A foundation for scikit-learn at Inria

Post by hiker512 » Thu, Sep 20, 2018, 5:56 AM UTC

zeec123 wrote: post #5

Unfortunately scikit-learn is a mess without an alternative. There is so much wrong with the api design of sklearn (how can one think "predict_proba" is a good function name?). I can understand this, since most of it was probably written by PhD students without the time and expertise to come up with a proper api; many of them without a CS background. Compare this to e.g. the API of google/guava. For example https://w…

The fact that LogisticRegression uses L2 is stated _very_ clearly in the documentation. Maybe someone who is not on mobile wants to check since when?

I'm not sure if the backends in use actually allow for a non-regularized use. I would assume so, but does someone know?

Post by screye » Thu, Sep 20, 2018, 1:26 PM UTC

rjdagost wrote: post #15

zeec123 wrote: post #11

Earlier quoted context omitted.

Ease of use might be the criteria when you are a student. However as soon as you start to depend on it for a living you realise that scikit-learn made enough serious mistakes such to have lost my trust in it and I am forced to pay the ~$15.000 for matlab until some alternative is available.

Often I will do some prototyping with scikit-learn and then write my own implementation in numpy / scipy for something that goes into production. But I have used scikit-learn in production as well without issue. I have used MATLAB a bit and it is quite nice for figuring things out / prototyping. But the issue I have with it is that it's not typically intended for production software. So then you often need to reimple…

Matlab does have code generation though. Dunno if the library you use supports it, but most code can be exported to C++ or Cuda code.

Re: A foundation for scikit-learn at Inria

Re: A foundation for scikit-learn at Inria