{"id":9965,"date":"2026-10-07T11:32:53","date_gmt":"2026-10-07T02:32:53","guid":{"rendered":"https:\/\/since2020.jp\/media\/?p=9965"},"modified":"2026-10-07T11:39:33","modified_gmt":"2026-10-07T02:39:33","slug":"timesfm-3-pandas-input-arrays","status":"publish","type":"post","link":"https:\/\/since2020.jp\/media\/timesfm-3-pandas-input-arrays\/","title":{"rendered":"\u3010\u5b9f\u88c5\u7de8\u3011TimesFM-3\u306e\u5165\u529b\u914d\u5217\u3092Pandas\u3067\u4f5c\u308b\uff5cshape\u78ba\u8a8d\u3068predict\u30fbpredict_batch\u306e\u4f7f\u3044\u5206\u3051"},"content":{"rendered":"\n

TimesFM-3\u3078\u6642\u7cfb\u5217\u30c7\u30fc\u30bf\u3092\u6e21\u3059\u306b\u306f\u3001DataFrame\u3092(\u7cfb\u5217\u6570, \u6642\u70b9\u6570)<\/code>\u306eNumPy\u914d\u5217\u3078\u5909\u63db\u3057\u307e\u3059\u3002<\/p>\n\n\n\n

\u672c\u8a18\u4e8b\u3067\u306f\u3001\u30c6\u30c8\u30a5\u30a2\u30f3\u5e02\u306e\u96fb\u529b\u6d88\u8cbb\u30c7\u30fc\u30bf\u304b\u30893\u7a2e\u985e\u306e\u5165\u529b\u914d\u5217\u3092\u4f5c\u6210\u3057\u3001predict<\/code>\u3068predict_batch<\/code>\u3067\u4e88\u6e2c\u3059\u308b\u3068\u3053\u308d\u307e\u3067\u5b9f\u88c5\u3057\u307e\u3059\u3002<\/p>\n\n\n\n

\u5165\u529b\u306e\u5206\u985e\u57fa\u6e96\u3084\u5358\u5909\u91cf\u30fb\u591a\u5909\u91cf\u4e88\u6e2c\u306e\u9055\u3044\u306f\u3001\u5165\u9580\u7de8<\/a>\u3067\u8aac\u660e\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n

\u5165\u529b<\/th>\u5fc5\u8981\u306a\u6642\u9593\u7bc4\u56f2<\/th>\u4eca\u56de\u306eshape<\/th><\/tr><\/thead>
target<\/td>context<\/td>(3, 720)<\/code><\/td><\/tr>
past-only covariate<\/td>context<\/td>(5, 720)<\/code><\/td><\/tr>
past-future covariate<\/td>context + horizon<\/td>(5, 744)<\/code><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

Python\u30fbPandas\u30fbNumPy\u306e\u57fa\u672c\u64cd\u4f5c\u304c\u3067\u304d\u3001TimesFM-3\u3078\u6e21\u3059\u5165\u529b\u914d\u5217\u3092\u518d\u73fe\u53ef\u80fd\u306a\u5f62\u3067\u4f5c\u308a\u305f\u3044\u8aad\u8005\u3092\u60f3\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n

TimesFM-3\u306e\u5165\u529b\u914d\u5217\u3092\u4f5c\u308b\u624b\u9806<\/h2>\n\n\n\n

\u3053\u3053\u304b\u3089\u306f\u3001\u30b3\u30fc\u30c9\u3092\u4e0a\u304b\u3089\u9806\u306b\u5b9f\u884c\u3067\u304d\u308b\u3088\u3046\u306b\u3001\u6b21\u306e\u9806\u756a\u3067\u5165\u529b\u30c7\u30fc\u30bf\u3092\u4f5c\u308a\u307e\u3059\u3002<\/p>\n\n\n\n

    \n
  1. CSV\u3092\u8aad\u307f\u8fbc\u307f\u3001\u65e5\u6642index\u3092\u4f5c\u308b<\/li>\n\n\n\n
  2. 10\u5206\u5024\u30921\u6642\u9593\u5024\u3078\u96c6\u8a08\u3057\u3001\u6b20\u640d\u3092\u691c\u67fb\u3059\u308b<\/li>\n\n\n\n
  3. context\u3068horizon\u306e\u65e5\u6642\u7bc4\u56f2\u3092\u6c7a\u3081\u308b<\/li>\n\n\n\n
  4. \u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u3092DataFrame\u3068\u3057\u3066\u4f5c\u308b<\/li>\n\n\n\n
  5. NumPy\u914d\u5217\u3078\u5909\u63db\u3057\u3001shape\u3092\u691c\u67fb\u3059\u308b<\/li>\n<\/ol>\n\n\n\n

    \u4eca\u56de\u306e\u8a2d\u5b9a\u306f\u30012017\u5e742\u67081\u65e50\u6642\u3092\u4e88\u6e2c\u958b\u59cb\u6642\u523b\u3068\u3057\u3001\u76f4\u524d30\u65e5\u9593\u304b\u3089\u6b21\u306e24\u6642\u9593\u3092\u4e88\u6e2c\u3059\u308b\u3082\u306e\u3067\u3059\u3002<\/p>\n\n\n\n

    context = 30\u65e5 \u00d7 24\u6642\u9593 = 720\nhorizon = 1\u65e5 \u00d7 24\u6642\u9593 = 24\ncontext + horizon = 744<\/code><\/pre><\/div>\n\n\n\n
    \u2460 CSV\u3092\u8aad\u307f\u8fbc\u307f\u3001\u65e5\u6642index\u3092\u4f5c\u308b<\/h3>\n\n\n\n
    \u307e\u305a\u3001CSV\u3092\u8aad\u307f\u8fbc\u307f\u3001\u65e5\u6642\u3068\u4f7f\u7528\u5217\u3092\u78ba\u5b9a\u3057\u307e\u3059\u3002
    \u203bZone 2\u3068Zone 3\u306e\u5143\u5217\u540d\u306b\u306f\u7a7a\u767d\u304c2\u3064\u542b\u307e\u308c\u308b\u305f\u3081\u3001\u5b9f\u969b\u306eCSV\u3068\u540c\u3058\u6587\u5b57\u5217\u3092\u6307\u5b9a\u3057\u307e\u3059\u3002(6\u884c\u76ee\u306e\u30ab\u30e9\u30e0\u540d)<\/p>\n\n\n\n
    UCI Machine Learning Repository\u3067\u516c\u958b\u3055\u308c\u3066\u3044\u308b\u300cPower Consumption of Tetouan City<\/a>\u300d\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u4f7f\u7528\u3057\u307e\u3059\u3002CSV\u30d5\u30a1\u30a4\u30eb\u3092\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u3001data\/Tetuan City power consumption.csv<\/code>\u3068\u3057\u3066\u914d\u7f6e\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u5225\u306e\u5834\u6240\u306b\u4fdd\u5b58\u3057\u305f\u5834\u5408\u306f\u3001DATA_PATH<\/code>\u3092\u5909\u66f4\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n
