flennerhag / mlens / 442

92%

master: 92%

push

travis-ci

Module (1/2)

First commit introducing the module API. Major changes:
- ParallelProcessing now operates a map and stack methods,
  and can be run outside of the context manager.
- BaseParallel now implements generic setup method
- Base parallel classes now implements a parameter safe API
  with re-inits for param changes
- Module module implements estimators for Layer, Learner and
  Transformer that obey Scikit-learn API.
- Use dot-notation for nested dependencies.
- Indexers no longer explicit params in Learners and Transformers
  to avoid cloning issues. Use the set_indexer method, or pass
  an indexer via the indexer kwarg
- Learner names separated from learner cache_names to avoid implicit
  modifications to name
- __fitted__ is now a generic property checking for param changes
- Indexers now communicate on __fitted__ flag
- formatting upgraded to always assign unique estimator names
- formatting also ensures estimators are cloned

Outstanding issues:
- Ensemble classes raise the "mutate on fit" error due to layers
  explicitly appearing in get_params. Possible solution is to
  build ensembles as monkey patches.

1155 of 1360 branches covered (84.93%)

2994 of 3274 relevant lines covered (91.45%)

2.72 hits per line