Tracking issue: Split label out of example object

[jackgerrits]

As a part of the atomization project we would like to split the label object out of example and additionally change the signature of learn/predict from:

void learn_or_predict(data&, LEARNER::single_learner&, example&);
void learn_or_predict(data&, LEARNER::multi_learner&, multi_ex&);

to:

void learn(data&, LEARNER::single_learner&, example&, label&);
void learn(data&, LEARNER::multi_learner&, multi_ex&, std::vector<label>&);

void predict(data&, LEARNER::single_learner&, example&);
void predict(data&, LEARNER::multi_learner&, multi_ex&);

To do:

• [ ] Infrastructure to support learn/predict interop

  • 2323

• [ ] Convert finish example
• [ ] Convert output example
• [ ] Convert sensitivity branch
• [ ] Transform all reductions
  • CCB, Binary #2323
  • Active branch
  • Audit_regressor branch
    • This reduction does not use a label, so I implemented a forwarding label. It made it clear though, that for the other reductions that just forward they will only work one the machinery goes away.
  • Autolink - requires forwarding
  • Baseline branch
  • Bfgs branch
  • CB stack branch -> depends on sensitivity
  • NN, update, multipredict GD, FTRL, baseline branch
  • Active cover branch -> depends on sensitivity
  • Boosting branch
  • Recall tree branch
  • Replay and basic label traits branch
  • Marginal, print, sender, topk branch
  • Reductions left to do:
    • cb_adf, cb_algs, cb_explore, cbify, cbifyldf, classweight, confidence, cs_active, CSOAA::csldf, CSOAA::csoaa, ect, explore_eval, gd_mf, interact, kernel_svm, lda, log_multi, lrq, lrqfa, mf, multilabel_oaa, mwt, noop, oaa, OjaNewton, scorer, Search, shared_feature_merger, stagewise_poly, svrg, warm_cb
• [ ] Convert multipredict (See above)
• [ ] Make label ref const - may not be feasible
• [ ] Remove adaption infrastructure
• [ ] Remove label from example
• [ ] Type trait for reduction -> label type
• [ ] Constrain predict signature to predict(features,labels)

[Shivanshmundra]

@jackgerrits are you taking up this issue or is it free to take? Since I have experience with labels in the example, I would love to take this issue.

[jackgerrits]

This is an in progress task that I am working on, this issue is to track progress.