From 4ea5a020991975e96dd948477b176876799a9ef5 Mon Sep 17 00:00:00 2001
From: sasha <sasha_kobzev18@mail.ru>
Date: Thu, 13 Nov 2025 08:10:08 +0300
Subject: [PATCH] add .ipynb file

---
 labworks/LW2/lw2.ipynb | 474 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 474 insertions(+)
 create mode 100644 labworks/LW2/lw2.ipynb

diff --git a/labworks/LW2/lw2.ipynb b/labworks/LW2/lw2.ipynb
new file mode 100644
index 0000000..ccb6f47
--- /dev/null
+++ b/labworks/LW2/lw2.ipynb
@@ -0,0 +1,474 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "mWA2s-HPT5L9"
+   },
+   "outputs": [],
+   "source": [
+    "10 % 3"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "8_zCzPa3T_VE"
+   },
+   "outputs": [],
+   "source": [
+    "from google.colab import drive\n",
+    "drive.mount('/content/drive')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "hBLeHBeiUOAy"
+   },
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "os.chdir('/content/drive/MyDrive/Colab Notebooks/is_lab2')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "dREn2NGEUfIy"
+   },
+   "outputs": [],
+   "source": [
+    "!wget -N http://uit.mpei.ru/git/main/is_dnn/raw/branch/main/labworks/LW2/lab02_lib.py"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "FVyzYbb6Up8S"
+   },
+   "outputs": [],
+   "source": [
+    "!wget -N http://uit.mpei.ru/git/main/is_dnn/raw/branch/main/labworks/LW2/data/letter_train.txt\n",
+    "!wget -N http://uit.mpei.ru/git/main/is_dnn/raw/branch/main/labworks/LW2/data/letter_test.txt\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "QGPmkl3FU2Et"
+   },
+   "outputs": [],
+   "source": [
+    "# импорт модулей\n",
+    "import numpy as np\n",
+    "import lab02_lib as lib\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "Hlmf3aY4dbg6"
+   },
+   "source": [
+    "2. Сгенерировать индивидуальный набор двумерных данных в пространстве\n",
+    "признаков с координатами центра (k, k), где k – номер бригады. Вывести\n",
+    "полученные данные на рисунок и в консоль.\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "r7gSmNgWdXo2"
+   },
+   "outputs": [],
+   "source": [
+    "k = 10\n",
+    "data = lib.datagen(k, k, 1000, 2)\n",
+    "print('Размерность данных:', data.shape)\n",
+    "print('Пример данных:', data)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "Na7tT2TReMcu"
+   },
+   "source": [
+    "3-4. Создать и обучить автокодировщик AE1 простой архитектуры, выбрав\n",
+    "небольшое количество эпох обучения. Зафиксировать в таблице вида\n",
+    "табл.1 количество скрытых слоёв и нейронов в них.\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "emmU0JGJeOVD"
+   },
+   "outputs": [],
+   "source": [
+    "# AE1 — простая сеть, меньше эпох\n",
+    "patience = 300\n",
+    "ae1_trained, IRE1, IREth1 = lib.create_fit_save_ae(\n",
+    "    data, 'out/AE1.h5', 'out/AE1_ire_th.txt',\n",
+    "    1000, False, patience,  verbose_every_n_epochs = 100, early_stopping_delta = 0.001\n",
+    ")\n",
+    "lib.ire_plot('training', IRE1, IREth1, 'AE1')\n",
+    "print('AE1 IREth =', IREth1)\n",
+    "\n",
+    "#1\n",
+    "#1"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "1Pu38Xylsvjr"
+   },
+   "source": [
+    "5-6. Создать и обучить второй автокодировщик AE2 с усложненной\n",
+    "архитектурой, задав большее количество эпох обучения.\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "CToPhfVgs4Ya"
+   },
+   "outputs": [],
+   "source": [
+    "# AE2 — более сложная сеть, больше эпох\n",
+    "ae2_trained, IRE2, IREth2 = lib.create_fit_save_ae(\n",
+    "    data, 'out/AE2.h5', 'out/AE2_ire_th.txt',\n",
+    "    3000, False, patience, verbose_every_n_epochs = 500, early_stopping_delta = 0.001\n",
+    ")\n",
+    "lib.ire_plot('training', IRE2, IREth2, 'AE2')\n",
+    "print('AE2 IREth =', IREth2)\n",
+    "\n",
+    "# 5 3 2 3 5"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "aiMYnnw5ybt2"
+   },
+   "source": [
+    "8. Рассчитать характеристики качества обучения EDCA для AE1 и AE2.\n",
+    "Визуализировать и сравнить области пространства признаков,\n",
+    "распознаваемые автокодировщиками AE1 и AE2. Сделать вывод о\n",
+    "пригодности AE1 и AE2 для качественного обнаружения аномалий."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "N6F98jegybGj"
+   },
+   "outputs": [],
+   "source": [
+    "#AE1\n",
+    "numb_square = 20\n",
+    "xx, yy, Z1 = lib.square_calc(numb_square, data, ae1_trained, IREth1, '1', True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "Rc5-WXjC9rAk"
+   },
+   "outputs": [],
+   "source": [
+    "#AE2\n",
+    "numb_square = 20\n",
+    "xx, yy, Z2 = lib.square_calc(numb_square, data, ae2_trained, IREth2, '2', True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "id6lkBOi9y1v"
+   },
+   "outputs": [],
+   "source": [
+    "#Сравнение\n",
+    "lib.plot2in1(data, xx, yy, Z1, Z2)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "NZRB1Vg1y0ta"
+   },
+   "source": [
+    "9. Изучить сохраненный набор данных и пространство признаков. Создать\n",
+    "тестовую выборку, состоящую, как минимум, из 4ёх элементов, не\n",
+    "входящих в обучающую выборку. Элементы должны быть такими, чтобы\n",
