The Altair Community is migrating to a new platform to provide a better experience for you. In preparation for the migration, the Altair Community is on read-only mode from October 28 - November 6, 2024. Technical support via cases will continue to work as is. For any urgent requests from Students/Faculty members, please submit the form linked here
Probabilities from LogisticRegression
Contributor II
Hi, I am trying to find a way of extracting probabilities out of a LogisticRegression model.
I found a post which talks about "converting" confidence values into probabilies using the PlattScaling operator.
However, I am finding that the confidence values which are produced following PlattScaling are not the logistic regression probabilies that I would expect (i.e. taking the callibration parameters for the logistic regression generated by RM (which are spot-on) and manually computing the probability as prob = exp / [1 + exp] ).
what am i doing wrong? and is there a way to extract the logit probabilities?
many thx.
I found a post which talks about "converting" confidence values into probabilies using the PlattScaling operator.
However, I am finding that the confidence values which are produced following PlattScaling are not the logistic regression probabilies that I would expect (i.e. taking the callibration parameters for the logistic regression generated by RM (which are spot-on) and manually computing the probability as prob = exp / [1 + exp] ).
what am i doing wrong? and is there a way to extract the logit probabilities?
many thx.
0
Answers
I think this is because the PlattScaling uses another algorithm to determine the probabilities. But I'm not that familiar with this algorithm, that I could describe what's happening there...
Greetings,
Sebastian
further to my original post: the LogLikelihood figure produced by RM is what i would expect if regression probabilities were p = exp / [1+exp ]
which is not the case if one was to compute the LogLikelihood with produced Confidence values both before and after PlattScaling. (both of which result in poorer fit, in my particular case).
Try using the Weka's LogistRegression operator. It's faster and it will give you the probabilities you want (after ModelApplier).