Unsupervised_Learning_Health_Index_Construction_Models.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "collapsed_sections": [
        "p3TMLfmy3ccX"
      ],
      "authorship_tag": "ABX9TyPTtjNJIY5Sgnvn7CIjLHyo",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Ogunfool/Prognostics-Strategies-An-Aero-engine-Use-case/blob/main/Unsupervised_Learning_Health_Index_Construction_Models.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "This notebook contains different unsupervised learning models used for HI construction. It includes, PCA, Autoencoder and Convolutional autoencoder.\n",
        "This work is still in progress, more sophisticated unsupervised HI construction models will are currently being worked upon."
      ],
      "metadata": {
        "id": "O10PQBFMG3bN"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Import Libraries"
      ],
      "metadata": {
        "id": "EAL2YwgEwbjd"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "import scipy\n",
        "import math\n",
        "from sklearn.cluster import KMeans\n",
        "from sklearn.model_selection import train_test_split \n",
        "from sklearn.metrics import mean_absolute_error, mean_squared_error, mean_absolute_percentage_error, r2_score\n",
        "from sklearn.decomposition import PCA\n",
        "import seaborn as sns"
      ],
      "metadata": {
        "id": "L51gDtlPowvB"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "vmW7SgcfzLO8"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from keras.models import Model, Sequential\n",
        "from keras import losses, optimizers, layers, activations, regularizers\n",
        "from keras.layers import Dense, Conv1D, LSTM, GRU, Input, Concatenate, MaxPool1D, Flatten, GlobalMaxPool1D\n",
        "from keras.losses import SparseCategoricalCrossentropy, categorical_crossentropy"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.optimizers import Adam, RMSprop, Adagrad, SGD\n",
        "from keras.callbacks import ModelCheckpoint, EarlyStopping"
      ],
      "metadata": {
        "id": "tZfcDDB12DDs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "qjThQcYqA6zN"
      },
      "outputs": [],
      "source": [
        "# !pip install -q -U keras-tuner\n",
        "import keras_tuner as kt"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install git+https://github.com/tensorflow/docs"
      ],
      "metadata": {
        "id": "kMRjc0UJo_iI"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import tensorflow_docs as tfdocs\n",
        "import tensorflow_docs.modeling\n",
        "import tensorflow_docs.plots"
      ],
      "metadata": {
        "id": "OshwNxI8pGLE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Next, we'll confirm that we can connect to the GPU with tensorflow:\n",
        "device_name = tf.test.gpu_device_name()\n",
        "if device_name != '/device:GPU:0':\n",
        "  raise SystemError('GPU device not found')\n",
        "print('Found GPU at: {}'.format(device_name))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2GTXIN1z12VY",
        "outputId": "8a1c6d8c-7bbe-4143-c01e-9b120de62bed"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Found GPU at: /device:GPU:0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Helper Functions"
      ],
      "metadata": {
        "id": "WQuNt9Km8Bbj"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "np.set_printoptions(suppress=True, linewidth=100, precision=2)"
      ],
      "metadata": {
        "id": "MnQh6z29olxC"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Checkpoints - List and npy files\n",
        "# np.save() - Saves a single array in a binary numpy format\n",
        "def checkpoints(filename, checkpoint_data):\n",
        "  np.save(filename, checkpoint_data)\n",
        "  checkpoint_variable = np.load(filename + '.npy') #Load so that we always have an on-hand version of the checkpoint\n",
        "  return(checkpoint_variable)\n",
        "\n",
        "# List Checkpoint\n",
        "def list_checkpoints(filename, checkpoint_data):\n",
        "  np.save(filename, checkpoint_data, allow_pickle=True)\n",
        "  checkpoint_variable = np.load(filename + '.npy', allow_pickle=True) #Load so that we always have an on-hand version of the checkpoint\n",
        "  return(checkpoint_variable)"
      ],
      "metadata": {
        "id": "n7i_27BpmXIF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Load Data"
      ],
      "metadata": {
        "id": "00T-W-RfwfiX"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from google.colab import files\n",
        "uploaded = files.upload()"
      ],
      "metadata": {
        "id": "HW-VrJguBvYJ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Selected Feature Matrix (x_train and x_val) and Original health indices (y_train and y_val)."
      ],
      "metadata": {
        "id": "lENLBbdXJstx"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "x_train = np.load('/content/x-train.npy')\n",
        "y_train = np.load('/content/y-train.npy')\n",
        "x_val = np.load('/content/x-val.npy')\n",
        "y_val = np.load('/content/y-val.npy')"
      ],
      "metadata": {
        "id": "2H9qv7wv8Ji6"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "3D Data conversion of the selected feature matrix - \"See HI prediction models\" for how data is prepared. "
      ],
      "metadata": {
        "id": "dOEBgtmGKD6D"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "NTD_x_train = np.load('/content/NTD-X-train.npy')\n",
        "NTD_x_val = np.load('/content/NTD-X-val.npy')"
      ],
      "metadata": {
        "id": "fFwHYAstweLs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Master dataframe that contains all HI's from different models up till now. "
      ],
      "metadata": {
        "id": "mS6QfiSvKyEA"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# Load HItrain and HIval back in\n",
        "HItrain_df = pd.read_csv('/content/HItrain-df (1)')\n",
        "HIval_df = pd.read_csv('/content/HIval-df (1)')"
      ],
      "metadata": {
        "id": "QmAMCJFoIs_E"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "HItrain_df.head()"
      ],
      "metadata": {
        "id": "sasGa2bzrbLo"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Take median of the maxcycle of nearest residuals or neighbours\n",
        "# Instances maxcycle_df\n",
        "max_cycle_df = HItrain_df.groupby('Unit_id').max()['Time'].reset_index().rename(columns={'Time':'max_cycle'})\n",
        "max_cycle_df.head()"
      ],
      "metadata": {
        "id": "blbVWO5yWl_z"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "max_cycle_df.set_index('Unit_id',inplace=True)\n",
        "max_cycle_df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 237
        },
        "id": "p6yWKgvvgI4x",
        "outputId": "b6adcef5-774b-434a-ad05-f3962a355f92"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "         max_cycle\n",
              "Unit_id           \n",
              "1              149\n",
              "2              269\n",
              "3              206\n",
              "4              235\n",
              "5              154"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-04e16453-8a46-4cd3-989c-9894ce3a7f26\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>max_cycle</th>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Unit_id</th>\n",
              "      <th></th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>149</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>269</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>206</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>235</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>154</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-04e16453-8a46-4cd3-989c-9894ce3a7f26')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-04e16453-8a46-4cd3-989c-9894ce3a7f26 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-04e16453-8a46-4cd3-989c-9894ce3a7f26');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 46
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Health index constructor with PCA."
