I wonder if anyone has managed to fix what is questioned here.
My experiment works as follows: - Train a LogisticRegression with a dataset of a feature vector of 2 positions (1 feature of 2 categories) (each row can only have a feature vector with just one position different from 0). As a result, the model has 2 coefficients and an intercept. - Build a new LogisticRegression passing model1 as the starting point (overriding setInitialModel method by making it public) - Train the new LogisticRegression with a training dataset where only one category appears.
As a result, I would expect two coefficients, one equivalent to its value in model1 and the second one with the value modified because of the second training dataset. However, I obtain a coefficients vector with only one coefficient different from 0, which is the one referring to the category appearing in the second training dataset.
Any ideas? What am I missing to override?
