Research of effectiveness new window_size estimators

[valer1435]

Due to implementation new methods for estimation window size it's important to compare they performance.

What I assume to do:

• [ ] Make an experiment for evaluating accuracy of the methods. Let take M4 dataset. We know that data have different interval parameters (from hourly to yearly). Theoretically real seasonality for data must be close to priori known values (for monthly data it should looks like 12k, for daily like 30k). We can firstly calculate time for seasonality estimation and secondly how different theoretical and empirical values for different methods. Find trade-off between cost and performance
• [ ] Run Fedot ts_forecasting task with estimated window_size parameters and compare effectiveness with default fedot (with static and dynamic structure)
• [ ] Run Fedot_ind ts_classification task with estimated window_size parameters and compare effectiveness with default fedot_ind presets