We randomly selected a different number of units (from 5 to 150) to train the logistic regression (python, scikit-learn 0.23.2) on 50% of time bins and to test it on the remaining 50%, and for each ...