still some errors in learner

ai-se / SMOTE

0 stars 0 forks source link

still some errors in learner #13

Closed WeiFoo closed 8 years ago

WeiFoo commented 8 years ago

By looking at the HPC results, there're still some issues that the learner can't work well(I mean errors, either the data issues or some others) when tuning learner is set. need to debug and fix...

WeiFoo commented 8 years ago

To check where this bug happens, run the experiment individually for the following learners(feature number = 100 for simplicity):

[x] Naive_learner
[x] SMOTE
[x] Tune_Learner
[x] Tune_SMOTE

WeiFoo commented 8 years ago

the bug is due to the distribution of training data. In some circumstances, randomly splitting data might cause training data only including samples with one type of label, like "yes" label , without "no" samples!

fix: stratified sampling !