      ],
      "metadata": {
        "id": "p3TMLfmy3ccX"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Retain about 80% variance."
      ],
      "metadata": {
        "id": "0T6sM8jbHppV"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from scipy.linalg import norm\n",
        "# Reconstruction error\n",
        "def reconsruction_error(original, reconstructed):\n",
        "  return norm((original-reconstructed))"
      ],
      "metadata": {
        "id": "06fPYPFOvsLc"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Sklearn PCA automatically does mean normalization but not standardization (i.e doesn't divide by std - determine if this is required for your dataset yourself)\n",
        "pca = PCA()\n",
        "pca.fit(x_train)\n",
        "orig_transform = pca.transform(x_train)\n",
        "orig_reconstruction = pca.inverse_transform(orig_transform)  # Transform back to original shape \n",
        "error_1 = reconsruction_error(original=x_train, reconstructed=orig_reconstruction)"
      ],
      "metadata": {
        "id": "TGHyB1_FHnSr"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "orig_transform.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d3b130d1-94fd-4e3a-cd54-e1dadbfaa649",
        "id": "vWn-1vPnHnSr"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(41456, 8)"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "error_1"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "991c12fa-14bd-4717-90cb-7d917bf4d6e8",
        "id": "K0R8kTv7HnSr"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "4.397569313483957e-13"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "out_data_transform = pca.transform(x_val) # Use the n_components to transform validation data\n",
        "out_data_reconstruction = pca.inverse_transform(out_data_transform)  # Transform back to original shape \n",
        "error_2 = reconsruction_error(original=x_val, reconstructed=out_data_reconstruction)"
      ],
      "metadata": {
        "id": "tgRxzt7CHnSs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "out_data_transform.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "45a98556-8d32-401e-e71c-ddec98f78a94",
        "id": "fcN7aLzpHnSs"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(12303, 8)"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "error_2"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ee76a4b3-bb62-4718-c9f4-68a19be2b3f4",
        "id": "u8csYHylHnSs"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "2.4550701580408806e-13"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# K components was chosen such that 99% of the variance was retained\n",
        "print(pca.explained_variance_ratio_.round(2),pca.explained_variance_.round(2))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "aa06d71f-3d0f-45f4-d43e-53a65f74c428",
        "id": "YubfEJYLHnSs"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[0.55 0.22 0.06 0.06 0.05 0.03 0.03 0.01] [4.4  1.76 0.48 0.46 0.38 0.27 0.2  0.05]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Eigen Values\n",
        "contr = pca.explained_variance_ratio_.round(2)"
      ],
      "metadata": {
        "id": "HI8dbkE_edki"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's create multi-dimensional PCA dataframes"
      ],
      "metadata": {
        "id": "2ShN2Na5PN_b"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "PCA_HItrain_df = HItrain_df[['Unit_id','Time','ori_HI','Maxcycle','RUL']]\n",
        "PCA_HItrain_df.head()"
      ],
      "metadata": {
        "id": "sCC2ygvoPNeS"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "PCA_HIval_df = HIval_df[['Unit_id','Time','ori_HI','Maxcycle','RUL']]\n",
        "PCA_HIval_df.head()"
      ],
      "metadata": {
        "id": "AAEkFHX_QJDU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "PCA_HItrain_df.loc[:,('HI_PCA1','HI_PCA2','HI_PCA3')] = orig_transform[:,:3]\n",
        "PCA_HItrain_df.head()"
      ],
      "metadata": {
        "id": "3CxTc4zAQXBK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "PCA_HIval_df.loc[:,('HI_PCA1','HI_PCA2','HI_PCA3')] = out_data_transform[:,:3]\n",
        "PCA_HIval_df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "XCFhmRSWSGCW",
        "outputId": "e0a6ef6e-a539-444d-b659-254b36c8a658"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unit_id  Time    ori_HI  Maxcycle  RUL   HI_PCA1   HI_PCA2   HI_PCA3\n",
              "0      201     1  1.000000       191  190 -3.938672  0.349602  0.160232\n",
              "1      201     2  0.994737       191  189 -2.332807 -0.573674 -0.667446\n",
              "2      201     3  0.989474       191  188 -2.512478 -0.037324 -0.308155\n",
              "3      201     4  0.984211       191  187 -2.180602 -0.386085  0.215751\n",
              "4      201     5  0.978947       191  186 -3.277371  0.223124 -0.253054"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-0f5bd883-451f-4471-8ae6-0266ac096868\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Unit_id</th>\n",
              "      <th>Time</th>\n",
              "      <th>ori_HI</th>\n",
              "      <th>Maxcycle</th>\n",
              "      <th>RUL</th>\n",
              "      <th>HI_PCA1</th>\n",
              "      <th>HI_PCA2</th>\n",
              "      <th>HI_PCA3</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>201</td>\n",
              "      <td>1</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>191</td>\n",
              "      <td>190</td>\n",
              "      <td>-3.938672</td>\n",
              "      <td>0.349602</td>\n",
              "      <td>0.160232</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>201</td>\n",
              "      <td>2</td>\n",
              "      <td>0.994737</td>\n",
              "      <td>191</td>\n",
              "      <td>189</td>\n",
              "      <td>-2.332807</td>\n",
              "      <td>-0.573674</td>\n",
              "      <td>-0.667446</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>201</td>\n",
              "      <td>3</td>\n",
