Bug in heuristic computation of CELOE

SmartDataAnalytics / DL-Learner

A tool for supervised Machine Learning in OWL and Description Logics

http://dl-learner.org

GNU General Public License v3.0

152 stars 34 forks source link

Bug in heuristic computation of CELOE #62

Closed aris-github closed 6 years ago

aris-github commented 6 years ago

https://github.com/SmartDataAnalytics/DL-Learner/blob/7f3af9181b8c69cca5e19f32e25ea341b8f4c8f9/components-core/src/main/java/org/dllearner/algorithms/celoe/OEHeuristicRuntime.java#L66

Hi @JensLehmann @LorenzBuehmann @SimonBin , I am using DL-Learner for my research. I found a bug in the heuristic computation in the above line: the accuracy gain is wrong, it shall be conversely computed as:

 score += (score - parentAccuracy) * gainBonusFactor;

LorenzBuehmann commented 6 years ago

Hi Yingbing Hua,

seems like you're right. The formula for the accuracy gain is

acc_gain(C) = acc_c(C, t) − acc_c(C′, t)

with C being the current concept and C´ it's parent in the refinement tree.

LorenzBuehmann commented 6 years ago

@kit-hua Has been fixed in develop branch. thank you very much for the bug report.