pip install utilsforecast
conda install -c conda-forge utilsforecast
from utilsforecast.data import generate_series
series = generate_series(3, with_trend=True, static_as_categorical=False)
series
from utilsforecast.plotting import plot_series
fig = plot_series(series, plot_random=False, max_insample_length=50, engine='matplotlib')
fig.savefig('imgs/index.png', bbox_inches='tight')
from utilsforecast.preprocessing import fill_gaps
serie = series[series['unique_id'].eq(0)].tail(10)
with_gaps = serie.sample(frac=0.5, random_state=0).sort_values('ds')
with_gaps
fill_gaps(with_gaps, freq='D')
from functools import partial
import numpy as np
from utilsforecast.evaluation import evaluate
from utilsforecast.losses import mape, mase
valid = series.groupby('unique_id').tail(7).copy()
train = series.drop(valid.index)
rng = np.random.RandomState(0)
valid['seas_naive'] = train.groupby('unique_id')['y'].tail(7).values
valid['rand_model'] = valid['y'] * rng.rand(valid['y'].shape[0])
daily_mase = partial(mase, seasonality=7)
evaluate(valid, metrics=[mape, daily_mase], train_df=train)