The last two weeks have been very interesting. Testing of the implemented evaluation metrics is complete and the code is now ready for the pull request.
The tests conducted on the implementations checked the accuracy and of the results they produced by comparing their results with those of their corresponding scikit-learn implementations. This was done by first generating randomized inputs which were then fed to both of the implementations. The scikit-learn implementation was accessed by embedding the python code within ECL.
Setting up scikit learn within the ECL python embedding was a slightly complex procedure, that involved installing the library by accessing the python package manager within the embedded python code.
Once set up, the data was then prepared to be sent to each of these implementations, keeping in mind ECL’s multi-workitem functionality, which needed additional code to replicate in python.
On testing, it was seen that the implemented machine learning evaluation metrics match the scikit-learn implementations to a high degree of precision, and further support the myriad interface perfectly.
The goal for the next few weeks is to initiate the pull request, and tailor the code as per the suggestions received on code review.
HPCC Systems Project Journal 