ibayer

@jwasserman2 > Are you still planning on having the 3 solvers ALS, SGD, and MCMC? Yes, BUT imo sgd is the least interesting solver and implemented more for completeness. ALS...

@jwasserman2 This makes sense. fastfm uses probit regression (same as libfm) for `als` classification which is less stable the the sigmoid transform used for the sgd classification. > If it...

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...

What bothers you? The whole to string serialization? Python side looks fine to me.

Can you open PR for the cpp stuff, easier to discuss details there.

> It is perfectly reasonable to me that the number of columns in X_test could be less than or equal to the number of columns in X_train. While it's reasonable...

@AllardJM That's currently not implemented, but excluding certain columns from the second order interactions shouldn't be to complicated to do. What's you motivation? For the ALS/CD solver the following hack...

Link doesn't work.

I see, here is the proper formatted link. [ An Efficient Alternating Newton Method for Learning Factorization Machines ](https://www.csie.ntu.edu.tw/~cjlin/papers/fm/) Thanks for the heads up the paper looks interesting. I wanted...

http://www.jmlr.org/papers/volume17/15-355/15-355.pdf See equation 1.