Maciej Kula

Go into the cloned directory and run `pip install .`: this should work.

I'm afraid the predict API here is a little wonky at the moment. Fixes are in the works but not ready to be merged yet. For now, please pass in...

Syntax problem, amended my answer above.

I include some of the predict API fixes in this branch: https://github.com/maciejkula/spotlight/pull/31 You can give that a try as well.

Yes, these are the scores corresponding to your inputs. On 2 Aug 2017 23:44, "Juan Carlos Rendon" wrote: > I'm only get this: > [ 0.02556127 0.03148273 0.02408132] > >...

These will be arranged from 0 to however many items you have.

The [`ImplicitFactorizationModel`](https://maciejkula.github.io/spotlight/factorization/implicit.html#spotlight.factorization.implicit.ImplicitFactorizationModel) is probably the right choice. The predicted scores only make sense as defining a ranking over products. You can have a look at my StackOverflow answer [here](https://stackoverflow.com/questions/41902860/interpreting-results-from-lightfm/45467232#45467232).

If you believe that the important part of the signal is likes, and dislikes, you should use the explicit model. Then, the predicted scores should be roughly between 0 and...

Explicit feedback models do not output a probability. They merely try to approximate the ratings you give them. Because in your case they are between 1 and 0, the resulting...

Machine learning models can be sensitive to your choice of hyperparameters. I would suggest playing with different hyperparameters and models until you get a better result. Have you had a...