Release 0.1.6

Release 0.1.6 adds the following scoring methods:

• h2o_auc_score
• h2o_logloss_score

It also dramatically improves performance on several classes that rely on H2OFrame rbind operations. Since they are lazily evaluated, some tasks could cause maximum recursion depth to be reached easily. Now, rbinds are handled in batches and should not cause the same issue...

Release 0.1.5

Loads of bug fixes (some minor, some major), migration towards Python 3 compatibility, and more...

More release notes to come; see it all here

Release 0.1.3

Version 0.1.3 does not offer as many additions in functionality as 0.1.0 and 0.1.2. What it does offer is better examples, more docstrings, slight bug fixes and one large bug fix in the skutil.odr.dqrsl module.

Release 0.1.2

v0.1.2 adds:

• skutil.utils.metaestimators.if_delegate_isinstance
  • Injects methods at runtime if the delegate is an instance of a delegated type
• skutil.h2o.h2o_f_classif
  • One way ANOVA given a classification target
• skutil.h2o.H2OFScoreKBestSelector
  • Select the top K features based on the F-score,
    using the h2o_f_classif method.
• skutil.h2o.H2OFScorePercentileSelector
  • Select the top percentile of features based on the F-score,
    using the h2o_f_classif method.
• download_pojo method in H2OPipeline and H2O grid searches.
• Several bug fixes

v0.1.0

Re-release of v0.1.0 to incorporate several hotfixes.

Release 0.1.0 adds:

• Sphinx documentation
• More efficient skutil.h2o.metrics for scoring H2O functions
• H2OSelectiveScaler
• New H2OPipeline functionality
• The exclude_features keyword to H2OTransformers