Predicting Probability

[econeer]

Probably fastFM is more for the classification but do we have a way to predict probability using fastFM? Although fastFM has predict_proba() function, it seems like it's not an actual probability (more like a score it seems). For example, I had data (over 100,000) which has 30% True events and 70% False events. Therefore, I expected average output of predict_prob() to be 30% (which I get from Logistic Regression) but fastFM generated average probability of 43%. Are there parameters I should change?

[ibayer]

The Logistic Regression is calibrated by default which is not the case for many model. fastFM uses probit regression (with als solver) for classification. However, the probabilities can be calibrated see http://scikit-learn.org/stable/modules/calibration.html for more details.

[econeer]

I'll check it out. Thanks!