Krakenous is a Python "backend" for machine-learning tasks. It provides various storage options (currently, SQLite as a backend is implemented), and a lot of helper methods/functions. Designed to help with Kaggle competitions.

Because time spent on organizing data loading/storage can be spent doing more interesting things (like writing complex functions to classificate squids based on hi-res photos of their tentacle-thingies). You write the data/feature-extracting functions, Krakenous takes care of writing them to disk.

Why "Krakenous"? Krakens are awesome, and "krakenous" would make one hell of an adjective.

Store data. Extract features and store them (you'll have to write the extractor function yourself). You can call do something like:

mydataset.extract_feature_simple(some_data_feature_extractor_function, ('filename', ))

and it will extract and store the feature for the whole dataset. It comes with helper functions to build base datasets from CSV files, files in folders (for example - images in folders), convert things into numpy arrays. You can work as a team, too - Krakenous supports basic merging of different datasets (Alice extracts some of the features, Bob extracts some other features, they combine their datasets and save time).

It is modular - you can plug in your own serializers, functions to create the initial dataset (search for files in folders, parse CSV, whatever you need). Even writing your own backend is not that hard (probably).

Support more backend types, export to CSV.

v0.3 (08.01.2015) Added some additional methods, major cleanup in the DataSet class. Shelve has been deprecated as a backend, only SQLite support for now. The yield_data_records method now utilizes WHERE clauses in the backend and so should work faster. Fixed error when attempting to deserialize a non-existent record (serializer fails on None), this should help work with Null values (probably, not tested yet).

v0.2.1 (06.01.2015) Fixed imports and renamed source folder

v0.2 (05.01.2015) The first more-or-less stable version. No docstrings yet. Serialization / deserialization support, probably more-or-less stable function signatures and some working pre-rolled functions are in place.

This was a development version - with everything-breaking changes pushed daily (hourly, even). Writing a specific changelog of things that have changed between v0.1 and v0.2 would be about as much use as documenting pure primordial chaos. Sorry.