              "      <td>0.989474</td>\n",
              "      <td>191</td>\n",
              "      <td>188</td>\n",
              "      <td>-2.512478</td>\n",
              "      <td>-0.037324</td>\n",
              "      <td>-0.308155</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>201</td>\n",
              "      <td>4</td>\n",
              "      <td>0.984211</td>\n",
              "      <td>191</td>\n",
              "      <td>187</td>\n",
              "      <td>-2.180602</td>\n",
              "      <td>-0.386085</td>\n",
              "      <td>0.215751</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>201</td>\n",
              "      <td>5</td>\n",
              "      <td>0.978947</td>\n",
              "      <td>191</td>\n",
              "      <td>186</td>\n",
              "      <td>-3.277371</td>\n",
              "      <td>0.223124</td>\n",
              "      <td>-0.253054</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-0f5bd883-451f-4471-8ae6-0266ac096868')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-0f5bd883-451f-4471-8ae6-0266ac096868 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-0f5bd883-451f-4471-8ae6-0266ac096868');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 30
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "PCA Fusion: Linear combination weighted by the percentage of contribution of the chosen principal components - Simply, dot product of the PCA elements (matrix) and the contribution (vector)."
      ],
      "metadata": {
        "id": "04eUc3vmbY66"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# from 8 dim to 3 dim\n",
        "HI_PCA_B = np.dot(orig_transform[:,:2], contr[:2])\n",
        "HI_PCA_B.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dKEKbLz0ceg9",
        "outputId": "a43cf544-e7d7-4262-d0d1-8576badf4d6a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(41456,)"
            ]
          },
          "metadata": {},
          "execution_count": 132
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.plot(HI_PCA_B[:500]);"
      ],
      "metadata": {
        "id": "ZHLmRTPGfKi7"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Feature scaling: Scale the HI to range [0-1] - z = x-min/max-min"
      ],
      "metadata": {
        "id": "x76uwapEfgyo"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "min_x = min(HI_PCA_B)\n",
        "max_x = max(HI_PCA_B)\n",
        "range = max_x - min_x\n",
        "HI_z = (HI_PCA_B - min_x)/(range)"
      ],
      "metadata": {
        "id": "CNNXx5WMfX1_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "HI_PCA_B"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "eePYZXkbgyug",
        "outputId": "eaf18a90-8531-4c80-b1e6-e0b44180b60b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([-0.84, -0.71, -0.51, ...,  2.77,  3.61,  3.59])"
            ]
          },
          "metadata": {},
          "execution_count": 134
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "HI_z"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fovA9jl1g16f",
        "outputId": "2753748f-55be-4cb5-f1ca-39072b30d743"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(41456,)"
            ]
          },
          "metadata": {},
          "execution_count": 64
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "arr_z = np.array([HI_PCA_B,HI_z]).T\n",
        "arr_z.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gqxFNa3BiyPC",
        "outputId": "9e117dbe-9490-4187-8d17-953dc45bbd9d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(41456, 2)"
            ]
          },
          "metadata": {},
          "execution_count": 135
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "HItrain_df.loc[:,['HI_PCA_L','HI_PCA_Lz']] = arr_z\n",
        "PCA_HItrain_df.loc[:,['HI_PCA_L','HI_PCA_Lz']] = arr_z\n",
        "PCA_HItrain_df.head()"
      ],
      "metadata": {
        "id": "57TTFq4GhDAN"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Validation data feature fusion."
      ],
      "metadata": {
        "id": "0BH0wKTLjaxk"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# from 8 dim to 3 dim\n",
        "HI_PCA_B = np.dot(out_data_transform[:,:2], contr[:2])\n",
        "HI_PCA_B.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "124cbfcf-8944-4e81-c2cc-31542df142ec",
        "id": "hflAFxpTjogJ"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(12303,)"
            ]
          },
          "metadata": {},
          "execution_count": 137
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.plot(HI_PCA_B[:500]);"
      ],
      "metadata": {
        "id": "NIz1ZAyMjogK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Feature scaling: Scale the HI to range [0-1] - z = x-min/max-min"
      ],
      "metadata": {
        "id": "th5uS_HKjogK"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "min_x = min(HI_PCA_B)\n",
        "max_x = max(HI_PCA_B)\n",
        "range = max_x - min_x\n",
        "HI_z = (HI_PCA_B - min_x)/(range)"
      ],
      "metadata": {
        "id": "U3M4Gix-jogK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "arr_z = np.array([HI_PCA_B,HI_z]).T\n",
        "arr_z.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1bda7b26-87ba-4647-9ded-134a60afd7be",
        "id": "Klq71aBijogK"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(12303, 2)"
            ]
          },
          "metadata": {},
          "execution_count": 140
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "HIval_df.loc[:,['HI_PCA_L','HI_PCA_Lz']] = arr_z\n",
        "PCA_HIval_df.loc[:,['HI_PCA_L','HI_PCA_Lz']] = arr_z\n",
        "PCA_HIval_df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "oMXl_D4njogL",
        "outputId": "392ddcbe-d5da-4a5f-8fc3-6937c5a904cb"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unit_id  Time    ori_HI  Maxcycle  RUL   HI_PCA1   HI_PCA2   HI_PCA3  \\\n",
              "0      201     1  1.000000       191  190 -3.938672  0.349602  0.160232   \n",
              "1      201     2  0.994737       191  189 -2.332807 -0.573674 -0.667446   \n",
