Close #37

[lgienapp]

This PR introduces changes to the way run- and qrel-files are loaded from disk/memory (#37). The goal is to (1) have forced typing and column name validation with default arguments; and (2) be able to circumvent both automatic checks if needed with custom arguments.

Flexible column naming

Tl;dr: read methods now all have a header argument that allows to specify custom column names. load methods have a mapping argument to optionally map external columns to the internal representation.

• replace qrel_header argument in TrecQrel.read_qrel with immutable default
• add run_header argument in TrecRun.read_run to mirror the TrecQrel.read_qrel signature
• add column_mapping argument to TrecRun.load_run_from_dataframe to optionally map columns to internal representation correctly

Forced Typing & Validation

Tl;dr: Columns with default names have forced types. This can be circumvented by using the header/mapping argument during data loading, as custom column names are not subject to typechecking.

• add forced string typing to query column in TrecQrel
• add forced string typing to query, q0, and docid column in TrecRun
• add validate argument to TrecRun.load_run_from_dataframe to trigger column name validation (default True)
• column validation in TrecRun.load_run_from_dataframe now raises a ValueError instead of printing, and produces the full list of missing columns

Unittests

Tl;dr: unittest adaptation

• adapt testtrecrun.py unittests to forced string type
• adapt testtreceval.py unittests to forced string type
• remove print statement from testtreceval.test_get_precision

[joaopalotti]

Looks awesome, @lgienapp! Thanks for your PR!