>
> Just to check that there is no way of passing a vector of C's
>
> Use Case:
>
> Large Number of Dummy Variables or other sparse data.
> Normally you would normalise your inputs and have common C
> but then you lose sparsity increasing memory consumption and make
> calculations longer....
>
> Do you agree a) that one can't b) that its important?
> [ learning rate for stoch gradient descent would also benefit from indiv
> learning rate]
>
> b) no: it seems to me that you can rescale your variables to
> achieve the equivalent effect. How you rescale them will depend on the
> loss, and you'll have to work out the math.
>
> I agree I could rescale my inputs. But imagine I have 4 million rows with
> say 1000 dummy variables representing 10 categorical variables. If I leave
> the dummy variables unrescaled I will have only 10 numbers to store for
> each row. If I normalise by subtracting the column means/ dividing by
> standard deviation, I have to store 1000 variables per row.
>
> sean
>
>
------------------------------------------------------------------------------
Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
Discover the easy way to master current and previous Microsoft technologies
and advance your career. Get an incredible 1,500+ hours of step-by-step
tutorial videos with LearnDevNow. Subscribe today and save!
http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