              "2      201     3  0.989474       191  188 -2.512478 -0.037324 -0.308155   \n",
              "3      201     4  0.984211       191  187 -2.180602 -0.386085  0.215751   \n",
              "4      201     5  0.978947       191  186 -3.277371  0.223124 -0.253054   \n",
              "\n",
              "   HI_PCA_L  HI_PCA_Lz  \n",
              "0 -2.089357   0.013486  \n",
              "1 -1.409252   0.090581  \n",
              "2 -1.390074   0.092755  \n",
              "3 -1.284270   0.104748  \n",
              "4 -1.753467   0.051562  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-591c7051-49ad-4a48-b40c-d6f5b7c493a5\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Unit_id</th>\n",
              "      <th>Time</th>\n",
              "      <th>ori_HI</th>\n",
              "      <th>Maxcycle</th>\n",
              "      <th>RUL</th>\n",
              "      <th>HI_PCA1</th>\n",
              "      <th>HI_PCA2</th>\n",
              "      <th>HI_PCA3</th>\n",
              "      <th>HI_PCA_L</th>\n",
              "      <th>HI_PCA_Lz</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>201</td>\n",
              "      <td>1</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>191</td>\n",
              "      <td>190</td>\n",
              "      <td>-3.938672</td>\n",
              "      <td>0.349602</td>\n",
              "      <td>0.160232</td>\n",
              "      <td>-2.089357</td>\n",
              "      <td>0.013486</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>201</td>\n",
              "      <td>2</td>\n",
              "      <td>0.994737</td>\n",
              "      <td>191</td>\n",
              "      <td>189</td>\n",
              "      <td>-2.332807</td>\n",
              "      <td>-0.573674</td>\n",
              "      <td>-0.667446</td>\n",
              "      <td>-1.409252</td>\n",
              "      <td>0.090581</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>201</td>\n",
              "      <td>3</td>\n",
              "      <td>0.989474</td>\n",
              "      <td>191</td>\n",
              "      <td>188</td>\n",
              "      <td>-2.512478</td>\n",
              "      <td>-0.037324</td>\n",
              "      <td>-0.308155</td>\n",
              "      <td>-1.390074</td>\n",
              "      <td>0.092755</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>201</td>\n",
              "      <td>4</td>\n",
              "      <td>0.984211</td>\n",
              "      <td>191</td>\n",
              "      <td>187</td>\n",
              "      <td>-2.180602</td>\n",
              "      <td>-0.386085</td>\n",
              "      <td>0.215751</td>\n",
              "      <td>-1.284270</td>\n",
              "      <td>0.104748</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>201</td>\n",
              "      <td>5</td>\n",
              "      <td>0.978947</td>\n",
              "      <td>191</td>\n",
              "      <td>186</td>\n",
              "      <td>-3.277371</td>\n",
              "      <td>0.223124</td>\n",
              "      <td>-0.253054</td>\n",
              "      <td>-1.753467</td>\n",
              "      <td>0.051562</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-591c7051-49ad-4a48-b40c-d6f5b7c493a5')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-591c7051-49ad-4a48-b40c-d6f5b7c493a5 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-591c7051-49ad-4a48-b40c-d6f5b7c493a5');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 141
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "NOTE: After the HI construction for any model, the health indices are saved to a general dataframe (for train and validation data) named comp_df. This comp_df (meaning compiled HI dataframe) is used during testing with actual test set to choose the best model. "
      ],
      "metadata": {
        "id": "MR0WfQa-EueI"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Health Index Construction with Autoencoders."
      ],
      "metadata": {
        "id": "QfZciV_oFvF3"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Train-val loss plot helper function."
      ],
      "metadata": {
        "id": "dJc5tkZYvX_d"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# Helper function that plots, saves and shows plots\n",
        "def plot_loss_accuracy(filename):\n",
        "  loss = history.history['loss']\n",
        "  val_loss = history.history['val_loss']\n",
        "\n",
        "  epochs = range(len(loss))\n",
        "\n",
        "  plt.plot(epochs, loss, 'r', label='Training loss')\n",
        "  plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
        "  plt.title('Training and validation accuracy')\n",
        "  plt.legend(loc=0)\n",
        "  plt.figure()\n",
        "\n",
        "  plt.savefig(filename + '.png')\n",
        "  plt.show()"
      ],
      "metadata": {
        "id": "8NAScmNHg70e"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Basic Autoencoder Training - Using Subclass Method.\n",
        "\n",
        "\n",
        "*   This is the simplest autoencoder, you can simply think of it as PCA. We transform the input data (D=8) to latent space (D=3) then reconstruct it back to D=8.\n"
      ],
      "metadata": {
        "id": "LW7SUfkTUXMq"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "latent_dim = 3\n",
        "K = 8\n",
        "\n",
        "class Autoencoder(Model):\n",
        "  def __init__(self, latent_dim):\n",
        "    super(Autoencoder, self).__init__()\n",
        "    self.latent_dim = latent_dim   \n",
        "    self.encoder = tf.keras.Sequential([\n",
        "      layers.Dense(5, activation='relu'),\n",
        "      layers.Dense(3, activation='relu'),\n",
        "    ])\n",
        "    self.decoder = tf.keras.Sequential([\n",
        "      layers.Dense(5, activation='relu'),\n",
        "      layers.Dense(K),  # I am not sure of the activation function thing\n",
        "    ])\n",
