nikolavp / spaska Goto Github PK

We should preferably first enable the algorithms to run in multiple threads and then look into distributed usage through RMI(maybe)

The Naive Bayes algorithms is one of the best performing algorithms for classification. It's strange that it wasn't implemented in the first iteration.

Build a module that will be able to scan packages with reflection for new algorithms and preprocessing tasks.

Some of the algorithms need refactoring.

Note: This should be done after the task about unit tests is finished so we don't break something.

Currently we only have kmeans as a clustering algorithms. So we can't compare which is better for our data. We need to implement more clustering algorithms. We can look into the mahout implemented algorithms - Here;check the Perform Clustering subsection

The first idea was to use sonar, but as we currently don't have any hosting solution and most of us are with laptops - we should ditch that idea.

We will have to use the individual tools and run them through maven on every push to the repository.
Someone should read the git hooks documentation and do some tweaking so this happens automatically on push. Also a baseline for the rules should be decided - i.e. under what % of rule compliance we should have a warning.

Currently there is no way to evaluate the clustering algorithm heuristic. There should be some method similar to the one provided by WEKA. The idea is that the clusterer should group instances the same way they are grouped by the class attribute.

Currently we only support arff files which is inconvenient for most of the systems that want to build something with ML. We should support data importing from database

Add unit tests to the whole codebase so we can start the refactoring properly.

1. Add unit tests for functions/utility objects
2. Add integration tests that verify(more like validate) that we are not lowering the precision, recall with the refactoring.

The current documentation is in a doc file outside of the code. It's pretty hard to follow the code and lookup the documentation.

Also the documentation is currently in Bulgarian - this stops international development on the project.