    import numpy as np\nimport pandas as pd\n\nDATA_PATH = "data\/Tetuan City power consumption.csv"\n\nZONE_COLUMNS = [\n    "Zone 1 Power Consumption",\n    "Zone 2  Power Consumption",\n    "Zone 3  Power Consumption",\n]\nWEATHER_COLUMNS = [\n    "Temperature",\n    "Humidity",\n    "Wind Speed",\n    "general diffuse flows",\n    "diffuse flows",\n]\nVALUE_COLUMNS = ZONE_COLUMNS + WEATHER_COLUMNS\n\nraw = pd.read_csv(DATA_PATH)\nraw["date"] = pd.to_datetime(\n    raw["DateTime"],\n    format="%m\/%d\/%Y %H:%M",\n    errors="raise",\n)\n\nmissing_columns = set(VALUE_COLUMNS) - set(raw.columns)\nif missing_columns:\n    raise ValueError(f"\u5fc5\u8981\u306a\u5217\u304c\u3042\u308a\u307e\u305b\u3093: {sorted(missing_columns)}")\n\nindexed = (\n    raw.set_index("date")[VALUE_COLUMNS]\n    .apply(pd.to_numeric, errors="raise")\n    .sort_index()\n)\n\nif not indexed.index.is_unique:\n    raise ValueError("\u65e5\u6642\u306b\u91cd\u8907\u304c\u3042\u308a\u307e\u3059")<\/code><\/pre><\/div>\n\n\n\n
    \u6642\u7cfb\u5217\u306e\u9593\u9694\u3068\u6b20\u640d\u3092\u78ba\u8a8d\u3067\u304d\u308b\u3088\u3046\u306b\u3001\u65e5\u6642index\u3092\u6301\u3064DataFrame\u306e\u307e\u307e\u524d\u51e6\u7406\u3092\u9032\u3081\u307e\u3059\u3002<\/p>\n\n\n\n
    \u2461 10\u5206\u5024\u30921\u6642\u9593\u5024\u3078\u96c6\u8a08\u3057\u3001\u6b20\u640d\u3092\u691c\u67fb\u3059\u308b<\/h3>\n\n\n\n
    TimesFM-3\u306b\u306f\u6642\u9593\u9593\u9694\u3092\u6307\u5b9a\u3059\u308b\u5f15\u6570\u304c\u306a\u3044<\/strong>\u305f\u3081\u3001\u5229\u7528\u8005\u5074\u3067\u5165\u529b\u3092\u7b49\u9593\u9694\u306b\u305d\u308d\u3048\u307e\u3059\u3002<\/p>\n\n\n\n
    \u4eca\u56de\u306f\u30011\u6642\u9593\u3054\u3068\u306e\u4ee3\u8868\u7684\u306a\u96fb\u529b\u6d88\u8cbb\u6c34\u6e96\u3092\u6271\u3046\u305f\u3081\u5e73\u5747\u5024\u3092\u4f7f\u7528\u3057\u307e\u3059\u3002\u533a\u9593\u96fb\u529b\u91cf\u3092\u6c42\u3081\u308b\u5834\u5408\u306f\u3001\u5143\u30c7\u30fc\u30bf\u306e\u5358\u4f4d\u3092\u78ba\u8a8d\u3057\u305f\u3046\u3048\u3067\u96c6\u8a08\u65b9\u6cd5\u3092\u9078\u3093\u3067\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n
    \u307e\u305f\u300110\u5206\u9593\u9694\u306e\u30c7\u30fc\u30bf\u3067\u3042\u308b\u305f\u3081\u30011\u6642\u9593\u306b\u3064\u304d6\u4ef6\u3042\u308b\u3053\u3068\u3092\u78ba\u8a8d\u3057\u3066\u304b\u3089\u5e73\u5747\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    expected_index = pd.date_range(\n    start=indexed.index.min(),\n    end=indexed.index.max(),\n    freq="10min",\n)\nif not indexed.index.equals(expected_index):\n    raise ValueError("10\u5206\u9593\u9694\u3067\u9023\u7d9a\u3057\u3066\u3044\u306a\u3044\u65e5\u6642\u304c\u3042\u308a\u307e\u3059")\n\ncounts = indexed.resample("1h").count()\nif not counts.eq(6).all().all():\n    raise ValueError("1\u6642\u9593\u3042\u305f\u308a6\u89b3\u6e2c\u3067\u306a\u3044\u533a\u9593\u304c\u3042\u308a\u307e\u3059")\n\nhourly = indexed.resample("1h").mean()\n\nif hourly.isna().any().any():\n    missing = hourly.isna().sum()\n    raise ValueError(f"1\u6642\u9593\u5024\u306b\u6b20\u640d\u304c\u3042\u308a\u307e\u3059:\\n{missing[missing > 0]}")\n\nif not np.isfinite(hourly.to_numpy(dtype=float)).all():\n    raise ValueError("NaN\u307e\u305f\u306f\u7121\u9650\u5024\u304c\u542b\u307e\u308c\u3066\u3044\u307e\u3059")<\/code><\/pre><\/div>\n\n\n\n
    \u6b20\u640d\u3092\u81ea\u52d5\u88dc\u5b8c\u3059\u308b\u3068\u51e6\u7406\u5185\u5bb9\u3092\u8ffd\u8de1\u3057\u306b\u304f\u304f\u306a\u308b\u305f\u3081\u3001\u4eca\u56de\u306fNaN\u3084\u7121\u9650\u5024\u304c\u898b\u3064\u304b\u3063\u305f\u6642\u70b9\u3067\u51e6\u7406\u3092\u505c\u6b62\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u2462 context\u3068horizon\u306e\u65e5\u6642\u7bc4\u56f2\u3092\u6c7a\u3081\u308b<\/h3>\n\n\n\n
    1\u6642\u9593\u5024\u304c\u5b8c\u6210\u3057\u3066\u304b\u3089\u3001context\u3068horizon\u306e\u5883\u754c\u3092\u6c7a\u3081\u307e\u3059\u3002
    \u3053\u306e\u9806\u5e8f\u306b\u3059\u308b\u3053\u3068\u3067\u3001\u672a\u5b9a\u7fa9\u306ehourly<\/code>\u3092\u53c2\u7167\u3059\u308b\u554f\u984c\u3092\u9632\u3052\u307e\u3059\u3002<\/p>\n\n\n\n
