This PR switches our main CV method to 10-fold random CV, rather than time-based rolling window. Random V-fold seems to yield hyperparameters that perform slightly better on our test set, and does so more consistently than rolling-window.
This PR also expands the comparables search range by increasing the size of the bins (from 30 bins to 10 bins).
This PR switches our main CV method to 10-fold random CV, rather than time-based rolling window. Random V-fold seems to yield hyperparameters that perform slightly better on our test set, and does so more consistently than rolling-window.
This PR also expands the comparables search range by increasing the size of the bins (from 30 bins to 10 bins).