cjlin1
commented
7 months ago For the 1st, yes, we don't give probabilities if # data is too small.
For the 2nd, we agree that it's a problem. May you give us your data and settings so we can see if a better way can be done
On 2023-11-28 10:23, KKKaren wrote:
Hi,
I have some questions related to the implementation of the probability estimation of the one-class svm:
*
The first question is how to calculate the probability estimation when the number of positive or negative predictions is smaller than 5. I found that the probability estimation will not be calculated in this situation. Does it mean that the model can only generate the decision values rather than the probabilities? Why do we manually generate some predicted decision values in the positive area or the negative area to enable the probability estimation? *
The second question is when there are many decision values in the same value, the decision value of this value can be mapped into different intervals. Is it reasonable?
-- Reply to this email directly, view it on GitHub [1], or unsubscribe [2]. You are receiving this because you are subscribed to this thread.Message ID: @.> [ { @.": "http://schema.org", @.": "EmailMessage", "potentialAction": { @.": "ViewAction", "target": "https://github.com/cjlin1/libsvm/issues/210", "url": "https://github.com/cjlin1/libsvm/issues/210", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { @.***": "Organization", "name": "GitHub", "url": "https://github.com" } } ]
Links:
[1] https://github.com/cjlin1/libsvm/issues/210 [2] https://github.com/notifications/unsubscribe-auth/ABI3BHQHHY5NBDOW543LNGDYGVDI5AVCNFSM6AAAAAA746XKPWVHI2DSMVQWIX3LMV43ASLTON2WKOZSGAYTGNBWHE2DOOI
GoXian
Hi,
I have some questions related to the implementation of the probability estimation of the one-class svm:
The first question is how to calculate the probability estimation when the number of positive or negative predictions is smaller than 5. I found that the probability estimation will not be calculated in this situation. Does it mean that the model can only generate the decision values rather than the probabilities? Why do we manually generate some predicted decision values in the positive area or the negative area to enable the probability estimation?
The second question is when there are many decision values in the same value, the decision value of this value can be mapped into different intervals. Is it reasonable?