Open ojscholten opened 4 years ago
Update: the philander 2014 replication is going well, although the small sample size limits the reliability of the scores generated by each of the data mining approaches applied.
re: neural networks, the paper uses two (one regression, one classification). I can see both finding further application in the library, so shall be revising the core structure to include a 'machine_learning' module, although at this point it seems streamlining (merging) some modules may be a better idea so that the library doesn't become too fragmented. Also, a machine learning module would need to do something other than simply wrap around scikit learn's methods, so some sort of contextual helper method or something similar specific to the transaction analytics domain would need to be added. These are notes to self but may be useful for anyone following this project!
it makes sense that the machine learning module absorbs the clustering module, and that other modules are merged to make the library simpler overall
The 9 computational methods presented in Philander’s 2014 study are yet to be integrated into the library. Each method should have its own function, if possible parameterized in the same way as described in the paper.