refactor API for black-box users

[yorkerlin]

@karlnapf
Currently, the API in GP modular is similar to GPML. The API is good for researchers but not for black-box users. For example, users have to choose inference methods/optimizers/likelihoods.

Generally, we may need unified API for most of ML models (including GP) in shogun. We can learn from sklearn. (eg, fit and predict) We can design API for

• supervised ML models
• unsupervised ML models
• semi-supervised ML models (beyond GSoC?)

EDIT: I want to hide technical details from black-box users since they do not care. On the other hand, we should offer options for advanced users to customize ML models.

[yorkerlin]

@lisitsyn

[yorkerlin]

Ref: http://scikit-learn.org/stable/modules/classes.html#module-sklearn.base

[yorkerlin]

@sanuj

[karlnapf]

Absolutely! We dont really have GSoC project related to this. Maybe @Saurabh7 finds time, it is closest to his. Otherwise, you can also add these convenience wrapper classes easily