        "\n",
        "  def call(self, x):\n",
        "    encoded = self.encoder(x)\n",
        "    decoded = self.decoder(encoded)\n",
        "    return decoded\n",
        "\n",
        "autoencoder = Autoencoder(latent_dim)"
      ],
      "metadata": {
        "id": "_YwfHjsF4cyF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "autoencoder.compile(optimizer='adam', loss=losses.MeanSquaredError())"
      ],
      "metadata": {
        "id": "irHgOYO3acVw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "tf.keras.backend.clear_session()\n",
        "\n",
        "early_stopping = EarlyStopping(monitor='val_loss', patience=10)\n",
        "\n",
        "history = autoencoder.fit(x_train, x_train,\n",
        "                epochs=60,\n",
        "                shuffle=False,\n",
        "                callbacks=[early_stopping],\n",
        "                validation_data=(x_val, x_val))"
      ],
      "metadata": {
        "id": "bI-cd7mpfFXF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Latent representation and reconstructed validation data\n",
        "x_val_latent = autoencoder.encoder(x_val).numpy()\n",
        "x_val_reconstructed = autoencoder.decoder(x_val_latent).numpy()\n",
        "\n",
        "# Latent representation and reconstructed train data\n",
        "x_train_latent = autoencoder.encoder(x_train).numpy()\n",
        "x_train_reconstructed = autoencoder.decoder(x_train_latent).numpy()"
      ],
      "metadata": {
        "id": "5v5-GcDPiHNL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_val_latent.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1RbKDSaR7apX",
        "outputId": "ccb30ad2-0f68-43dc-ab0b-c3d6ae1df3e2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(12303, 3)"
            ]
          },
          "metadata": {},
          "execution_count": 121
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from scipy.linalg import norm\n",
        "# Reconstruction error - MSE\n",
        "def reconstruction_error(original, reconstructed):\n",
        "  return norm((original-reconstructed))"
      ],
      "metadata": {
        "id": "c8ZE_gJQmUF_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Reconstruction error\n",
        "rec_error_train = reconstruction_error(x_train, x_train_reconstructed)\n",
        "rec_error_val = reconstruction_error(x_val, x_val_reconstructed)"
      ],
      "metadata": {
        "id": "2ByBaoiqizGQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "rec_error_train"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JfXaT3sVl5mT",
        "outputId": "b21a2e78-e919-43fc-e741-1c81e117d353"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "236.37072233379027"
            ]
          },
          "metadata": {},
          "execution_count": 130
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "r2_score(x_train, x_train_reconstructed)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ne7DThtDuNZ_",
        "outputId": "e3105554-bffd-4474-fd47-e03680c8062e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.8315348852500315"
            ]
          },
          "metadata": {},
          "execution_count": 128
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plot_loss_accuracy(filename='Autoencoder')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 470
        },
        "id": "WqywIKQYnlXx",
        "outputId": "9a25116a-2384-4003-c068-670bfc32e435"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 0 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Model Evaluation: We should be focused on the the reconstruction error on the validation set to really have an idea of how well the model is behaving.In this example, the train and val losses are tight so we don't have the issue of overfitting. To quatify how much info is retained, we use the r2-score."
      ],
      "metadata": {
        "id": "x3UH-NhegPq0"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's create multi-dimensional AE dataframes."
      ],
      "metadata": {
        "id": "2Ka7WApeIs_F"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "AE_HItrain_df = HItrain_df[['Unit_id','Time','ori_HI','Maxcycle','RUL']]\n",
        "AE_HItrain_df.head()"
      ],
      "metadata": {
        "id": "27zk_sL6Is_F"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "AE_HIval_df = HIval_df[['Unit_id','Time','ori_HI','Maxcycle','RUL']]\n",
        "AE_HIval_df.head()"
      ],
      "metadata": {
        "id": "31hIYhjdIs_F"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "AE_HItrain_df.loc[:,('HI_AE1','HI_AE2','HI_AE3')] = x_train_latent\n",
        "AE_HItrain_df.head()"
      ],
      "metadata": {
        "id": "3t3qQnuTIs_F"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "AE_HIval_df.loc[:,('HI_AE1','HI_AE2','HI_AE3')] = x_val_latent\n",
        "AE_HIval_df.head()"
      ],
      "metadata": {
        "id": "fhG6KccIIs_F"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "8kgvxFloHjSH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Evaluation of above methods on validation set"
      ],
      "metadata": {
        "id": "vMv87IM2m_b5"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# This is the residual function we are using because the lent can be larger than some instance's maxcycles.....\n",
        "def residual_func(train_df, test_df, lent, name):\n",
        "  no_instances = train_df['Unit_id'].unique().shape[0]\n",
        "  res_mat = np.zeros((lent, no_instances))\n",
        "  cur_val = test_df[name][:lent]\n",
        "  for id in train_df['Unit_id'].unique():\n",
        "    # If it is already dead, then maybe you shouldn't be part of the computation OR add zeros below it\n",
        "    temp_df = train_df[train_df['Unit_id']==id]\n",
        "    if temp_df.shape[0] < lent:\n",
        "      diff_arr = np.zeros(lent- temp_df.shape[0])\n",
        "      cur_train = temp_df[name][:lent]\n",
        "      cur_train = np.concatenate((cur_train, diff_arr), axis=0)\n",
        "      res_mat[:,id-1] = cur_train\n",
