Towards better xvalidation

[karlnapf]

• Introduce iterator interface for pairs of train/validation indices in CSplittingStrategy

• Remove CFeatures training data from CMachine state

• Remove CLabels training data from CMachine state

• While the above two are being worked on, we could have a CMachine::clone_without_data that allows to clone a machine without cloning the data as well, to be used in xvalidation code.

• Introduce CMachine::train(CFeatures*, CLabels*) (user API)

• If possible, introduce CMachine::train(const CFeatures*, const CLabels*), via propagating constness in downstream calls. The above user API call would be a wrapper that just calls this method eventually.

• Introduce a view for features, in the lines of `CFeatures* CFeatures::subset(SGVector,index_t>), that returns a new instance with an active subset, but the same underlying data (no copy)

• Same view for CLabels

• Rewrite x-validation

• use iterator interface for subsets
• clone machine (not data) in every fold
• create new views for training/validation features/labels in every fold
• use const methods
• enable multithreading

[karlnapf]

@vinx13 this is for you. Note: this is for later, no need to start doing this now. Just wrote it down since I just had a look at the code

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[gf712]

@karlnapf do you know how much of this is still relevant?

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[karlnapf]

I think quite some stuff has been done, so closing.