Due to recent updates, all users are required to create an Altair One account to login to the RapidMiner community. Click the Register button to create your account using the same email that you have previously used to login to the RapidMiner community. This will ensure that any previously created content will be synced to your Altair One account. Once you login, you will be asked to provide a username that identifies you to other Community users. Email us at Community with questions.
When does a grid search end?
Hello everyone,
I got a quick question and I think you guys can help me out:
When I do a grid search for optimal parameters C (11 different values) and Gamma (10 different values) for my SVM, when will the search stop?
Does it stop immediately the first time it hits a 100% accuracy with an "optimal pair" and then disregards the other untested pairs (assuming there will be an 100% accuracy)?
I also ask this question against the background of time-intense calculations..
Thanx a lot in advance,
Sasch
I got a quick question and I think you guys can help me out:
When I do a grid search for optimal parameters C (11 different values) and Gamma (10 different values) for my SVM, when will the search stop?
Does it stop immediately the first time it hits a 100% accuracy with an "optimal pair" and then disregards the other untested pairs (assuming there will be an 100% accuracy)?
I also ask this question against the background of time-intense calculations..
Thanx a lot in advance,
Sasch
0
Answers
no, Grid Search will test ALL combinations. You might be surprised, but I NEVER saw any 100% accuracy if validated correctly on a real world data set. If you receive a 100% accuracy you rather should double check your process setup than searching for a way to abort calculations...
Greetings,
Sebastian
Ok from the start:
I have 54 example sets of EEG data (all filtered and baseline corrected), each containing ~180 attributes + 1 label.
I have 3 different labels (=classes) that means 18 example sets for each class. The labels are numbers but "transformed" into text with the import wizard.
So when I now import the whole dataset and do a grid search for best SVM params and cross-validations and so on, I get a 100% accuracy (?).
Can't that be? Or is it in the nature of the SVM that it can find a optimal 100% solution when the number of example sets are much less than the number of attributes (=> "curse of dimensionality"?) ?
Perhaps my process setup is also wrong. I'd appreciate it if you would be so kind to check it. I put the code below:
Sorry I have to cut it into 2 pieces. Hope you don't mind.
So here's part 1:
Sasch