"Changing C in linear SVM has no effect"
Elisa0815
New Altair Community Member
Hello,
For a machine learning problem (sentiment analysis), I try different classifiers, also SVM.
So, I used the regular Support Vector Machine by RapidMiner Studio with a linear kernel (dot) and made different tests by changing C. This actually has no effect on the results. I've set the C to 1.000.000 and still no effect.
I also tried the Support Vector Machine (linear) to see, if there is any difference but it's the same result.
Changing C SHOULD have an effect, since it define, how strongly any outliers are punished. I didn't change any of the other parameters.
Has may anyone an idea, how that can be? Is there any explanation, how it can be possible that C has no effect?
I'm wondering, if I may did a mistake or if it is reasonable. If I look it up in literature, it always tells me that C infects the results.
I'm sorry, if this topic already occured in the forum. I couldn't find it.
Greets
Elisa
For a machine learning problem (sentiment analysis), I try different classifiers, also SVM.
So, I used the regular Support Vector Machine by RapidMiner Studio with a linear kernel (dot) and made different tests by changing C. This actually has no effect on the results. I've set the C to 1.000.000 and still no effect.
I also tried the Support Vector Machine (linear) to see, if there is any difference but it's the same result.
Changing C SHOULD have an effect, since it define, how strongly any outliers are punished. I didn't change any of the other parameters.
Has may anyone an idea, how that can be? Is there any explanation, how it can be possible that C has no effect?
I'm wondering, if I may did a mistake or if it is reasonable. If I look it up in literature, it always tells me that C infects the results.
I'm sorry, if this topic already occured in the forum. I couldn't find it.
Greets
Elisa
0
Answers
-
Hi there,
There is a useful discussion on StackExchange on this topic, http://stats.stackexchange.com/questions/31066/what-is-the-influence-of-c-in-svms-with-linear-kernel. Looks like C's effect can be small, depending on the data.0