flennerhag / mlens / 432

92%

master: 92%

push

travis-ci

Generalization of Parallel

This commit introduces a major enhancement in parallel API structure.

  - A common BaseParallel class streamlines attribute and method
    requirements on classes interfacing ParallelProcessing.
  - The Layer class has been overhauled to generalize to an
    arbitrary number of Group instances. This commit introduce
    the push and pop interface to the Layer. For dynamic refitting.
  - Group classes binds learners, transformers and indexer and
    allow a minimal interface to learner and transformer
    construction, management and cloning.
  - Learners now directly calculate their output shape
  - Layers now directly calculate their output shape
  - Model selection: param dict now expects 'case__est' entries
    for streamlined API across package.
  - Formatting. A make_learners function allows black-box
    configuration of learners and transformers without accessing
    various formatting functions.
  - When one preprocessing case is empty, it is dropped and
    associated estimators are fitted as without an transformer.
  - Metrics: data now uses a 'layer-x/case__est---part' API
    for clearer differentiation.

919 of 1166 branches covered (78.82%)

2493 of 2847 relevant lines covered (87.57%)

2.6 hits per line