Reproducing published accuracy results

[rafkt]

I am trying to reproduce the accuracy results of CPT+ that were published on PAKDD paper, but when I run the implementation (as it currently is), the accuracies that I get are way off (for some datasets).

For example SIGN, MSNBC get around 2% and 30% respectively. In the paper for the same datasets CPT+ has accuracies around 33% and 61% respectively.

Could you please provide me the profiles that you used to publish the results of the paper ?

(Sorry if you have done already and I have missed it)

Rafael.

[tedgueniche]

@rafkt, I apologize for the ridiculous delay. As a reference for future reader: if you clone this repo and run ipredict/controllers/MainController.java you will get the similar results as the paper. I just got 33% accuracy for CPT+ on SIGN and 59% on MSNBC.

The profiles in this repo are the same as the one in the research papers.