flennerhag / mlens / 434

92%

master: 92%

push

travis-ci

Fixes for refit option, learner cloning, lyr push

Fitting with refit False caused errors for several reasons.

  - Changing the outer fit call to transform broke the sublearner
    job call. Sublearners now register the job in the constructor
    to allow the learner to dynamically change the job type if
    refit is false

  - Collect was always called, but transform does not cache
    sublearners, so the learner ended up dropping its fitted
    sublearners. A new __collect__ call is introduced in the
    _BaseEstimator class. It should be turned on only in a
    true gen_fit call, and is then turned off after collect.

Cloning with auxiliary was broken as the auxiliary was not cloned.
Solution is to explicitly introduce the auxiliary as a parameter.
The learner constructor will force-tie the indexer of the
auxiliary with its own.

Layer now allows multiple groups in push.

987 of 1174 branches covered (84.07%)

2607 of 2859 relevant lines covered (91.19%)

2.71 hits per line