    CONTEXT_LENGTH = 30 * 24\nHORIZON = 24\nforecast_start = pd.Timestamp("2017-02-01 00:00:00")\n\ncontext_dates = pd.date_range(\n    end=forecast_start - pd.Timedelta(hours=1),\n    periods=CONTEXT_LENGTH,\n    freq="1h",\n)\nforecast_dates = pd.date_range(\n    start=forecast_start,\n    periods=HORIZON,\n    freq="1h",\n)\n\nmissing_context = context_dates.difference(hourly.index)\nif not missing_context.empty:\n    raise ValueError(f"context\u306b\u5b58\u5728\u3057\u306a\u3044\u65e5\u6642\u304c\u3042\u308a\u307e\u3059: {missing_context[:5]}")\n\ncontext_frame = hourly.loc[context_dates]<\/code><\/pre><\/div>\n\n\n\n
    \u672a\u676524\u6642\u9593\u306e\u96fb\u529b\u5b9f\u7e3e\u306f\u3001\u5165\u529b\u914d\u5217\u306b\u306f\u542b\u3081\u307e\u305b\u3093\u3002
    forecast_dates<\/code>\u306f\u3001\u672a\u6765\u307e\u3067\u65e2\u77e5\u306e\u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u3092\u4f5c\u308b\u305f\u3081\u306b\u4f7f\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
    \u2463 \u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u3092DataFrame\u3068\u3057\u3066\u4f5c\u308b<\/h3>\n\n\n\n
    \u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u3082DataFrame\u3068\u3057\u3066\u5217\u3092\u8ffd\u52a0\u3057\u3001\u6700\u5f8c\u306b\u307e\u3068\u3081\u3066NumPy\u914d\u5217\u3078\u5909\u63db\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u3053\u306e\u5f62\u306a\u3089\u3001\u7279\u5fb4\u540d\u3001index\u3001\u5024\u3092\u5909\u63db\u76f4\u524d\u307e\u3067\u78ba\u8a8d\u3067\u304d\u3001\u5f8c\u6bb5\u306e\u30d0\u30c3\u30af\u30c6\u30b9\u30c8\u3067\u3082\u540c\u3058\u95a2\u6570\u3092\u518d\u5229\u7528\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
    CALENDAR_COLUMNS = [\n    "hour_sin",\n    "hour_cos",\n    "dow_sin",\n    "dow_cos",\n    "is_weekend",\n]\n\ndef make_calendar_frame(index: pd.DatetimeIndex) -> pd.DataFrame:\n    hour = index.hour.to_numpy()\n    day_of_week = index.dayofweek.to_numpy()\n\n    values = np.column_stack(\n        [\n            np.sin(2 * np.pi * hour \/ 24),\n            np.cos(2 * np.pi * hour \/ 24),\n            np.sin(2 * np.pi * day_of_week \/ 7),\n            np.cos(2 * np.pi * day_of_week \/ 7),\n            (day_of_week >= 5).astype(float),\n        ]\n    )\n\n    return pd.DataFrame(\n        values,\n        index=index,\n        columns=CALENDAR_COLUMNS,\n        dtype=np.float32,\n    )\n\nall_dates = context_dates.append(forecast_dates)\ncalendar = make_calendar_frame(all_dates)<\/code><\/pre><\/div>\n\n\n\n
    \u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u306f\u672a\u6765\u306e\u65e5\u6642\u304b\u3089\u4e8b\u524d\u306b\u8a08\u7b97\u3067\u304d\u308b\u305f\u3081\u3001context 720\u6642\u9593\u3068horizon 24\u6642\u9593\u3092\u5408\u308f\u305b\u305f744\u6642\u70b9\u3092\u6301\u3061\u307e\u3059\u3002<\/p>\n\n\n\n
    \u672a\u6765\u60c5\u5831\u306e\u6f0f\u6d29\u306b\u6ce8\u610f<\/strong>
    \u672a\u6765\u5074\u3078\u5165\u529b\u3059\u308b\u306e\u306f\u3001\u4e88\u6e2c\u6642\u70b9\u3067\u78ba\u5b9a\u3057\u3066\u3044\u308b\u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u3060\u3051\u3067\u3059\u3002\u672a\u6765\u306e\u96fb\u529b\u5b9f\u7e3e\u3084\u3001\u5f8c\u304b\u3089\u89b3\u6e2c\u3055\u308c\u305f\u6c17\u8c61\u5b9f\u7e3e\u306f\u5165\u529b\u3057\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u2464 NumPy\u914d\u5217\u3078\u5909\u63db\u3057\u3001shape\u3092\u691c\u67fb\u3059\u308b<\/h3>\n\n\n\n
    \u524d\u51e6\u7406\u3068\u7279\u5fb4\u91cf\u4f5c\u6210\u304c\u7d42\u308f\u3063\u305f\u6bb5\u968e\u3067\u3001DataFrame\u3092NumPy\u914d\u5217\u3078\u5909\u63db\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    DataFrame\u306f\u300c\u884c\uff1d\u6642\u523b\u3001\u5217\uff1d\u7cfb\u5217\u300d\u3067\u3059\u304c\u3001TimesFM-3\u3078\u6e21\u30592\u6b21\u5143\u914d\u5217\u306f\u300c\u884c\uff1d\u7cfb\u5217\u3001\u5217\uff1d\u6642\u523b\u300d\u3067\u3059\u3002<\/p>\n\n\n\n
