Backtest Scheme Generation with Rolling Window and multiple observations

[edwinnglabs]

Describe the bug It seems to me the rolling window length is compressed automatically (at least the graph shows something like that).

To Reproduce

min_train_len=100
# just some step to make prediction spot more spread
incremental_len=50
forecast_len=84

ex_splitter = TimeSeriesSplitter(df=analysis_data,
                                 min_train_len=min_train_len,
                                 incremental_len=incremental_len,
                                 forecast_len=forecast_len,
                                 n_splits=n_splits,
                                 window_type='rolling',
                                 date_col='order_day')

_ = ex_splitter.plot()

Expected behavior If the rolling window size + forecast_len + (n_splits - 1) * incremental length > total sample size, generate an error message instead.

Screenshots

Environment (please complete the following information):

• OS: Ubuntu
• Python Version: 3.8