java - Strange weka instance results -
There are strange results when using a J48 tree I need to classify the vector of 48 characteristics, which Works well, but when I try to "optimize", I get strange results.
I have a method classified:
classified public boolean (double feature 1, double feature 2, double [] featureweek) {for example twenty classified = new example (2+ feature a gender); ToBeClassified.setValue (0, Feature1); ToBeClassified.setValue (1, Feature2); For (Int i = 2; I
It works very well, and I hope that I am correcting things but I want to delete the example construct which is done on every method call , So I transferred the instance to twenty classified = new installation (48); Line in the body of the body - hence it is created only once, it actually works well, despite this fact, I get slightly different results compared to the other, say, 400 classification, is different (say Not for, wrong). But I do not see any reason for this ... I hope some people are using Waqas so that I will understand what's going on / wrong. (Yes, 2 + featurewalk lamps are 48).
Thanks and relation
It is not very likely that anything in J. 48 That's wrong. Classifier creation and classification are both determinants. I recommend posting a large chunk of your code, because it looks great.
For your 400 loop test: This should definitely produce the same result every time, no exceptions. Two thoughts:
-
Example values are similar to a model, so check it out. This will leave any bug in the instance.
-
Does the classification run in a multi-threading manner? Is there a shared data object?
Comments
Post a Comment