Fix reporting of our MRR measures

[pasky]

Right now, we report several MRR measures but it seems the important ones are not well presented:

(i) The Mb model file contains MRR measure that is irrelevant to the Mb model but is MRR of the Mb+clues logregression, is that right? Instead, please take another look at our answer scoring model files. They report cross-validation results, performance on the whole train set, and parameters used when training the model (like the mythical learning rate parameter in our case). Can we get that, please?

(ii) I added print('Mb MRR on train:', mrr(M,b,trainlist)). Can you confirm if this make sense, is this value really Mb MRR on whole train?