[Flink-3007] Implemented a parallel version of Hidden Markov Models. It uses a par…

[HilmiYildirim]

Implemented a parallel version of Hidden Markov Models. It uses a parallel version of the Baum-Welch algorithm to train the model and uses the Viterbi algorithm to predict the sequence labels.

The prediction quality of my examples is low. The reason for this is probably the small size of the model prameters

[chiwanpark]

Hi @HilmiYildirim, Thanks for opening pull request. I would like to shepherd this PR. Could you rename this PR to include JIRA issue number? I think "[FLINK-3003] Implement a parallel version of the Hidden Markov Model" would be better.

[uce]

3003 is wrong, its [FLINK-3007]

[chiwanpark]

@uce Thanks for correction. :-)

[HilmiYildirim]

I have renamed the pull request and adapted the source code so that all checks have passed.

[tillrohrmann]

Could you rename this file to HMMITSuite? Otherwise this test will always be run when Unit tests are executed.

[tillrohrmann]

Thanks for your contribution @HilmiYildirim. I had some comments.

I'm not so sure about the correctness of the implementation. Where did you get the testing data from @HilmiYildirim. Is that from some published source?

The current implementation lacks the integration with the FlinkML's pipelining mechanism. Furthermore, it only works on integers. What if the observations are characters for example.

The PR needs more documentation for better understanding the code and maintaining it.

Could we maybe add a smaller test file. One of the files is 2MB large. This should definitely be smaller.

I think this PR needs a little bit of more effort to get in shape.

[tillrohrmann]

Would be great to make the formatting consistent with the code base.

[zentol]

Closing since flink-ml is effectively frozen.