hi,
I am interested in knowing what happens if we try to
regress wiht SVM with noise free data. In this case ideally all the samples
should fit into the model.
In gernel, this concept is right.
If we use epsilon-SVM regression, epsilon have to be zero. But what happens
with Support vectors?
yes, all of the data are regarded as SVs.
If they lie outside epsilon-tube that means they do
not fall on the model developed. Does this mean that SVM regression is not
able to find the best model with noise-free data?
yes, this case is regarded as the hyper-parameters of SVR are not correctly
selected.
then, you can be use existed approaches to find the suitable
hyper-parameters for SVR.
best regards,
chen-chia chuang
----- Original Message -----
From: "Dulakshi Karunasingha" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Friday, October 01, 2004 10:58 AM
Subject: SVM regression with noise-free data
> Hi
> I am interested in knowing what happens if we try to
> regress wiht SVM with noise free data. In this case ideally all the
samples
> should fit into the model.
> If we use epsilon-SVM regression, epsilon have to be zero. But what
happens
> with Support vectors? If they lie outside epsilon-tube that means they do
> not fall on the model developed. Does this mean that SVM regression is not
> able to find the best model with noise-free data?
>
> thank you.
> Dulakshi
>
|