        "    else:\n",
        "      temp_df = train_df[train_df['Unit_id']==id]\n",
        "      cur_train = temp_df[name][:lent]\n",
        "      res_mat[:,id-1] = cur_train\n",
        "\n",
        "  l = res_mat - cur_val.to_numpy().reshape((-1,1))\n",
        "  residual = np.sqrt(np.mean(l**2, axis=0))\n",
        "  return residual"
      ],
      "metadata": {
        "id": "fITNcQzjne76"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def lenght(test_df, pct):\n",
        " lent = math.ceil(test_df.shape[0]*pct)\n",
        " return lent\n",
        "\n",
        "def neighbors(residual,nearest):\n",
        "  n_neighbors = np.argsort(residual)[:nearest]\n",
        "  return n_neighbors"
      ],
      "metadata": {
        "id": "jI8Ppmmlne76"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "eW3ublyTne76"
      },
      "outputs": [],
      "source": [
        "#  The function for new data will be a little different from that of val data\n",
        "def RUL_estimator(train_df, test_df, max_cycle_df, pct, nearest, name):\n",
        "  # Call Lenght Function\n",
        "  lent = lenght(test_df,pct)\n",
        "\n",
        "  # Call Residual Function\n",
        "  residual = residual_func(train_df, test_df, lent, name)\n",
        "\n",
        "  # Nearest neighbours function - where you can change the size of nearest neighbours\n",
        "  n_neighbors = neighbors(residual,nearest)\n",
        "\n",
        "  # RUL Estimation\n",
        "  # We want the closest 50 that are not yet dead\n",
        "  ens_RULs = max_cycle_df.loc[n_neighbors+1] - lent\n",
        "\n",
        "  true_RUL = test_df.shape[0] - lent\n",
        "\n",
        "  m = ens_RULs[ens_RULs>-10]\n",
        "  m[m.isna()] = 0\n",
        "\n",
        "  est_RUL = m['max_cycle'].median()\n",
        "\n",
        "  return (est_RUL, true_RUL, (ens_RULs))\n",
        "  # return lent, residual, n_neighbors"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def RUL_Estimator_coll(train_df, test_df, max_cycle_df, nearest, name):\n",
        "  # Instantiate pct list:\n",
        "  pct_list = [0.5,0.7,0.9]\n",
        "  # Val_data unique ID\n",
        "  inst_id = test_df['Unit_id'].unique()\n",
        "\n",
        "  # Create a results dictionary:\n",
        "  val_inst = {}\n",
        "  for i in inst_id:\n",
        "    val_inst[i] = {}\n",
        "    for j in [0.5, 0.7, 0.9]:\n",
        "      val_inst[i][j] = (0,0,())\n",
        "\n",
        "  # Loop through all val_instances\n",
        "  for k in inst_id:\n",
        "    # Working with a single test set:\n",
        "    for pct in [0.5, 0.7, 0.9]: \n",
        "      cur_val_df = test_df[test_df['Unit_id'] == k]\n",
        "      val_inst[k][pct] = RUL_estimator(train_df=train_df, test_df=cur_val_df, max_cycle_df=max_cycle_df, pct=pct, nearest=nearest, name=name)\n",
        "\n",
        "  return val_inst\n"
      ],
      "metadata": {
        "id": "myMmzUvSne77"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def result_analysis(val_inst):\n",
        "  fif = []\n",
        "  sev = []\n",
        "  nine = []\n",
        "  EST_RUL_5 = []\n",
        "  TRUE_RUL_5 = []\n",
        "  ENS_RUL_5 = []\n",
        "  EST_RUL_7 = []\n",
        "  TRUE_RUL_7 = []\n",
        "  ENS_RUL_7 = []\n",
        "  EST_RUL_9 = []\n",
        "  TRUE_RUL_9 = []\n",
        "  ENS_RUL_9 = []\n",
        "  id_list = []\n",
        "  # loop through all instances:\n",
        "  for id,info in val_inst.items():\n",
        "    id_list.append(id)\n",
        "\n",
        "    EST_RUL_5.append(info[0.5][0])\n",
        "    TRUE_RUL_5.append(info[0.5][1])\n",
        "    ENS_RUL_5.append(info[0.5][2])\n",
        "\n",
        "    EST_RUL_7.append(info[0.7][0])\n",
        "    TRUE_RUL_7.append(info[0.7][1])\n",
        "    ENS_RUL_7.append(info[0.5][2])\n",
        "\n",
        "    EST_RUL_9.append(info[0.9][0])\n",
        "    TRUE_RUL_9.append(info[0.9][1])\n",
        "    ENS_RUL_9.append(info[0.5][2])\n",
        "\n",
        "  fif = [EST_RUL_5, TRUE_RUL_5, ENS_RUL_5]\n",
        "  sev = [EST_RUL_7, TRUE_RUL_7, ENS_RUL_7]\n",
        "  nine = [EST_RUL_9, TRUE_RUL_9, ENS_RUL_9]\n",
        "  \n",
        "  return fif, sev, nine\n",
        "  # Calculate RMSE\n",
        "\n"
      ],
      "metadata": {
        "id": "ZvozRBJZne77"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def score(errors):\n",
        "  a1=10\n",
        "  a2=13\n",
        "  s1=0\n",
        "  s2=0\n",
        "  for err in errors:\n",
        "    if err < 0:\n",
        "      s1 += (np.exp(-1*(err/a1))) - 1\n",
        "    if ((err > 0) or (err == 0)):\n",
        "      s2 += (np.exp(err/a2)) - 1\n",
        "  return [s1 , s2]"
      ],
      "metadata": {
        "id": "b9uafycane77"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def RUL_metrics(a,b,c):\n",
        "  res_list = []\n",
        "  RMSE = []\n",
        "  MAE = []\n",
        "  SCORE = []\n",
        "\n",
        "  # Make an array for easy computation\n",
        "  for res in [a,b,c]:\n",
        "    res_list.append(np.array(res[:2]).T)\n",
        "  res_array = np.array(res_list)\n",
        "\n",
        "  for r in res_array:\n",
        "    errors = r[0]-r[1]\n",
        "    RMSE.append(mean_squared_error(r[0], r[1], squared=False))\n",
        "    MAE.append(mean_absolute_error(r[0],r[1]))\n",
        "    SCORE.append(score(errors))\n",
        "\n",
        "  return RMSE, MAE, SCORE\n"
      ],
      "metadata": {
        "id": "F5ksC0D9ne77"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "HItrain_df.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "RR2iJRzjoCHf",
        "outputId": "bd64d429-7584-4008-c00a-f9c24987045a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unit_id  Time    ori_HI  Maxcycle  RUL    new_HI  apprx_HI    HI_PCA  \\\n",
              "0        1     1  1.000000       149  148  0.582822  0.546977 -0.859024   \n",
