I am quite new to using WEKA and I am hoping someone can possibly help me or point out where I am going wrong with using GridSearch in WEKA to find the best libSVM cost and gamma parameters for my training dataset.
When I set up the following GridSearch with 5-fold cross-validation:
However, when I use these values in libSVM with the same setup and dataset but without GridSearch the accuracy drops down to ~65%.
If, for example, I select cost and gamma values myself of 1 and 0.01 respectively, I can achieve an accuracy of 73% - which although lower than GridSearch, it is much higher than what I get using the "optimal parameters".
So why do the GridSearch parameters not give the same accuracy when used with libSVM? Where have I went wrong?
(Also, if I have specified gamma (Y-coord) to search between -5 and 2, why does it give me a value of 10?)
Any help with understanding this so I can get my optimal parameters would be very much appreciated.
Thanks.
When I set up the following GridSearch with 5-fold cross-validation:
weka.classifiers.meta.GridSearch -E ACC -y-property classifier.gamma -y-min -5.0 -y-max 2.0 -y-step 1.0 -y-base 10.0 -y-expression pow(BASE,I) -filter weka.filters.AllFilter -x-property classifier.cost -x-min 1.0 -x-max 10.0 -x-step 1.0 -x-base 10.0 -x-expression I -sample-size 100.0 -traversal COLUMN-WISE -log-file "C:\\Program Files (x86)\\Weka-3-7" -num-slots 1 -S 1 -W weka.classifiers.functions.LibSVM -- -S 1 -K 2 -D 3 -G 0.0 -R 0.0 -N 0.5 -M 40.0 -C 1.0 -E 0.001 -P 0.1 -model "C:\\Program Files (x86)\\Weka-3-7" -seed 1
this gives the result:Evaluation: Accuracy
Coordinates: [10.0, 1.0]
Values: 10.0 (X coordinate), 10.0 (Y coordinate)
which I take as meaning that the optimal cost and gamma values for this dataset with libSVM are both 10. This gives an accuracy of ~75%.However, when I use these values in libSVM with the same setup and dataset but without GridSearch the accuracy drops down to ~65%.
If, for example, I select cost and gamma values myself of 1 and 0.01 respectively, I can achieve an accuracy of 73% - which although lower than GridSearch, it is much higher than what I get using the "optimal parameters".
So why do the GridSearch parameters not give the same accuracy when used with libSVM? Where have I went wrong?
(Also, if I have specified gamma (Y-coord) to search between -5 and 2, why does it give me a value of 10?)
Any help with understanding this so I can get my optimal parameters would be very much appreciated.
Thanks.