    \u305d\u306e\u305f\u3081\u3001\u6700\u5f8c\u306b.T<\/code>\u3067\u8ee2\u7f6e\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    targets = (\n    context_frame[ZONE_COLUMNS]\n    .to_numpy(dtype=np.float32)\n    .T\n)\npast_weather = (\n    context_frame[WEATHER_COLUMNS]\n    .to_numpy(dtype=np.float32)\n    .T\n)\nknown_calendar = (\n    calendar[CALENDAR_COLUMNS]\n    .to_numpy(dtype=np.float32)\n    .T\n)\n\nexpected_shapes = {\n    "targets": (3, 720),\n    "past_weather": (5, 720),\n    "known_calendar": (5, 744),\n}\n\narrays = {\n    "targets": targets,\n    "past_weather": past_weather,\n    "known_calendar": known_calendar,\n}\n\nfor name, array in arrays.items():\n    assert array.shape == expected_shapes[name]\n    assert array.dtype == np.float32\n    assert np.isfinite(array).all()<\/code><\/pre><\/div>\n\n\n\n
    \u4ee5\u4e0a\u3067\u3001target\u3001\u904e\u53bb\u306e\u6c17\u8c61\u3001\u65e2\u77e5\u306e\u30ab\u30ec\u30f3\u30c0\u30fc\u7279\u5fb4\u3092TimesFM-3\u306e\u5165\u529bshape\u3078\u5909\u63db\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    TimesFM-3\u3092\u521d\u671f\u5316\u3059\u308b<\/h2>\n\n\n\n
    \u4ee5\u964d\u306eforecaster.predict<\/code>\u3092\u5b9f\u884c\u3059\u308b\u306b\u306f\u3001\u5148\u306b\u30d1\u30c3\u30b1\u30fc\u30b8\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u3001\u30e2\u30c7\u30eb\u3092\u521d\u671f\u5316\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    PyTorch\u7248\u306e\u6700\u5c0f\u4f8b\u306f\u6b21\u306e\u3068\u304a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n
    python -m pip install "timesfm[torch]==3.0.2" "torch==2.14.1" "pandas==2.2.3" "numpy==2.1.3"<\/code><\/pre><\/div>\n\n\n\n
    import platform\nfrom importlib.metadata import version\n\nfrom timesfm3 import TimesFM3Forecaster\n\nMODEL_ID = "google\/timesfm-3.0-pytorch"\nMODEL_REVISION = "43046b85ec22d584a13f8098c2ed39c889e129c2"\n\nprint("Python:", platform.python_version())\nprint("timesfm:", version("timesfm"))\nprint("torch:", version("torch"))\nprint("pandas:", version("pandas"))\nprint("numpy:", version("numpy"))\nprint("OS:", platform.platform())\n\nforecaster = TimesFM3Forecaster.from_pretrained(\n    MODEL_ID,\n    revision=MODEL_REVISION,\n    device="cpu",  # \u672c\u8a18\u4e8b\u3067\u306fMac\u4e0a\u306eCPU\u5b9f\u884c\u306b\u56fa\u5b9a\n)\n\nprint("device:", forecaster.device)\nprint("model revision:", MODEL_REVISION)<\/code><\/pre><\/div>\n\n\n\n
    \u5b9f\u884c\u7d50\u679c\u306f\u6b21\u306e\u3068\u304a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n
    Python: 3.11.16\ntimesfm: 3.0.2\ntorch: 2.14.1\npandas: 2.2.3\nnumpy: 2.1.3\nOS: macOS-26.6-arm64-arm-64bit\ndevice: cpu\nmodel revision: 43046b85ec22d584a13f8098c2ed39c889e129c2<\/code><\/pre><\/div>\n\n\n\n
    \u672c\u8a18\u4e8b\u3067\u306f\u3001Apple Silicon\u642d\u8f09Mac\u306eCPU\u3067\u4e0a\u8a18\u74b0\u5883\u3092\u78ba\u8a8d\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    \u30e2\u30c7\u30ebrevision\u3092\u56fa\u5b9a\u3057\u3066\u3044\u308b\u305f\u3081\u3001Hugging Face\u4e0a\u306e\u30e2\u30c7\u30eb\u304c\u66f4\u65b0\u3055\u308c\u305f\u5f8c\u3082\u540c\u3058\u30e2\u30c7\u30eb\u30a6\u30a7\u30a4\u30c8\u3092\u53d6\u5f97\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
    \u521d\u56de\u5b9f\u884c\u6642\u306fHugging Face\u304b\u3089\u91cd\u307f\u3092\u53d6\u5f97\u3059\u308b\u305f\u3081\u3001\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u63a5\u7d9a\u3068\u5341\u5206\u306a\u4fdd\u5b58\u5bb9\u91cf\u304c\u5fc5\u8981\u3067\u3059\u3002<\/p>\n\n\n\n
    \u5229\u7528\u6761\u4ef6\u306e\u6ce8\u610f<\/strong>
    2026\u5e7410\u6708\u6642\u70b9\u3067\u3001TimesFM-3\u306e\u30e2\u30c7\u30eb\u30a6\u30a7\u30a4\u30c8\u306b\u306f\u975e\u5546\u7528\u30fb\u975e\u672c\u756a\u5229\u7528\u3092\u524d\u63d0\u3068\u3057\u305f\u30e9\u30a4\u30bb\u30f3\u30b9\u304c\u9069\u7528\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u696d\u52d9\u5229\u7528\u3084\u30b5\u30fc\u30d3\u30b9\u3078\u306e\u7d44\u307f\u8fbc\u307f\u3092\u691c\u8a0e\u3059\u308b\u5834\u5408\u306f\u3001\u5fc5\u305a\u6700\u65b0\u306e\u30e2\u30c7\u30eb\u30ab\u30fc\u30c9\u3068TimesFM-3\u516c\u5f0f\u30e9\u30a4\u30bb\u30f3\u30b9<\/a>\u3092\u78ba\u8a8d\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n
    predict\u3068predict_batch\u3092\u4f7f\u3044\u5206\u3051\u308b<\/h2>\n\n\n\n
    predict\u3068predict_batch\u306e\u5f15\u6570\u4e00\u89a7<\/h3>\n\n\n\n
    predict<\/code>\u3068predict_batch<\/code>\u306f\u3001\u307b\u307c\u540c\u3058\u5f15\u6570\u3092\u6301\u3061\u307e\u3059\u3002
