Closed h0bb3 closed 3 years ago
In addition, it is interesting to see how far you can get using simple naming.
Some implementations and ideas here: https://github.com/tobias-dv-lnu/s4rdm3x/tree/NaiveNameMatcher
I will probably not pursue this further at this momemt.
Well as I had a new idea I did pursue it further. Modules can now be equipped with keywords and these keywords will be used when performing the file-based mapping. Basically, it uses the path/package names and the file name to create a mapping using a special naive Bayes mapper. This can be tweaked to perform really well in most systems, and in simpler mappings, it will perform perfectly.
One idea is to suggest keywords to a user who will then map the keywords to the correct module. I.e. keyword X describes module Y. The keywords can be scanned from the actual data, and with some smart suggestions, it could become really efficient.
This has now been merged into dev: #ce3b8a354a074f74e44b5e06ec0bebe5367ae899
Using the hierarchical names of packages and files seems to work quite well for java projects. This could be utilized as both a simple naming classifier and as a boot-strapper for more complex classifiers to construct an initial set.