[0.2.dev4] Utilities for merging tables

[smmaurer]

This PR adds utilities to perform merges using implicit join keys (column names that match index names) instead of requiring Orca broadcasts. See issues #78 and #100.

New utilities

• utils.validate_table()
• utils.validate_all_tables()
• utils.merge_tables()

The validation tools check that tables conform to the new stricter spec: unique index or multi-index values, no duplicate index/column names within a table. Also prints some stats to indicate whether columns that match index names of other tables make sense as join keys.

The merge tool replaces orca.merge_tables() and is simpler and more deterministic. It also supports merging on multi-indexes (useful for interaction terms and sampling weights), which orca doesn't.

Other changes

• updates utils.get_data() to use the new merge tool instead of Orca's
• updates older model step templates (BinaryLogitStep and OLSRegressionStep) to use utils.get_data(), removing any reliance on broadcasts
• raises the pandas requirement to 0.23

(The Pandas API is moving toward allowing users to refer to index and non-index columns interchangeably by name, which is nice: it's easier, and aligns better with other data formats. Version 0.23 allows specification of merge keys by name whether or not they're indexes.)

Versioning

0.2.dev4

To do before merging

• finish and test utils.merge_tables()
• add support for tables from Orca
• add support for merges not strictly in order
• update utils.get_data() to use new merge tool
• make sure all the model step templates are using the new utilities, and not relying on broadcasts
• raise pandas requirement to 0.23
• finalize versioning
• update docs and changelog

[coveralls]

Coverage increased (+0.4%) to 91.894% when pulling 3f8bcb5 on data-utils into b23a284 on master.