    \u5927\u304d\u306a\u9055\u3044\u306f\u3001predict<\/code>\u304c1\u4ef6\u306e\u4e88\u6e2c\u554f\u984c<\/strong>\u3092\u53d7\u3051\u53d6\u308a\u3001predict_batch<\/code>\u304c\u8907\u6570\u4ef6\u3092list\u3067<\/strong>\u53d7\u3051\u53d6\u308b\u70b9\u3067\u3059\u3002<\/p>\n\n\n
    \u30b9\u30af\u30ed\u30fc\u30eb\u3067\u304d\u307e\u3059 <\/i><\/span><\/div>\n
    \u5f15\u6570<\/th>predict<\/code><\/th>predict_batch<\/code><\/th>\u5f79\u5272<\/th>\u672c\u8a18\u4e8b\u306e\u8a2d\u5b9a<\/th><\/tr><\/thead>
    context<\/code> \/ contexts<\/code><\/td>np.ndarray<\/code><\/td>list[np.ndarray]<\/code><\/td>\u4e88\u6e2c\u5bfe\u8c61\u3068\u306a\u308b\u904e\u53bb\u306etarget\u7cfb\u5217<\/td>\u4e0b\u8868\u300ccontext\u3068contexts\u306e\u5165\u529bshape\u306e\u9055\u3044\u300d\u3092\u53c2\u7167<\/td><\/tr>
    horizon<\/code><\/td>\u6574\u6570<\/td>\u6574\u6570<\/td>\u4e88\u6e2c\u3059\u308b\u6642\u70b9\u6570<\/td>24<\/code><\/td><\/tr>
    past_only_covariates<\/code><\/td>NumPy\u914d\u5217\u307e\u305f\u306fNone<\/code><\/td>NumPy\u914d\u5217\u307e\u305f\u306fNone<\/code>\u3092\u8981\u7d20\u306b\u6301\u3064list\u3001\u3082\u3057\u304f\u306fNone<\/code><\/td>\u904e\u53bb\u533a\u9593\u3060\u3051\u5229\u7528\u3067\u304d\u308b\u5171\u5909\u91cf<\/td>\u904e\u53bb720\u6642\u9593\u306e\u6c17\u8c615\u7cfb\u5217<\/td><\/tr>
    past_future_covariates<\/code><\/td>NumPy\u914d\u5217\u307e\u305f\u306fNone<\/code><\/td>NumPy\u914d\u5217\u307e\u305f\u306fNone<\/code>\u3092\u8981\u7d20\u306b\u6301\u3064list\u3001\u3082\u3057\u304f\u306fNone<\/code><\/td>\u904e\u53bb\u3068\u672a\u6765\u306e\u4e21\u65b9\u3067\u5229\u7528\u3067\u304d\u308b\u5171\u5909\u91cf<\/td>744\u6642\u9593\u306e\u30ab\u30ec\u30f3\u30c0\u30fc5\u7cfb\u5217<\/td><\/tr>
    ts_id<\/code> \/ ts_ids<\/code><\/td>str<\/code>\u307e\u305f\u306fNone<\/code><\/td>list[str]<\/code>\u307e\u305f\u306fNone<\/code><\/td>\u51fa\u529b\u3078\u4ed8\u3051\u308b\u7cfb\u5217ID<\/td>\u6307\u5b9a\u3057\u306a\u3044<\/td><\/tr>
    return_quantiles<\/code><\/td>bool<\/code><\/td>bool<\/code><\/td>\u5206\u4f4d\u70b9\u4e88\u6e2c\u3082\u8fd4\u3059\u304b<\/td>False<\/code><\/td><\/tr>
    use_symmetric_averaging<\/code><\/td>bool<\/code><\/td>bool<\/code><\/td>\u5143\u306e\u5165\u529b\u3068\u7b26\u53f7\u53cd\u8ee2\u3057\u305f\u5165\u529b\u306e\u4e88\u6e2c\u3092\u5e73\u5747\u3059\u308b\u304b<\/td>\u65e2\u5b9a\u5024\u306eFalse<\/code><\/td><\/tr>
    make_positive<\/code><\/td>bool<\/code><\/td>bool<\/code><\/td>\u975e\u8ca0\u306e\u5165\u529b\u7cfb\u5217\u306b\u5bfe\u3057\u3066\u4e88\u6e2c\u5024\u30920\u4ee5\u4e0a\u3078\u88dc\u6b63\u3059\u308b\u304b<\/td>\u65e2\u5b9a\u5024\u306eFalse<\/code><\/td><\/tr>
    sort_quantiles<\/code><\/td>bool<\/code><\/td>bool<\/code><\/td>\u5206\u4f4d\u70b9\u4e88\u6e2c\u3092\u5024\u306e\u5c0f\u3055\u3044\u9806\u306b\u4e26\u3079\u308b\u304b<\/td>\u65e2\u5b9a\u5024\u306eTrue<\/code><\/td><\/tr>
    use_znorm<\/code><\/td>bool<\/code><\/td>bool<\/code><\/td>\u63a8\u8ad6\u524d\u306b\u5404\u7cfb\u5217\u3092Z\u6b63\u898f\u5316\u3059\u308b\u304b<\/td>False<\/code><\/td><\/tr>
    padding_mode<\/code><\/td>\"none\"<\/code>\u307e\u305f\u306f\"edge\"<\/code><\/td>\"none\"<\/code>\u307e\u305f\u306f\"edge\"<\/code><\/td>\u672a\u6765\u5171\u5909\u91cf\u306e\u53f3\u5074\u3092\u672b\u5c3e\u5024\u3067\u88dc\u3046\u304b<\/td>\"none\"<\/code><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
    \u672c\u8a18\u4e8b\u3067\u306f\u5165\u529bshape\u306e\u9055\u3044\u3092\u660e\u78ba\u306b\u3059\u308b\u305f\u3081\u3001\u5206\u4f4d\u70b9\u914d\u5217\u306f\u8fd4\u3055\u305a\uff08return_quantiles=False<\/code>\uff09\u3001\u5bfe\u79f0\u5e73\u5747\u3001\u975e\u8ca0\u88dc\u6b63\u3001\u8ffd\u52a0\u306eZ\u6b63\u898f\u5316\u3082\u4f7f\u7528\u3057\u307e\u305b\u3093\u3002sort_quantiles<\/code>\u306f\u65e2\u5b9a\u5024\u306eTrue<\/code>\u3068\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    context<\/code>\u3068contexts<\/code>\u306e\u5165\u529bshape\u306e\u9055\u3044<\/h3>\n\n\n\n
    predict_batch<\/code>\u3067\u306f\u3001\u5916\u5074\u306elist\u306e\u5404\u8981\u7d20\u304c1\u4ef6\u306e\u4e88\u6e2c\u554f\u984c\u3067\u3059\u3002\u4e00\u65b9\u3001\u5404\u914d\u5217\u306e\u5148\u982d\u8ef8\u306f\u3001\u305d\u306e1\u4ef6\u306b\u542b\u307e\u308c\u308btarget\u7cfb\u5217\u3092\u8868\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u547c\u3073\u51fa\u3057\u65b9<\/th>\u672c\u8a18\u4e8b\u3067\u6e21\u3059\u5165\u529b<\/th>\u610f\u5473<\/th><\/tr><\/thead>
