selection

[amarchenko26]

• train it on 1/4 of the data, predict on the other 3/4
• plot the predicted in the pre-period
• do it separately for sh and non sh