I haven't looked at Theano's SGD yet - do they calibrate the learning
rate on held-out data or do they use an heuristic?

best,
 Peter

PS: the patching up that I did was not related to learning rate or the
learning algorithm in general. For the approach that I use I need to
mask features (much like removing them from the feature space - but
more efficient than making a copy of X).

2011/10/4 Gael Varoquaux <[email protected]>:
> On Tue, Oct 04, 2011 at 07:45:07AM +0200, Peter Prettenhofer wrote:
>> @alexandre: thanks; basically yes -> I use the SGD classifier from
>> Bolt instead of sklearn because I had to patch it up a bit.
>
> By the way, Nicolas Pinto mentionned to me that he had better
> classification rates with the theano SGD than the scikit's SGD. He
> mentionned that it might be related to rules tuning parameters such as
> the learning rate. I couldn't get him to tell me more (there was some
> beer involved) and he hasn't shown up on the mailing list giving more
> details.
>
> That's just a data point to say that it might be worth inspecting the
> theano SGD.
>
> Gael
>
> ------------------------------------------------------------------------------
> All the data continuously generated in your IT infrastructure contains a
> definitive record of customers, application performance, security
> threats, fraudulent activity and more. Splunk takes this data and makes
> sense of it. Business sense. IT sense. Common sense.
> http://p.sf.net/sfu/splunk-d2dcopy1
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>



-- 
Peter Prettenhofer

------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure contains a
definitive record of customers, application performance, security
threats, fraudulent activity and more. Splunk takes this data and makes
sense of it. Business sense. IT sense. Common sense.
http://p.sf.net/sfu/splunk-d2dcopy1
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to