I have successfully implemented the triplet clustering method from the 2nd paper. I am able to follow Abinash’s comments this time. He also appears happy with my coding so far. The results are all cool 🙂
They are 99% accurate if the convergence is fast. I am checking accuracy by comparing it against the MPLP code by Sonag. The results deviate when the number of iterations increase. This happens to be the issue I have spent my time mostly on in the last few days. The conclusion is that … the deviation is not of the result of an algorithmic error but loss of precision in rounding off fractions. I think we can bear such inaccuracies. Very minute indeed.
My next expectation is to write off a fantastic ipython notebook as a part of documentation showcasing the algorithm step by step so that a normal user doesn’t have to get into cumbersome details. Another expectation is ofcourse to get the current pr( pull 420 ) merged. I know that I need to implement the 3rd paper also, but I just feel that writing the proper documentation in the form of an ipython notebook and then extending the work would be a better option.
Okay then. I must retire now. Bye bye