Informer のエンコーダとデコーダへの入力の期間を確認するだけ

20210213_Informer_Inputs.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Informer のエンコーダとデコーダへの入力の期間を確認するだけ\n",
    "\n",
    "このノートを動作させるには手元に以下の2つのリポジトリをcloneし、Informer2020の直下で作業する。  \n",
    "https://github.com/zhouhaoyi/Informer2020  \n",
    "https://github.com/zhouhaoyi/ETDataset  \n",
    "  \n",
    "デフォルトの設定では、ETDataset の1時間毎のデータ（ETDataset/ETT-small/ETTh1.csv）を、\n",
    "- 現時点までの96ステップ（seq_len）を使って未来の24ステップ（pred_len）を予測する。\n",
    "- ただし、デコーダには現時点までの48ステップ（label_len）をフィードする。\n",
    "- つまり、エンコーダへの入力の長さは96、デコーダへの入力の長さは72になる。\n",
    "    - デコーダへの入力は、予測すべき箇所はゼロで塗りつぶしておく。\n",
    "\n",
    "|  date  |  x = encoder input  |  y = decoder target  |  decoder input  | seq_len = 96 | label_len = 48 | pred_len = 24 |\n",
    "| ---- | ---- | ---- | ---- | ---- | ---- | ---- |\n",
    "|  2016-07-01 00:00:00  |  X  |  -  |  -  | ＊ |  -  |  -  |\n",
    "|  2016-07-01 01:00:00  |  X  |  -  |  -  | ＊ |  -  |  -  |\n",
    "|  ***  |  X  |  -  |  -  | ＊ |  -  |  -  |\n",
    "|  2016-07-02 23:00:00  |  X  |  -  |  -  | ＊ |  -  |  -  |\n",
    "|  2016-07-03 00:00:00  |  X  |  X  |  X  | ＊ |  ＊  |  -  |\n",
    "|  ***  |  X  |  X  |  X  | ＊ |  ＊  |  -  |\n",
    "|  2016-07-04 23:00:00  |  X  |  X  |  X  | ＊ |  ＊  |  -  |\n",
    "|  2016-07-05 00:00:00  |  -  |  X  |  O  |  -  |  -  |  ＊  |\n",
    "|  ***  |  -  |  X  |  O  |  -  |  -  |  ＊  |\n",
    "|  2016-07-05 23:00:00  |  -  |  X  |  O  |  -  |  -  |  ＊  |"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 元データを確認する"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from IPython.core.display import display, HTML\n",
    "import pandas as pd\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "df_raw = pd.read_csv('../ETDataset/ETT-small/ETTh1.csv')\n",
    "display(HTML('<h4>生データ</h4>'))\n",
    "display(HTML(df_raw.head().to_html(index=True)))\n",
    "\n",
    "scaler = StandardScaler()\n",
    "df_data = df_raw.iloc[:,1:]\n",
    "data = scaler.fit_transform(df_data.values)\n",
    "df_raw.iloc[:,1:] = data\n",
    "display(HTML('<h4>スケール後（学習時はスケールされるのでスケール後と見比べる）</h4>'))\n",
    "display(HTML(df_raw.head().to_html(index=True)))\n",
    "display(HTML(df_raw.iloc[46:,:].head().to_html(index=True)))\n",
    "display(HTML(df_raw.iloc[94:,:].head().to_html(index=True)))\n",
    "display(HTML(df_raw.iloc[118:,:].head().to_html(index=True)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1バッチ目の1つ目のデータを確認し元データと見比べる\n",
    "\n",
    "batch_x, batch_y が上のデータと合致する。  \n",
    "dec_inp は予測すべき箇所がゼロで塗りつぶされている。  \n",
    "batch_x_mark, batch_y_mark は「月, 日, 曜日, 時」のタイムスタンプであり、データと同じ次元に埋め込まれデータに足される。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from data.data_loader import Dataset_ETT_hour\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "seq_len = 24 * 4\n",
    "label_len = 24 * 2\n",
    "pred_len = 24\n",
    "\n",
    "data_set = Dataset_ETT_hour(\n",
    "    root_path='../ETDataset/ETT-small/',\n",
    "    data_path='ETTh1.csv',\n",
    "    flag='train', \n",
    "    size=[seq_len, label_len, pred_len],\n",
    "    features='M',  # M の場合は HUFL, HULL, MUFL, MULL, LUFL, LULL を利用, S の場合は OT のみ利用\n",
    "    scale=True,\n",
    ")\n",
    "data_loader = DataLoader(\n",
    "    data_set,\n",
    "    batch_size=32,\n",
    "    shuffle=False,\n",
    "    num_workers=0,\n",
    "    drop_last=True\n",
    ")\n",
    "\n",
    "for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(data_loader):\n",
    "    print('----- batch_x -----')\n",
    "    print(batch_x.size())\n",
    "    print(batch_x[0,:3,:])\n",
    "    print(batch_x[0,-3:,:])\n",
    "    \n",
    "    print('\\n----- batch_y -----')\n",
    "    print(batch_y.size())\n",
    "    print(batch_y[0,:3,:])\n",
    "    print(batch_y[0,-3:,:])\n",
    "    \n",
    "    print('\\n----- batch_x_mark -----')\n",
    "    print(batch_x_mark.size())\n",
    "    print(batch_x_mark[0,:3,:])\n",
    "    print(batch_x_mark[0,-3:,:])\n",
    "    \n",
    "    print('\\n----- batch_y_mark -----')\n",
    "    print(batch_y_mark.size())\n",
    "    print(batch_y_mark[0,:3,:])\n",
    "    print(batch_y_mark[0,-3:,:])\n",
    "\n",
    "    # decoder input\n",
    "    dec_inp = torch.zeros_like(batch_y[:,-pred_len:,:]).double()\n",
    "    dec_inp = torch.cat([batch_y[:,:label_len,:], dec_inp], dim=1)\n",
    "    \n",
    "    print('\\n----- dec_inp -----')\n",
    "    print(dec_inp.size())\n",
    "    print(dec_inp[0,:3,:])\n",
    "    print(dec_inp[0,-26:-22,:])\n",
    "    print(dec_inp[0,-3:,:])\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}