Support for business week

[ngupta23]

Since many business activities happen only on weekdays, missing weekends can be an issue for time series models. Add support for data that is available only on weekdays.

Indirectly related: #1853

[ngupta23]

https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html

[ngupta23]

Double check (from Konrad's 3rd presentation)

Prophet supports irregularly spaced time series.
Exponential Smoothing and ARIMA do not.

[ngupta23]

This notebook shows a workaround for now (i.e. pass seasonal period manually): https://gist.github.com/ngupta23/420fef55fb6a6c92c6035432660f6552

But this (along with all other frequency types per this link) should be easy to handle.

[ngupta23]

To test,

import numpy as np
import pandas as pd
from pycaret.time_series import TSForecastingExperiment
import datetime


N = 106
data = pd.DataFrame(
    range(N) + np.random.randn(N),
    columns=["Value"],
    index=pd.date_range(start="2022-01-01", end="2022-05-30", freq="C").tolist(),
)
exp = TSForecastingExperiment()
exp.setup(data=data, fh=6, session_id=42)
exp.plot_model()
# best_model = exp.compare_models(turbo=True)
best_model = exp.create_model("ets")
exp.predict_model(best_model)
exp.plot_model(best_model)

N = 48
data = pd.DataFrame(
    range(N) + np.random.randn(N),
    columns=["Value"],
    index=pd.date_range(start="2022-01-01", end="2030-01-01", freq="2M").tolist(),
)
exp = TSForecastingExperiment()
exp.setup(data=data, fh=6, session_id=42)
exp.plot_model()
# best_model = exp.compare_models(turbo=True)
best_model = exp.create_model("ets")
exp.predict_model(best_model)
exp.plot_model(best_model)