>.Please go through a paper "Ensemble of Exemplar-SVMs for Object Detection 
and Beyond" by Tomasz Malisiewicz (ICCV 2011)
Thanks, will go through the paper.

>During testing, a special calibration techniques used to choose the best 
output. On terms of calibration, differently sized templates will produce 
different scores so you have to do something like Platt's method. 


> Why do you want to do that?
> To be more specific, how do you quantify success of your task?
> By definition, probabilities are "normalized" in the sense that the
> are guaranteed to live in the [0-1] range. However the  classifier
> models can be predict arbitrarily bad probabilities. For instance a
> badly trained or badly parameterized binary classifier could predict 0
> proba for the positive class 100% of the time.


The following sums the problem:
> [Classifiers are trained on similar type datasets, difference being their 
sizes and the way each result might be used after classification]. I am
 using SGDClassifier to train the individual classifiers, and need to choose
 the best amongst them. But as I understand I would need to normalize first
 before comparing them and was not sure how to calibrate them as such. Any
 pointers to would be helpful.
- I would quantify a most similar document as a  successful choice from the 
  results of ensemble of classifiers. I was a bit confused on how to 
  weight the predictions from each classifier so as to compare their
  values.
- The motivation for having multiple classifiers was mostly due to the
 logical separation of the dataset documents' property in my problem.
 [From the implementation perspective the training multiple classifiers
 saves on time (~7 min vs ~17 min) and space (2.2G vs 12G)]
--[Apologies on the duplicate thread, seems that it might take more than
 6hrs for the post to show up]

Thanks,
Abhi









------------------------------------------------------------------------------
Rapidly troubleshoot problems before they affect your business. Most IT 
organizations don't have a clear picture of how application performance 
affects their revenue. With AppDynamics, you get 100% visibility into your 
Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro!
http://pubads.g.doubleclick.net/gampad/clk?id=84349351&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to