+    "AE1 распознавал их как норму, а AE2 детектировал как аномалии."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "on7YZMauy4ZS"
+   },
+   "outputs": [],
+   "source": [
+    "with open('data_test.txt', 'w') as file:\n",
+    "    file.write(\"10.5 9.5\\n\")\n",
+    "    file.write(\"9.0 10.2\\n\")\n",
+    "    file.write(\"9.6 10.5\\n\")\n",
+    "    file.write(\"9.0 9.0\\n\")\n",
+    "data_test = np.loadtxt('data_test.txt', dtype=float)\n",
+    "print(data_test)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "8FuMOxLu_wVZ"
+   },
+   "outputs": [],
+   "source": [
+    "data_test = np.loadtxt('data_test.txt', dtype=float)\n",
+    "print(data_test)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "cib_DiqazAWM"
+   },
+   "source": [
+    "10. Применить обученные автокодировщики AE1 и AE2 к тестовым данным\n",
+    "и вывести значения ошибки реконструкции для каждого элемента тестовой\n",
+    "выборки относительно порога на график и в консоль."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "IprWT-m4zDR1"
+   },
+   "outputs": [],
+   "source": [
+    "# AE1\n",
+    "predicted_labels1, ire1 = lib.predict_ae(ae1_trained, data_test, IREth1)\n",
+    "lib.anomaly_detection_ae(predicted_labels1, ire1, IREth1)\n",
+    "lib.ire_plot('test', ire1, IREth1, 'AE1')\n",
+    "\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "3jD3wEhr5f6t"
+   },
+   "outputs": [],
+   "source": [
+    "# AE2\n",
+    "predicted_labels2, ire2 = lib.predict_ae(ae2_trained, data_test, IREth2)\n",
+    "lib.anomaly_detection_ae(predicted_labels2, ire2, IREth2)\n",
+    "lib.ire_plot('test', ire2, IREth2, 'AE2')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "67myAxUazRQ1"
+   },
+   "source": [
+    "11. Визуализировать элементы обучающей и тестовой выборки в областях\n",
+    "пространства признаков, распознаваемых автокодировщиками AE1 и AE2."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "jzFhJTrNzS4A"
+   },
+   "outputs": [],
+   "source": [
+    "lib.plot2in1_anomaly(data, xx, yy, Z1, Z2, data_test)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "hpP0MebIzZFi"
+   },
+   "source": [
+    "12-13 таблицы и выводы\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "RF2lQ2NMzdIZ"
+   },
+   "source": [
+    "Часть 2\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "R_YSpxv0zj0W"
+   },
+   "source": [
+    "2. Загрузить многомерную обучающую выборку реальных данных\n",
+    "name_train.txt."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "XRZI769HzjMQ"
+   },
+   "outputs": [],
+   "source": [
+    "train = np.loadtxt('letter_train.txt', dtype=float)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "sDimn5kIzvff"
+   },
+   "source": [
+    "3. Вывести полученные данные и их размерность в консоли."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "dbb7jXUjzzNT"
+   },
+   "outputs": [],
+   "source": [
+    "print('train.shape =', train.shape)\n",
+    "print('Пример train:', train)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "cCoweu23z4sx"
+   },
+   "source": [
+    "4-6. Создать и обучить автокодировщик с подходящей для данных\n",
+    "архитектурой. Выбрать необходимое количество эпох обучения"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "mqiCTs1H0HL5"
+   },
+   "outputs": [],
+   "source": [
+    "patience = 5000\n",
+    "ae3_trained, IRE3, IREth3 = lib.create_fit_save_ae(\n",
+    "    train, 'out/AE3.h5', 'out/AE3_ire_th.txt',\n",
+    "    100000, False, patience, verbose_every_n_epochs = 1000, early_stopping_delta = 0.001, early_stopping_value = 0.01\n",
+    ")\n",
+    "\n",
+    "\n",
+    "# 32 16 12 8 7 8 12 16 32\n",
+    "\n",
+    "# 30 24 19 15 11 15 19 24 30\n",
+    "\n",
+    "# 28 21 16 12 9 12 16 21 28\n",
+    "# 32 26 22 18 15 12 15 18 22 26 32\n",
+    "# 35 28 21 14 7 14 21 28 35"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "S-bT1nnDckWS"
+   },
+   "outputs": [],
+   "source": [
+    "lib.ire_plot('training', IRE3, IREth3, 'AE3')\n",
+    "print('AE3 IREth =', IREth3)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "HUoWlB7T4lPH"
+   },
+   "source": [
+    "8. Подать тестовую выборку на вход обученного автокодировщика для\n",
+    "обнаружения аномалий. Вывести график ошибки реконструкции\n",
+    "элементов тестовой выборки относительно порога."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "m9o31pOLG5IR"
+   },
+   "outputs": [],
+   "source": [
+    "test = np.loadtxt('letter_test.txt', dtype=float)\n",
+    "print('test.shape =', test.shape)\n",
+    "print('Пример train:', test)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "gyGWxfCt4jT5"
+   },
+   "outputs": [],
+   "source": [
+    "predicted_labels3, ire3 = lib.predict_ae(ae3_trained, test, IREth3)\n",
+    "lib.anomaly_detection_ae(predicted_labels3, ire3, IREth3)\n",
+    "lib.ire_plot('test', ire3, IREth3, 'AE3')"
+   ]
+  }
+ ],
+ "metadata": {
+  "accelerator": "GPU",
+  "colab": {
+   "authorship_tag": "ABX9TyP7I3xYtv8vwrxTVttwH659",
+   "gpuType": "T4",
+   "provenance": []
+  },
+  "kernelspec": {
+   "display_name": "Python 3",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}