utilsforecast

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)
# drop some points
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)