    \u591a\u5909\u91cf
    predict<\/code><\/td>    		targets<\/code>\uff1a(3, 720)<\/code><\/td>3\u7cfb\u5217\u3092\u542b\u30801\u4ef6\u3068\u3057\u3066
    \u5171\u540c\u4e88\u6e2c<\/td><\/tr>
    \u5358\u5909\u91cfpredict_batch<\/code><\/td>[targets[0], targets[1], targets[2]]<\/span><\/span><\/code>
    \uff1a\u5404\u8981\u7d20\u306f(720,)<\/code><\/span><\/td>    		1\u7cfb\u5217\u305a\u3064\u306e3\u4ef6\u3068\u3057\u3066
    \u72ec\u7acb\u4e88\u6e2c<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
    \u5358\u5909\u91cfpredict_batch\u3067\u306f\u5171\u5909\u91cf\u3082\u30d0\u30c3\u30c1\u4ef6\u6570\u5206\u305d\u308d\u3048\u308b<\/h3>\n\n\n\n
    3\u5730\u57df\u3092\u72ec\u7acb\u3057\u305f\u5358\u5909\u91cf\u554f\u984c\u3068\u3057\u3066\u51e6\u7406\u3059\u308b\u5834\u5408\u3001target\u3060\u3051\u3067\u306a\u304f\u5171\u5909\u91cf\u30823\u8981\u7d20\u306elist\u306b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u5404\u8981\u7d20\u306eshape\u3068\u9806\u5e8f\u3092\u5bfe\u5fdc\u3055\u305b\u308b\u3053\u3068\u304c\u91cd\u8981\u3067\u3059\u3002<\/p>\n\n\n\n
    \u540c\u4e00\u306e\u6c17\u8c61\u30fb\u30ab\u30ec\u30f3\u30c0\u30fc\u914d\u5217\u3092\u5168\u5730\u57df\u3067\u5171\u6709\u3059\u308b\u5834\u5408\u3001\u547c\u3073\u51fa\u3057\u5074\u3067\u6570\u5024\u30c7\u30fc\u30bf\u3092\u30c7\u30a3\u30fc\u30d7\u30b3\u30d4\u30fc\u3059\u308b\u5fc5\u8981\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u6b21\u306e\u3088\u3046\u306b\u3001\u63a8\u8ad6\u4e2d\u306b\u5909\u66f4\u3057\u306a\u3044\u540c\u3058\u914d\u5217\u3078\u306e\u53c2\u7167\u3092list\u3078\u4e26\u3079\u307e\u3059\u3002<\/p>\n\n\n\n
    # target: \u5404\u8981\u7d20\u306ftargets\u306e1\u884c\u3092\u53c2\u7167\u3059\u308b1\u6b21\u5143\u914d\u5217 (720,)\nindependent_contexts = [\n    targets[i]\n    for i in range(targets.shape[0])\n]\n\n# \u5171\u901a\u306e\u5171\u5909\u91cf\u306f\u3001\u540c\u3058\u914d\u5217\u3078\u306e\u53c2\u7167\u3092\u30d0\u30c3\u30c1\u4ef6\u6570\u5206\u4e26\u3079\u308b\nindependent_past_weather = [\n    past_weather\n    for _ in independent_contexts\n]\nindependent_known_calendar = [\n    known_calendar\n    for _ in independent_contexts\n]<\/code><\/pre><\/div>\n\n\n\n
    \u3053\u306elist\u306f\u540c\u3058\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u3092\u53c2\u7167\u3059\u308b\u305f\u3081\u3001\u63a8\u8ad6\u524d\u5f8c\u306b\u914d\u5217\u3092in-place\u66f4\u65b0\u3057\u306a\u3044\u3053\u3068\u304c\u524d\u63d0\u3067\u3059\u3002<\/p>\n\n\n\n
    \u5730\u57df\u3054\u3068\u306b\u5171\u5909\u91cf\u3092\u5909\u66f4\u3059\u308b\u5834\u5408\u3060\u3051\u3001\u305d\u306e\u8981\u7d20\u3092\u500b\u5225\u306b\u751f\u6210\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    3\u3064\u306elist\u3092\u5bfe\u5fdc\u3055\u305b\u3066predict_batch<\/code>\u3078\u6e21\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    independent_outputs = list(\n    forecaster.predict_batch(\n        contexts=independent_contexts,\n        horizon=HORIZON,\n        past_only_covariates=independent_past_weather,\n        past_future_covariates=independent_known_calendar,\n        return_quantiles=False,\n        use_znorm=False,\n        padding_mode="none",\n    )\n)\n\nassert len(independent_outputs) == 3\nassert all(out.forecast.shape == (24,) for out in independent_outputs)\n\nprint(\n    "predict_batch:",\n    [out.forecast.shape for out in independent_outputs],\n)<\/code><\/pre><\/div>\n\n\n\n
    \u5b9f\u884c\u7d50\u679c\u306f\u6b21\u306e\u3068\u304a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n
    predict_batch: [(24,), (24,), (24,)]<\/code><\/pre><\/div>\n\n\n\n
    3\u5730\u57df\u305d\u308c\u305e\u308c\u306b\u3064\u3044\u3066\u300124\u6642\u9593\u5206\u306e\u4e88\u6e2c\u304c\u8fd4\u3055\u308c\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    contexts<\/code>\u3060\u3051\u3092list\u5316\u3057\u3001\u5171\u5909\u91cf\u3092(5, 720)<\/code>\u3084(5, 744)<\/code>\u306e2\u6b21\u5143\u914d\u5217\u306e\u307e\u307e\u6e21\u3057\u3066\u306f\u3044\u3051\u307e\u305b\u3093\u3002
