Informer を動かすだけ

20210214_Informer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Informer を動かすだけ\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": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== batch 0 =====\n",
      "----- batch_x -----\n",
      "torch.Size([32, 96, 7])\n",
      "----- batch_y -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- batch_x_mark -----\n",
      "torch.Size([32, 96, 4])\n",
      "----- batch_y_mark -----\n",
      "torch.Size([32, 72, 4])\n",
      "----- dec_inp -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- outputs -----\n",
      "torch.Size([32, 24, 7])\n",
      "----- loss -----\n",
      "tensor(2.6383, dtype=torch.float64, grad_fn=<MseLossBackward>)\n",
      "===== batch 1 =====\n",
      "----- batch_x -----\n",
      "torch.Size([32, 96, 7])\n",
      "----- batch_y -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- batch_x_mark -----\n",
      "torch.Size([32, 96, 4])\n",
      "----- batch_y_mark -----\n",
      "torch.Size([32, 72, 4])\n",
      "----- dec_inp -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- outputs -----\n",
      "torch.Size([32, 24, 7])\n",
      "----- loss -----\n",
      "tensor(0.3763, dtype=torch.float64, grad_fn=<MseLossBackward>)\n",
      "===== batch 2 =====\n",
      "----- batch_x -----\n",
      "torch.Size([32, 96, 7])\n",
      "----- batch_y -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- batch_x_mark -----\n",
      "torch.Size([32, 96, 4])\n",
      "----- batch_y_mark -----\n",
      "torch.Size([32, 72, 4])\n",
      "----- dec_inp -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- outputs -----\n",
      "torch.Size([32, 24, 7])\n",
      "----- loss -----\n",
      "tensor(0.4360, dtype=torch.float64, grad_fn=<MseLossBackward>)\n",
      "===== batch 3 =====\n",
      "----- batch_x -----\n",
      "torch.Size([32, 96, 7])\n",
      "----- batch_y -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- batch_x_mark -----\n",
      "torch.Size([32, 96, 4])\n",
      "----- batch_y_mark -----\n",
      "torch.Size([32, 72, 4])\n",
      "----- dec_inp -----\n",
      "torch.Size([32, 72, 7])\n",
      "----- outputs -----\n",
      "torch.Size([32, 24, 7])\n",
      "----- loss -----\n",
      "tensor(1.6757, dtype=torch.float64, grad_fn=<MseLossBackward>)\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "from torch import optim\n",
    "from data.data_loader import Dataset_ETT_hour\n",
    "from torch.utils.data import DataLoader\n",
    "from models.model import Informer\n",
    "torch.manual_seed(0)\n",
    "\n",
    "seq_len = 24 * 4  # エンコーダへの入力ステップ数\n",
    "label_len = 24 * 2  # デコーダへシードとしてフィードするステップ数\n",
    "pred_len = 24  # 予測するステップ数\n",
    "\n",
    "# エンコーダへの入力は以下の2テンソル\n",
    "#   [バッチサイズ, seq_len, 入力次元数]\n",
    "#   [バッチサイズ, seq_len, タイムスタンプ次元数]\n",
    "# デコーダへの入力は以下の2テンソル\n",
    "#   [バッチサイズ, label_len + pred_len, 入力次元数]  # [:,-pred_len:,:] はゼロ埋め\n",
    "#   [バッチサイズ, label_len + pred_len, タイムスタンプ次元数]\n",
    "\n",
    "# データ\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 も利用, 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",
    "# モデル\n",
    "model = Informer(\n",
    "    enc_in=7, dec_in=7, c_out=7,  # ETTh1 データの次元数は 7 (HUFL, HULL, MUFL, MULL, LUFL, LULL, OT)\n",
    "    seq_len=seq_len, label_len=label_len, out_len=pred_len,\n",
    "    d_ff=1024,\n",
    "    data='ETTh',  # データによって埋め込みの仕方が変わるため指定\n",
    ").double()\n",
    "# モデルの詳細を表示\n",
    "# print(model)\n",
    "# 学習対象パラメータを表示\n",
    "# for name, param in model.named_parameters():\n",
    "#     print(name.ljust(14), param.size())\n",
    "\n",
    "# 訓練\n",
    "model_optim = optim.Adam(model.parameters(), lr=0.0001)\n",
    "criterion =  nn.MSELoss()\n",
    "for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(data_loader):\n",
    "    print(f'===== batch {i} =====')\n",
    "    batch_x = batch_x.double()\n",
    "    batch_y = batch_y.double()\n",
    "    batch_x_mark = batch_x_mark.double()\n",
    "    batch_y_mark = batch_y_mark.double()\n",
    "    print('----- batch_x -----')\n",
    "    print(batch_x.size()) \n",
    "    print('----- batch_y -----')\n",
    "    print(batch_y.size())\n",
    "    print('----- batch_x_mark -----')\n",
    "    print(batch_x_mark.size())\n",
    "    print('----- batch_y_mark -----')\n",
    "    print(batch_y_mark.size())\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).double()\n",
    "    print('----- dec_inp -----')\n",
    "    print(dec_inp.size())\n",
    "    \n",
    "    outputs = model(batch_x, batch_x_mark, dec_inp, batch_y_mark)\n",
    "    print('----- outputs -----')\n",
    "    print(outputs.size())\n",
    "    \n",
    "    batch_y = batch_y[:,-pred_len:,:]\n",
    "    loss = criterion(outputs, batch_y)\n",
    "    print('----- loss -----')\n",
    "    print(loss)\n",
    "\n",
    "    loss.backward()\n",
    "    model_optim.step()\n",
    "    \n",
    "    if i == 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
}