ia/ml-labs/Laboratorul_4_Nearest_Neighbors.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Laboratorul 4 Nearest Neighbors.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "p5NIiW7xXteX",
        "colab_type": "code",
        "outputId": "44a9f244-ccc4-4ba9-d19a-991c2a80d596",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 215
        }
      },
      "source": [
        "!wget https://fmi-unibuc-ia.github.io/ia/Data/data_MNIST.zip"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-04-27 06:42:41--  https://fmi-unibuc-ia.github.io/ia/Data/data_MNIST.zip\n",
            "Resolving fmi-unibuc-ia.github.io (fmi-unibuc-ia.github.io)... 185.199.108.153, 185.199.110.153, 185.199.109.153, ...\n",
            "Connecting to fmi-unibuc-ia.github.io (fmi-unibuc-ia.github.io)|185.199.108.153|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 629644 (615K) [application/zip]\n",
            "Saving to: ‘data_MNIST.zip’\n",
            "\n",
            "\rdata_MNIST.zip        0%[                    ]       0  --.-KB/s               \rdata_MNIST.zip      100%[===================>] 614.89K  --.-KB/s    in 0.07s   \n",
            "\n",
            "2020-04-27 06:42:42 (8.50 MB/s) - ‘data_MNIST.zip’ saved [629644/629644]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jyp-qk0hX4bz",
        "colab_type": "code",
        "outputId": "17ed5598-934d-41c2-ea45-c3bddc550ebd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 107
        }
      },
      "source": [
        "!unzip data_MNIST.zip"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Archive:  data_MNIST.zip\n",
            "  inflating: data/test_images.txt    \n",
            "  inflating: data/test_labels.txt    \n",
            "  inflating: data/train_images.txt   \n",
            "  inflating: data/train_labels.txt   \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JBlXX-44Y9n0",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "train_images = np.loadtxt('data/train_images.txt')\n",
        "train_labels = np.loadtxt('data/train_labels.txt').astype(np.int64)\n",
        "\n",
        "test_images = np.loadtxt('data/test_images.txt')\n",
        "test_labels = np.loadtxt('data/test_labels.txt').astype(np.int64)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qKbMu6qcX5n-",
        "colab_type": "code",
        "outputId": "340a4af5-acc4-41fa-81ea-aa19c861d017",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 283
        }
      },
      "source": [
        "class KNNClassifier:\n",
        "    def __init__(self, train_images, train_labels):\n",
        "        self.train_images = train_images\n",
        "        self.train_labels = train_labels\n",
        "\n",
        "    def classify_image(self, test_image, num_neighbors=3, metric='l2'):\n",
        "        if metric == 'l2':\n",
        "            distances = np.sum((train_images - test_image) ** 2, axis=-1)\n",
        "        elif metric == 'l1':\n",
        "            distances = np.sum(np.abs(train_images - test_image), axis=-1)\n",
        "\n",
        "        sorted_indexes = np.argsort(distances)\n",
        "\n",
        "        top_neighbors = self.train_labels[sorted_indexes[:num_neighbors]]\n",
        "\n",
        "        class_counts = np.bincount(top_neighbors)\n",
        "        \n",
        "        return np.argmax(class_counts)\n",
        "\n",
        "clf = KNNClassifier(train_images, train_labels)\n",
        "print('Predicted class:', clf.classify_image(train_images[0]))\n",
        "\n",
        "plt.imshow(train_images[0].reshape(28, 28), cmap='gray')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Predicted class: 4\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rz211w7kecPg",
        "colab_type": "code",
        "outputId": "093e7c13-d1f8-477d-d140-0efd0fe351b8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "predictions = []\n",
        "for test_image in test_images:\n",
        "    pred_label = clf.classify_image(test_image)\n",
        "    predictions.append(pred_label)\n",
        "\n",
        "pred_labels = np.array(predictions)\n",
        "\n",
        "np.savetxt('predictii_3nn_l2_mnist.txt', pred_labels)\n",
        "\n",
        "correct_count = np.sum(pred_labels == test_labels)\n",
        "total_count = len(test_labels)\n",
        "\n",
        "accuracy = correct_count / total_count\n",
        "\n",
        "print(f'Accuracy: {accuracy * 100}%')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy: 89.8%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "51vlzbfVhmcN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "ks = [1, 3, 5, 7, 9]\n",
        "\n",
        "def compute_accuracy(metric):\n",
        "    accuracies = []\n",
        "\n",
        "    for k in ks:\n",
        "        correct_count = 0\n",
        "        for test_image, test_label in zip(test_images, test_labels):\n",
        "            pred_label = clf.classify_image(test_image, num_neighbors=k, metric=metric)\n",
        "            if pred_label == test_label:\n",
        "                correct_count += 1\n",
        "\n",
        "        accuracy = correct_count / len(test_images)\n",
        "            \n",
        "        accuracies.append(accuracy)\n",
        "\n",
        "    np.savetxt(f'acuratete_{metric}.txt', np.array(accuracies))\n",
        "\n",
        "compute_accuracy('l2')\n",
        "compute_accuracy('l1')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-he3iL7ykYHq",
        "colab_type": "code",
        "outputId": "3fc89056-a226-498a-c6ba-7dd283e88b93",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        }
      },
      "source": [
        "l2_accs = np.loadtxt('acuratete_l2.txt')\n",
        "l1_accs = np.loadtxt('acuratete_l1.txt')\n",
        "\n",
        "plt.plot(l2_accs, label='L2')\n",
        "plt.plot(l1_accs, label='L1')\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}