    \u3053\u306e\u8aa4\u308a\u306f\u3001\u5fc5\u305ashape\u30a8\u30e9\u30fc\u306b\u306a\u308b\u3068\u306f\u9650\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u516c\u958b\u5b9f\u88c5\u3067\u306f\u5171\u5909\u91cf\u306blist(...)<\/code>\u3092\u9069\u7528\u3059\u308b\u305f\u3081\u3001(5, 720)<\/code>\u306e5\u884c\u304c5\u3064\u306e\u30d0\u30c3\u30c1\u8981\u7d20\u3068\u3057\u3066\u5206\u89e3\u3055\u308c\u3001\u5148\u982d3\u884c\u3060\u3051\u304c3\u5730\u57df\u30781\u672c\u305a\u3064\u8aa4\u3063\u3066\u5272\u308a\u5f53\u3066\u3089\u308c\u308b\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    predict_batch<\/code>\u306fpast_only_covariates[idx]<\/code>\u3068past_future_covariates[idx]<\/code>\u3092\u5404context\u306b\u5bfe\u5fdc\u4ed8\u3051\u308b\u305f\u3081\u30013\u3064\u306e\u5f15\u6570\u306f\u5916\u5074\u306e\u8981\u7d20\u6570\u3068\u9806\u5e8f\u3092\u660e\u793a\u7684\u306b\u305d\u308d\u3048\u307e\u3059\u3002<\/p>\n\n\n\n
    \u591a\u5909\u91cfpredict\u3067\u306f\u5171\u5909\u91cf\u3092list\u5316\u3057\u306a\u3044<\/h3>\n\n\n\n
    3\u5730\u57df\u3092\u5171\u540c\u4e88\u6e2c\u3059\u308b\u5834\u5408\u306f\u30011\u4ef6\u306e\u591a\u5909\u91cf\u554f\u984c\u3068\u3057\u30662\u6b21\u5143\u914d\u5217\u3092\u76f4\u63a5\u6e21\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    joint_output = forecaster.predict(\n    context=targets,                         # (3, 720)\n    horizon=HORIZON,\n    past_only_covariates=past_weather,       # (5, 720)\n    past_future_covariates=known_calendar,   # (5, 744)\n    return_quantiles=False,\n    use_znorm=False,\n    padding_mode="none",\n)\n\nassert joint_output.forecast.shape == (3, 24)\n\nprint("predict:", joint_output.forecast.shape)<\/code><\/pre><\/div>\n\n\n\n
    \u5b9f\u884c\u7d50\u679c\u306f\u6b21\u306e\u3068\u304a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n
    predict: (3, 24)<\/code><\/pre><\/div>\n\n\n\n
    \u51fa\u529bshape\u306f(3, 24)<\/code>\u3068\u306a\u308a\u30013\u5730\u57df\u305d\u308c\u305e\u308c\u306b\u3064\u3044\u306624\u6642\u9593\u5206\u306e\u4e88\u6e2c\u304c\u8fd4\u3055\u308c\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    TimesFM-3\u56fa\u6709\u306e\u30d1\u30c7\u30a3\u30f3\u30b0\u3068\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0<\/h2>\n\n\n\n
    \u672c\u8a18\u4e8b\u3067\u306f\u3001\u6700\u521d\u306e\u691c\u8a3c\u6761\u4ef6\u3068\u3057\u3066padding_mode=\"none\"<\/code>\u3001use_znorm=False<\/code> \u3092\u6307\u5b9a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    context\u9577\u306e\u8abf\u6574\u306fAPI\u304c\u5185\u90e8\u3067\u884c\u3046\u305f\u3081\u3001720\u6642\u70b9\u306e\u914d\u5217\u3092\u5229\u7528\u8005\u5074\u3067\u30d1\u30c7\u30a3\u30f3\u30b0\u3059\u308b\u5fc5\u8981\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u307e\u305f\u3001\u5165\u529b\u7279\u5fb4\u3092\u610f\u56f3\u3057\u305f\u5024\u306e\u307e\u307e\u78ba\u8a8d\u3067\u304d\u308b\u3088\u3046\u306b\u3001\u8ffd\u52a0\u306eZ\u6b63\u898f\u5316\u306f\u7121\u52b9\u306b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u305d\u308c\u305e\u308c\u306e\u5185\u90e8\u52d5\u4f5c\u3092\u78ba\u8a8d\u3057\u305f\u3044\u5834\u5408\u306f\u3001\u4ee5\u4e0b\u3092\u5c55\u958b\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n
    \u3010\u8a73\u7d30\u3011TimesFM-3\u306e\u30d1\u30c7\u30a3\u30f3\u30b0\u3068padding_mode<\/summary>\n
    \u516c\u958b\u4e2d\u306egoogle\/timesfm-3.0-pytorch<\/code>\u306f\u5165\u529b\u30d1\u30c3\u30c1\u957732\u3092\u57fa\u672c\u5358\u4f4d\u3068\u3057\u307e\u3059\u304c\u3001context\u9577\u3092\u5229\u7528\u8005\u304c32\u306e\u500d\u6570\u3078\u52a0\u5de5\u3059\u308b\u5fc5\u8981\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    predict_batch<\/code>\u3078\u7570\u306a\u308bcontext\u9577\u306e\u914d\u5217\u3092\u6e21\u3059\u3068\u3001\u540c\u3058\u30d0\u30c3\u30c1\u5185\u306e\u6700\u5927\u9577\u309232\u306e\u500d\u6570\u3078\u5207\u308a\u4e0a\u3052\u3001\u77ed\u3044\u7cfb\u5217\u3092\u5de6\u5074\u304b\u3089\u30bc\u30ed\u3067\u57cb\u3081\u305f\u3046\u3048\u3067\u30de\u30b9\u30af\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u3053\u306e\u516c\u958b\u5b9f\u88c5\u306e\u6700\u5927context\u306f15,360\u6642\u70b9\u3067\u3001\u305d\u308c\u3088\u308a\u9577\u3044\u5834\u5408\u306f\u53e4\u3044\u5074\u304c\u5207\u308a\u6368\u3066\u3089\u308c\u307e\u3059\u3002<\/p>\n\n\n\n
