commit image classification phase 1

backend/ImageSearch.ipynb

@@ -452,6 +452,184 @@
    "source": [
     "## **STEP 2 : Define Loss / Optimizer / Metrics**"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "vscode": {
+     "languageId": "plaintext"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "optimizer = optim.Adam(\n",
+    "                    model.parameters(), \n",
+    "                    lr = learning_rate\n",
+    "                      )\n",
+    "\n",
+    "class_weights = dataset.compute_class_weights()\n",
+    "\n",
+    "criterion = nn.CrossEntropyLoss(weight = class_weights)\n",
+    "\n",
+    "def accuracy(Y, Y_hat):\n",
+    "    correct_pairs = torch.sum(Y == Y_hat).item()\n",
+    "    return correct_pairs/ len(Y_hat)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## **STEP 3 : Model Training**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "vscode": {
+     "languageId": "plaintext"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "Train_Loss = []\n",
+    "Train_Accuracy = []\n",
+    "\n",
+    "Test_Loss = []\n",
+    "Test_Accuracy = []\n",
+    "    \n",
+    "def train_loop():    \n",
+    "    for epoch in range(epoches):\n",
+    "\n",
+    "        train_loss = 0\n",
+    "        train_accuracy = 0 \n",
+    "\n",
+    "        test_loss = 0\n",
+    "        test_accuracy = 0 \n",
+    "\n",
+    "        model.train()\n",
+    "\n",
+    "        for Xtrain, Ytrain in tqdm.tqdm(train_loader):\n",
+    "\n",
+    "            optimizer.zero_grad()\n",
+    "            \n",
+    "            Xtrain = Xtrain.to(device)\n",
+    "            Ytrain = Ytrain.type(torch.LongTensor).to(device)\n",
+    "\n",
+    "            Ytrain_hat = model(Xtrain).to(device)\n",
+    "\n",
+    "            loss =  criterion(Ytrain_hat, Ytrain)   \n",
+    "            loss.backward()\n",
+    "\n",
+    "            optimizer.step()\n",
+    "\n",
+    "            Ytrain_hat = torch.argmax(Ytrain_hat, dim=1)\n",
+    "            Ytrain_hat = Ytrain_hat.type(torch.LongTensor).to(device)\n",
+    "\n",
+    "            train_loss += loss.item()\n",
+    "            train_accuracy += accuracy(Ytrain, Ytrain_hat)\n",
+    "\n",
+    "        train_loss /= len(train_loader)\n",
+    "        train_accuracy /= len(train_loader)\n",
+    "\n",
+    "        Train_Loss.append(train_loss)\n",
+    "        Train_Accuracy.append(train_accuracy)\n",
+    "\n",
+    "        model.eval()\n",
+    "\n",
+    "        with torch.no_grad():\n",
+    "            for Xtest, Ytest in test_loader:\n",
+    "\n",
+    "                Xtest = Xtest.to(device)\n",
+    "                Ytest = Ytest.type(torch.LongTensor).to(device)\n",
+    "\n",
+    "                Ytest_hat = model(Xtest).to(device)\n",
+    "\n",
+    "                loss =  criterion(Ytest_hat, Ytest)   \n",
+    "\n",
+    "                Ytest_hat = torch.argmax(Ytest_hat, dim=1)\n",
+    "                Ytest_hat = Ytest_hat.type(torch.LongTensor).to(device)\n",
+    "\n",
+    "                test_loss += loss.item()\n",
+    "                test_accuracy += accuracy(Ytest, Ytest_hat)\n",
+    "\n",
+    "        test_loss /= len(test_loader)\n",
+    "        test_accuracy /= len(test_loader)\n",
+    "\n",
+    "        Test_Loss.append(test_loss)\n",
+    "        Test_Accuracy.append(test_accuracy)\n",
+    "\n",
+    "        print('Epoch : {}  -->  Train Loss : {}  -->  Train Accuracy  -->  {}  Test Loss  -->  {}  Test Accuracy  -->  {}'.format(\n",
+    "                                                                                                                            epoch + 1, \n",
+    "                                                                                                                            round(train_loss, 3),\n",
+    "                                                                                                                            round(train_accuracy, 3), \n",
+    "                                                                                                                            round(test_loss, 3),\n",
+    "                                                                                                                            round(test_accuracy, 3)\n",
+    "                                                                                                                            ))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# **STEP 4 : Saving / Loading Model for Inference**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "vscode": {
+     "languageId": "plaintext"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "def save_model(model, PATH):\n",
+    "    torch.save(model.state_dict(), PATH)\n",
+    "    \n",
+    "def load_model(model, PATH):\n",
+    "    model.load_state_dict(torch.load(PATH))\n",
+    "    model = model.to(device)\n",
+    "    model.eval()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# **STEP 5 : Visualize Performance Metrices**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "vscode": {
+     "languageId": "plaintext"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "def visualize_performance():\n",
+    "    plt.plot(Train_Loss, color = 'r', label='train')\n",
+    "    plt.plot(Test_Loss, color = 'b', label='test')\n",
+    "    plt.xlabel('Epoch ID')\n",
+    "    plt.ylabel('Loss')\n",
+    "    plt.title('Loss Vairation')\n",
+    "    plt.legend()\n",
+    "    plt.show()\n",
+    "    \n",
+    "    plt.plot(Train_Accuracy, color = 'r', label='train')\n",
+    "    plt.plot(Test_Accuracy, color = 'b', label='test')\n",
+    "    plt.xlabel('Epoch ID')\n",
+    "    plt.ylabel('Accuracy')\n",
+    "    plt.title('Accuracy Vairation')\n",
+    "    plt.legend()\n",
+    "    plt.show()"
+   ]
   }
  ],
  "metadata": {