Patterns for research in machine learning

Great stuff. Are you aware of other posts like this? I would love to read more by this or other authors.

A few other patterns I've found are useful:

experiments: have a shared folder which contains numbered throwaway experiments, each one with a description. The folder contains code, data or links to data, and any scripts or make files necessary to make it self contained, self documenting, and repeatable. Experiments are listed with descriptions in a centralized place.

meta organization: above the project level, store this year's projects in a 2014/ directory; next year's projects go in 2015/. Projects that remain under active development get moved forward.

tags: tag projects with keywords to make them easy to find later. tags can include the name of programming languages used, names of idioms used, names of functions used, and any descriptive tag. You can also tag output files with the version number of tools in the tool chain, in case the version affects the output.

executable documentation: instead of documenting the steps for something like pre-processing, make a script that does all the steps. Also related, replace most comments in the code with print/echo commands, which serve the commenting purpose while also showing what is happening while the script is running.