    720\u6642\u9593\u306f32\u3067\u5272\u308a\u5207\u308c\u306a\u3044\u305f\u3081\u3001\u5185\u90e8\u3067\u306f736\u6642\u70b9\u5206\u306e\u9818\u57df\u3078\u5de6\u30d1\u30c7\u30a3\u30f3\u30b0\u3055\u308c\u307e\u3059\u304c\u3001\u5229\u7528\u8005\u304c\u30bc\u30ed\u3092\u8ffd\u52a0\u3059\u308b\u5fc5\u8981\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u306a\u304a\u3001padding_mode=\"none\" | \"edge\"<\/code>\u306f\u3001\u3053\u306econtext\u7528\u30d1\u30c7\u30a3\u30f3\u30b0\u3092\u5207\u308a\u66ff\u3048\u308b\u5f15\u6570\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u4e88\u6e2chorizon\u3092\u51fa\u529b\u30d1\u30c3\u30c1\u957764\u3078\u5185\u90e8\u3067\u5207\u308a\u4e0a\u3052\u305f\u969b\u3001past-future covariate\u306e\u672a\u6765\u5074\u3092\u672b\u5c3e\u5024\u3067\u53f3\u30d1\u30c7\u30a3\u30f3\u30b0\u3059\u308b\u304b\u3092\u6307\u5b9a\u3059\u308b\u5f15\u6570\u3067\u3059\u3002<\/p>\n\n\n\n
    \u4eca\u56de\u306ehorizon 24\u3067\u306f\u3001padding_mode=\"edge\"<\/code>\u306a\u3089\u4e0d\u8db3\u3059\u308b40\u6642\u70b9\u304c\u672b\u5c3e\u5024\u3067\u88dc\u308f\u308c\u3001\u65e2\u5b9a\u306e\"none\"<\/code>\u306a\u3089\u88dc\u308f\u308c\u307e\u305b\u3093\u3002<\/p>\n<\/details>\n\n\n\n
    \u3010\u8a73\u7d30\u3011use_znorm\u306e\u5bfe\u8c61\u3068\u624b\u52d5\u6a19\u6e96\u5316\u3068\u306e\u9055\u3044<\/summary>\n
    TimesFM-3\u306f\u30e2\u30c7\u30eb\u5185\u90e8\u3067\u3082\u7cfb\u5217\u3054\u3068\u306e\u6b63\u898f\u5316\u3092\u884c\u3046\u305f\u3081\u3001\u6c17\u6e29\u30fb\u96fb\u529b\u30fbsin\/cos\u3092\u540c\u3058\u5c3a\u5ea6\u3078\u624b\u4f5c\u696d\u3067\u305d\u308d\u3048\u308b\u3053\u3068\u306f\u3001API\u5229\u7528\u4e0a\u306e\u5fc5\u9808\u6761\u4ef6\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u3053\u308c\u3068\u306f\u5225\u306b\u3001\u516c\u958bAPI\u306b\u306f\u63a8\u8ad6\u524d\u306e\u8ffd\u52a0\u51e6\u7406\u3068\u3057\u3066use_znorm<\/code>\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    \u65e2\u5b9a\u5024\u306fFalse<\/code>\u3067\u3059\u3002<\/p>\n\n\n\n
    True<\/code>\u306b\u3059\u308b\u3068\u3001target\u3001past-only covariate\u3001past-future covariate\u306e\u5404\u884c<\/strong>\u304c\u3001\u305d\u308c\u305e\u308c\u306e\u5e73\u5747\u3068\u6a19\u6e96\u504f\u5dee\u3067Z\u6b63\u898f\u5316\u3055\u308c\u307e\u3059\u3002<\/p>\n\n\n\n
    target\u306e\u4e88\u6e2c\u3060\u3051\u306f\u3001\u63a8\u8ad6\u5f8c\u306b\u5143\u306e\u5c3a\u5ea6\u3078\u623b\u3055\u308c\u307e\u3059\u3002<\/p>\n\n\n\n
    past-future covariate\u306e\u5e73\u5747\u30fb\u6a19\u6e96\u504f\u5dee\u306f\u3001\u5165\u529b\u3068\u3057\u3066\u6e21\u3057\u305fcontext\u3068horizon\u3092\u5408\u308f\u305b\u305f\u5168\u533a\u9593\u304b\u3089\u8a08\u7b97\u3055\u308c\u307e\u3059\u3002\u672a\u6765\u5074\u3082\u4e88\u6e2c\u6642\u70b9\u3067\u65e2\u77e5\u306e\u5024\u3067\u3059\u304c\u3001fold\u3054\u3068\u306b\u9031\u672b\u306e\u6bd4\u7387\u306a\u3069\u304c\u5909\u308f\u308c\u3070\u5909\u63db\u7d50\u679c\u3082\u5909\u308f\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    \u91cd\u8981\u306a\u306e\u306f\u3001use_znorm<\/code>\u306b\u300c\u6c17\u6e29\u3060\u3051\u6b63\u898f\u5316\u3057\u3001is_weekend<\/code>\u306f\u6b63\u898f\u5316\u3057\u306a\u3044\u300d\u3068\u3044\u3063\u305f\u7279\u5fb4\u5225\u306e\u6307\u5b9a\u304c\u306a\u3044\u3053\u3068\u3067\u3059\u3002<\/p>\n\n\n\n
    0\/1\u30d5\u30e9\u30b0\u3092Z\u6b63\u898f\u5316\u3057\u3066\u30822\u5024\u6027\u305d\u306e\u3082\u306e\u304c\u6d88\u3048\u308b\u308f\u3051\u3067\u306f\u306a\u304f\u3001sin\/cos\u3082\u30a2\u30d5\u30a3\u30f3\u5909\u63db\u3060\u3051\u3067\u5468\u671f\u60c5\u5831\u304c\u76f4\u3061\u306b\u5931\u308f\u308c\u308b\u308f\u3051\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n\n\n\n
    \u305f\u3060\u3057\u3001\u5024\u306e\u4e2d\u5fc3\u30fb\u632f\u5e45\u30fb2\u3064\u306e\u72b6\u614b\u9593\u306e\u8ddd\u96e2\u306f\u5909\u308f\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    \u7279\u306b\u3001\u77ed\u3044\u7a93\u3067\u9031\u672b\u306e\u5272\u5408\u304c\u7570\u306a\u308b\u5834\u5408\u3084\u3001sin\/cos\u304c\u6574\u6570\u5468\u671f\u3092\u542b\u307e\u306a\u3044\u5834\u5408\u306f\u3001\u7a93\u3054\u3068\u306b\u5909\u63db\u7d50\u679c\u304c\u5909\u5316\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
    \u305d\u306e\u305f\u3081\u3001\u6570\u5024\u7279\u5fb4\u3068\u30d0\u30a4\u30ca\u30ea\u30fb\u5468\u671f\u7279\u5fb4\u304c\u6df7\u5728\u3059\u308b\u4eca\u56de\u306e\u69cb\u6210\u3067\u306f\u3001\u6b21\u3092\u900f\u660e\u6027\u306e\u9ad8\u3044\u521d\u671f\u8a2d\u5b9a\u3068\u3057\u307e\u3059\u3002<\/p>\n\n\n\n