tomonori-masui

import requests
from io import BytesIO

wpi1 = requests.get("https://www.stata-press.com/data/r12/wpi1.dta").content
data = pd.read_stata(BytesIO(wpi1))
ts_wpi = data.set_index("t").wpi

tsplot(ts_wpi)

tomonori-masui

param_grid = {"window_length": [5, 10, 15, 20, 25, 30]}
forecaster = create_forecaster()
nl_lgb_mae, nl_lgb_mape = grid_serch_forecaster(nl_train, nl_test, forecaster, param_grid)

tomonori-masui

fh = np.arange(test_len) + 1
forecast = forecaster.predict(fh=fh)
forecast_int = forecaster.predict_interval(fh=fh, coverage=coverage)['Coverage'][coverage]
nl_arima_mae, nl_arima_mape = plot_forecast(nl_train, nl_test, forecast, forecast_int)

tomonori-masui

test_len = int(len(ts_nl) * 0.3)
nl_train, nl_test = ts_nl.iloc[:-test_len], ts_nl.iloc[-test_len:]

forecaster = AutoARIMA(suppress_warnings=True)
forecaster.fit(nl_train)
forecaster.summary()

tomonori-masui

ts_nl_diff = (ts_nl - ts_nl.shift(1)).dropna()
tsplot(ts_nl_diff)

tomonori-masui

ts_nl = sm.datasets.get_rdataset("Nile").data
ts_nl = ts_nl.set_index('time').value

tsplot(ts_nl)

tomonori-masui

from sktime.forecasting.compose import make_reduction, TransformedTargetForecaster
from sktime.forecasting.model_selection import ExpandingWindowSplitter, ForecastingGridSearchCV
from sktime.performance_metrics.forecasting import MeanAbsolutePercentageError
import lightgbm as lgb

def create_forecaster():

    # creating forecaster with LightGBM
    regressor = lgb.LGBMRegressor()
    forecaster = make_reduction(regressor, window_length=5, strategy="recursive")

tomonori-masui

fh = np.arange(test_len) + 1
forecast, forecast_int = forecaster.predict(fh=fh, return_pred_int=True, alpha=0.05)
sun_arima_mae, sun_arima_mape = plot_forecast(sun_train, sun_test, forecast, forecast_int)

tomonori-masui

from sklearn.metrics import mean_absolute_error
from sklearn.metrics import mean_absolute_percentage_error


def plot_forecast(series_train, series_test, forecast, forecast_int=None):

    mae = mean_absolute_error(series_test, forecast)
    mape = mean_absolute_percentage_error(series_test, forecast)

    plt.figure(figsize=(12, 6))

tomonori-masui

from sktime.forecasting.arima import AutoARIMA

forecaster = AutoARIMA(start_p=8, max_p=9, suppress_warnings=True)
sun_train.index = sun_train.index.astype(int)
forecaster.fit(sun_train)
forecaster.summary()