              "1        1     2  0.993243       149  147  0.582912  0.581504 -0.628524   \n",
              "2        1     3  0.986486       149  146  0.582965  0.601732 -0.404679   \n",
              "3        1     4  0.979730       149  145  0.583290  0.726264 -1.859147   \n",
              "4        1     5  0.972973       149  144  0.582879  0.568990 -0.310075   \n",
              "\n",
              "   new_HI_PCA  HI_PCA_L  HI_PCA_Lz  \n",
              "0   -1.192820 -0.794293   0.187668  \n",
              "1   -1.197747 -0.701790   0.197499  \n",
              "2   -1.202504 -0.405560   0.228983  \n",
              "3   -1.171129 -1.376195   0.125823  \n",
              "4   -1.204507 -0.534039   0.215328  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-06668608-c0e6-420b-92ca-6f8e34b29d60\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Unit_id</th>\n",
              "      <th>Time</th>\n",
              "      <th>ori_HI</th>\n",
              "      <th>Maxcycle</th>\n",
              "      <th>RUL</th>\n",
              "      <th>new_HI</th>\n",
              "      <th>apprx_HI</th>\n",
              "      <th>HI_PCA</th>\n",
              "      <th>new_HI_PCA</th>\n",
              "      <th>HI_PCA_L</th>\n",
              "      <th>HI_PCA_Lz</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>149</td>\n",
              "      <td>148</td>\n",
              "      <td>0.582822</td>\n",
              "      <td>0.546977</td>\n",
              "      <td>-0.859024</td>\n",
              "      <td>-1.192820</td>\n",
              "      <td>-0.794293</td>\n",
              "      <td>0.187668</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>2</td>\n",
              "      <td>0.993243</td>\n",
              "      <td>149</td>\n",
              "      <td>147</td>\n",
              "      <td>0.582912</td>\n",
              "      <td>0.581504</td>\n",
              "      <td>-0.628524</td>\n",
              "      <td>-1.197747</td>\n",
              "      <td>-0.701790</td>\n",
              "      <td>0.197499</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "      <td>0.986486</td>\n",
              "      <td>149</td>\n",
              "      <td>146</td>\n",
              "      <td>0.582965</td>\n",
              "      <td>0.601732</td>\n",
              "      <td>-0.404679</td>\n",
              "      <td>-1.202504</td>\n",
              "      <td>-0.405560</td>\n",
              "      <td>0.228983</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1</td>\n",
              "      <td>4</td>\n",
              "      <td>0.979730</td>\n",
              "      <td>149</td>\n",
              "      <td>145</td>\n",
              "      <td>0.583290</td>\n",
              "      <td>0.726264</td>\n",
              "      <td>-1.859147</td>\n",
              "      <td>-1.171129</td>\n",
              "      <td>-1.376195</td>\n",
              "      <td>0.125823</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1</td>\n",
              "      <td>5</td>\n",
              "      <td>0.972973</td>\n",
              "      <td>149</td>\n",
              "      <td>144</td>\n",
              "      <td>0.582879</td>\n",
              "      <td>0.568990</td>\n",
              "      <td>-0.310075</td>\n",
              "      <td>-1.204507</td>\n",
              "      <td>-0.534039</td>\n",
              "      <td>0.215328</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-06668608-c0e6-420b-92ca-6f8e34b29d60')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-06668608-c0e6-420b-92ca-6f8e34b29d60 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-06668608-c0e6-420b-92ca-6f8e34b29d60');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 99
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='ori_HI')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "wxlkuqoHnuLM"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Y7U5ZiXlorUK",
        "outputId": "fd2ab59d-d1e7-43d9-ecac-cad4322379f8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[3.2596012026013246, 2.2638462845343543, 0.3535533905932738] [3.25, 2.25, 0.25] [[0.7689263561692603, 0], [0.5054281748479112, 0], [0.05127109637602412, 0.0]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='apprx_HI')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "L6ZkviDCn0dw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6d81e2fb-925a-42dc-af1f-634520f4452e",
        "id": "vcEtT0gro077"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[12.55487952948972, 15.972632844963288, 11.667261889578034] [10.25, 12.25, 8.25] [[0.3498588075760032, 2.8426177733663716], [0.22140275816016985, 4.644994542233771], [0, 2.5581144982364594]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='HI_PCA')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "hEyILV20n9tP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1114c8fa-7bd4-470d-fa66-6e3cca44c272",
        "id": "yJ0RJvf2o5j6"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[18.85802216564611, 14.564511663629508, 10.253048327204938] [14.75, 11.25, 7.25] [[0.3498588075760032, 6.678786589066299], [0.22140275816016985, 3.840040444150536], [0, 2.0507413156095278]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full, k_components = 3\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='HI_PCA_L')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "IZdcU1NRoWEp"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "730e0deb-163f-4d03-a07a-8727b874c70c",
        "id": "TJfMO_Fmo7_W"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[20.967236346261757, 12.103718436910205, 3.5355339059327378] [16.25, 9.5, 2.5] [[0.3498588075760032, 8.671957984132842], [0.22140275816016985, 2.6976308645666607], [0, 0.46904919384901667]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full, k_components = 8\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='HI_PCA_L')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "2VJaE-OSswtL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "29bf930e-1862-47ae-946c-d81ffe099160",
