The utilities are arffmerge, arffselect, arffsplit, and
arffmetrics. You can find brief examples in the module docstrings
and more examples in the tests.
Read in a main csv (could be an arff) and an additional csv (could be an arff), insert columns from the additional csv into the main csv (joining on a key column), and write an output csv (or arff). For example usage please see the arffmerge docstring.
Select columns from a csv (could be an arff), either by column-position or (for an arff) by column-name. For example usage please see the arffselect docstring.
Read in a main csv (could be an arff) and N possibly-overlapping sets of keys, and split the rows of the csv into N csvs (or arffs) depending on which set(s) each row's key is in. For example usage please see the arffsplit docstring.
Generate statistics on the numerical features either for a single arff or comparing two arffs. The main use of the latter is to ensure that features in, say, a training arff and a test arff have similar distributions. We use the Kolmogoroff-Smirnoff statistic to sort features by mismatch: the more different the training set data and the test set data are for a feature, the higher it ranks. For a bit of detail please see the arffmetrics docstring.