-02],\n",
      "        [1.7122e-03, 1.9755e-02, 9.7853e-01],\n",
      "        [9.5709e-04, 1.2792e-05, 9.9903e-01],\n",
      "        [9.9904e-01, 3.0893e-04, 6.5262e-04],\n",
      "        [1.3995e-03, 2.1600e-06, 9.9860e-01],\n",
      "        [1.1233e-05, 9.9982e-01, 1.6597e-04],\n",
      "        [1.1525e-03, 1.9800e-03, 9.9687e-01],\n",
      "        [6.9876e-08, 9.9678e-01, 3.2172e-03]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 1, 0, 0, 1, 2, 0, 2, 1, 1, 2, 1, 0, 2, 1, 2, 1, 2, 1, 0])\n",
      "prediction:  tensor([[9.9995e-01, 5.0535e-05, 3.9883e-06],\n",
      "        [1.7916e-06, 9.9851e-01, 1.4914e-03],\n",
      "        [9.9997e-01, 1.8166e-05, 1.1357e-05],\n",
      "        [9.9996e-01, 4.9436e-10, 3.7340e-05],\n",
      "        [1.3672e-04, 9.9862e-01, 1.2465e-03],\n",
      "        [3.8422e-04, 2.0720e-04, 9.9941e-01],\n",
      "        [9.9921e-01, 2.0055e-04, 5.8510e-04],\n",
      "        [1.1395e-04, 9.0067e-05, 9.9980e-01],\n",
      "        [1.5275e-06, 9.9814e-01, 1.8558e-03],\n",
      "        [1.1199e-04, 9.9757e-01, 2.3222e-03],\n",
      "        [4.9818e-05, 6.3204e-07, 9.9995e-01],\n",
      "        [4.0390e-07, 9.9987e-01, 1.3412e-04],\n",
      "        [9.9967e-01, 6.3651e-06, 3.2314e-04],\n",
      "        [1.6147e-04, 1.4060e-03, 9.9843e-01],\n",
      "        [1.5375e-06, 9.9993e-01, 6.4247e-05],\n",
      "        [1.5713e-05, 4.5684e-05, 9.9994e-01],\n",
      "        [2.3198e-05, 9.9994e-01, 3.6564e-05],\n",
      "        [6.7125e-02, 1.4680e-06, 9.3287e-01],\n",
      "        [1.5333e-03, 9.9834e-01, 1.3007e-04],\n",
      "        [9.9959e-01, 2.9880e-05, 3.8121e-04]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([2, 1, 2, 1, 0, 0, 0, 2, 1, 0, 2, 0, 0, 0, 1, 2, 0, 0, 0, 2])\n",
      "prediction:  tensor([[1.3016e-03, 5.4703e-04, 9.9815e-01],\n",
      "        [1.5217e-06, 9.9956e-01, 4.4191e-04],\n",
      "        [1.7745e-05, 2.1204e-05, 9.9996e-01],\n",
      "        [4.5583e-07, 9.9922e-01, 7.7719e-04],\n",
      "        [9.9934e-01, 2.8398e-05, 6.3201e-04],\n",
      "        [9.9970e-01, 6.3483e-06, 2.9604e-04],\n",
      "        [9.9959e-01, 3.7911e-04, 3.4749e-05],\n",
      "        [1.2950e-06, 5.7610e-01, 4.2390e-01],\n",
      "        [5.2745e-08, 1.0000e+00, 4.8252e-07],\n",
      "        [9.9968e-01, 5.0644e-06, 3.1712e-04],\n",
      "        [1.0815e-03, 8.2689e-04, 9.9809e-01],\n",
      "        [9.9452e-01, 2.8172e-04, 5.1942e-03],\n",
      "        [9.9865e-01, 9.0141e-05, 1.2609e-03],\n",
      "        [9.9887e-01, 9.4291e-04, 1.8316e-04],\n",
      "        [1.1267e-09, 9.9966e-01, 3.3619e-04],\n",
      "        [3.4879e-05, 3.9930e-04, 9.9957e-01],\n",
      "        [9.4087e-01, 3.4747e-02, 2.4386e-02],\n",
      "        [9.9861e-01, 2.7431e-04, 1.1121e-03],\n",
      "        [9.9837e-01, 7.6406e-04, 8.6838e-04],\n",
      "        [5.7653e-04, 1.6385e-04, 9.9926e-01]], grad_fn=<SoftmaxBackward>)\n",
      "label:  tensor([0, 0, 2, 0, 1, 0, 0, 1, 2, 1, 1, 1, 1, 2, 1, 2, 1, 2, 2, 1])\n",
      "prediction:  tensor([[9.9994e-01, 7.4461e-08, 6.2097e-05],\n",
      "        [9.9999e-01, 7.0859e-06, 5.7879e-07],\n",
      "        [2.5142e-02, 1.8107e-04, 9.7468e-01],\n",
      "        [1.0000e+00, 2.8709e-07, 4.4151e-06],\n",
      "        [3.1580e-05, 9.9758e-01, 2.3928e-03],\n",
      "        [9.9989e-01, 6.5291e-07, 1.0720e-04],\n",
      "        [9.9716e-01, 8.3668e-07, 2.8385e-03],\n",
      "        [2.4011e-07, 9.9992e-01, 8.1011e-05],\n",
      "        [4.0241e-04, 1.5967e-05, 9.9958e-01],\n",
      "        [1.0513e-03, 9.9786e-01, 1.0846e-03],\n",
      "        [1.5204e-05, 9.5141e-01, 4.8577e-02],\n",
      "        [4.2098e-05, 8.7753e-01, 1.2242e-01],\n",
      "        [8.9246e-05, 9.9987e-01, 3.8460e-05],\n",
      "        [3.8828e-05, 5.1888e-05, 9.9991e-01],\n",
      "        [1.9447e-04, 9.9366e-01, 6.1466e-03],\n",
      "        [4.9068e-04, 1.0640e-05, 9.9950e-01],\n",
      "        [1.2005e-06, 9.9522e-01, 4.7750e-03],\n",
      "        [6.2633e-04, 5.8407e-05, 9.9932e-01],\n",
      "        [5.4494e-03, 7.7848e-04, 9.9377e-01],\n",
      "        [3.3171e-05, 9.9996e-01, 2.7176e-06]], grad_fn=<SoftmaxBackward>)\n"
     ]
    }
   ],
   "source": [
    "net = Net(60,30,5,500,150)\n",
    "net.load_state_dict(torch.load(\"./Best_CNN_model.pt\"))\n",
    "for data, label in train_loader:\n",
    "    prediction = net(data.float())\n",
    "    print(\"label: \", label)\n",
    "    print(\"prediction: \", prediction)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[296   0   4]\n",
      " [  0 297   3]\n",
      " [  3   2 295]]\n"
     ]
    }
   ],
   "source": [
    "#compute confusion matrix\n",
    "ground_truth = []\n",
    "predictions = []\n",
    "with torch.no_grad():\n",
    "    for data in train_loader:\n",
    "        images, labels = data\n",
    "        outputs = net(images)\n",
    "        _, predicted = torch.max(outputs.data, 1)\n",
    "        predictions += predicted.tolist()\n",
    "        ground_truth += labels.tolist()\n",
    "print(confusion_matrix(ground_truth, predictions))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}