        "id": "H1tInZm4swtL"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[20.967236346261757, 14.212670403551895, 3.5355339059327378] [16.25, 11.0, 2.5] [[0.3498588075760032, 8.671957984132842], [0.22140275816016985, 3.657419495658906], [0, 0.46904919384901667]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full, k_components = 5\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='HI_PCA_L')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "ubUSe0QHtEIO"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4a148564-be11-47e4-9c75-a7103211f928",
        "id": "O9SDMz-FtEIO"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[20.967236346261757, 14.212670403551895, 3.5355339059327378] [16.25, 11.0, 2.5] [[0.3498588075760032, 8.671957984132842], [0.22140275816016985, 3.657419495658906], [0, 0.46904919384901667]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full, k_components = 2\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='HI_PCA_L')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "Yp95avdlt8Za"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d1eb6f91-7b90-4788-93d9-851850c70b99",
        "id": "2vWZucA0t8Zb"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[20.967236346261757, 14.212670403551895, 3.8890872965260113] [16.25, 11.0, 2.75] [[0.3498588075760032, 8.671957984132842], [0.22140275816016985, 3.657419495658906], [0, 0.5266517261980206]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full\n",
        "val_inst = RUL_Estimator_coll(train_df=HItrain_df, test_df=HIval_df, max_cycle_df=max_cycle_df, nearest=50, name='HI_PCA_Lz')\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "KD04YyeJoWru"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1d87cb32-bc7a-452d-87d0-936996d79272",
        "id": "oxNyoJaqo_oE"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[23.430749027719962, 19.144189719076646, 9.192388155425117] [18.0, 14.5, 6.5] [[0.3498588075760032, 11.660178782560491], [0.22140275816016985, 6.979877631319886], [0, 1.718281828459045]]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Comments: We did a quick experiment with the no of k-components and realised that the best kcomponents is 3."
      ],
      "metadata": {
        "id": "U-b_cBTHEMHF"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Validation Error Scores on Autoencoder.\n",
        "\n",
        "*   Note: The Latent Space of the autoencoder contains 3 feature vectors; we simply took the sum of squares (i.e weights =1) for all features.\n",
        "*   Other methods to be tried includes: Using a latent space of size 1 or choosing the latent feature vectors that give best results or doing a weighted sum depending on the degree of relevance which can be chosen by how much weights each feature vector carries.\n",
        "\n",
        "*   The residual_func had to be refined a bit to accomodate this change.\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "a55EJc23NC1f"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Residual Function for multi-dimensional data."
      ],
      "metadata": {
        "id": "D_08UpRxKclw"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# Recall, in residual function, we are working with a single test set and all train set\n",
        "# Function will take a names list, D\n",
        "# name will be a list\n",
        "# res_mat is now 3D - NxLxD\n",
        "# cur_val is a matrix - pass list of names LxD\n",
        "# diff_arr is now a matrix LxD\n",
        "# cur_train is a matrix LxD\n",
        "# train_df and test_df are new PCA dataframe"
      ],
      "metadata": {
        "id": "Y_W9yQuOKcEa"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Test\n",
        "D = 3\n",
        "lent = 191\n",
        "train_df = AE_HItrain_df\n",
        "test_df = AE_HIval_df[AE_HIval_df['Unit_id']==201]\n",
        "name = ['HI_AE1','HI_AE2','HI_AE3']"
      ],
      "metadata": {
        "id": "bRLcrxKJTD4a"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def residual_func(train_df, test_df, lent, name):\n",
        "  no_instances = train_df['Unit_id'].unique().shape[0]\n",
        "  res_mat = np.zeros((no_instances,lent,D))\n",
        "  cur_val = test_df[name][:lent]\n",
        "  for id in train_df['Unit_id'].unique():\n",
        "    # If it is already dead, then maybe you shouldn't be part of the computation OR add zeros below it\n",
        "    temp_df = train_df[train_df['Unit_id']==id]\n",
        "    if temp_df.shape[0] < lent:\n",
        "      diff_arr = np.zeros((lent- temp_df.shape[0],D))\n",
        "      cur_train = temp_df[name][:lent]\n",
        "      cur_train = np.concatenate((cur_train, diff_arr), axis=0)\n",
        "      res_mat[id-1,:,:] = cur_train\n",
        "    else:\n",
        "      temp_df = train_df[train_df['Unit_id']==id]\n",
        "      cur_train = temp_df[name][:lent]\n",
        "      res_mat[id-1,:,:] = cur_train\n",
        "\n",
        "  err = ((res_mat - cur_val.to_numpy())**2).sum(axis=2)\n",
        "  residual = np.sqrt(np.mean(err, axis=1))\n",
        "\n",
        "  return residual"
      ],
      "metadata": {
        "id": "b-l78s-FSgnq"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# L is Full\n",
        "val_inst = RUL_Estimator_coll(train_df=AE_HItrain_df, test_df=AE_HIval_df, max_cycle_df=max_cycle_df, nearest=50, name=['HI_AE1','HI_AE2','HI_AE3'])\n",
        "a,b,c = result_analysis(val_inst)\n",
        "RMSE, MAE, SCORE = RUL_metrics(a,b,c)"
      ],
      "metadata": {
        "id": "AEfSleGTNB75"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(RMSE,MAE,SCORE)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "o--s6V2b4MvA",
        "outputId": "b12ca50a-e17c-4099-fdce-f78d815f460f"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[20.615528128088304, 16.32482771731451, 0.7071067811865476] [16.0, 12.5, 0.5] [[0.3498588075760032, 8.307022574766952], [0.22140275816016985, 4.866339056828965], [0, 0.07995899942818663]]\n"
          ]
        }
      ]
    }
  ]
}