On 1 February 2012 06:19, Olivier Grisel <[email protected]> wrote:
> 2012/1/31 Jeff Farris <[email protected]>:
> > I'm currently using pickle to persist models (e.g. SVC). After
> upgrading
> > sklearn, these pickled models from a previous version of sklearn don't
> tend
> > to work and then I need to retrain. Is there some version independent
> way
> > of saving models (e.g. libsvm model format) or other recommendations on
> how
> > to go about doing this without retraining all my persisted models after
> each
> > sklearn upgrade?
>
> Indeed this is an issue we will need to address at some point. I don't
> think python pickles are the solution for this.
>
> For some models it might be possible to write PRML importer and exporters:
>
> http://en.wikipedia.org/wiki/Predictive_Model_Markup_Language
> http://www.dmg.org/
>
> That would give us the additional benefit to be interoperable with the
> Google Prediction API and other tools from the analytics industry:
>
> https://code.google.com/apis/predict/docs/pmml-schema.html
> http://en.wikipedia.org/wiki/Predictive_Model_Markup_Language#PMML_Products
>
> --
> Olivier
> http://twitter.com/ogrisel - http://github.com/ogrisel
>
>
> ------------------------------------------------------------------------------
> Keep Your Developer Skills Current with LearnDevNow!
> The most comprehensive online learning library for Microsoft developers
> is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3,
> Metro Style Apps, more. Free future releases when you subscribe now!
> http://p.sf.net/sfu/learndevnow-d2d
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
For models that do not require storing a large dataset, JSON might be a
good option as well.
The interoperability of JSON is huge, and would also mean very little would
need to change between versions.
(if a new keyword is added, it can be ignored, with warning, by previous
versions and used in future versions).
Further, there isn't as much coding to be done for implementing a different
format (i.e. use simplejson and have the base classifier try to guess the
representation using introspection).
As for models that do require this, I'm not sure if more efficient
encodings exist for JSON.
BSON comes up in a google search, but I have no experience with that.
Thoughts?
- Robert
--
Public key at: http://pgp.mit.edu/ Search for this email address and select
the key from "2011-08-19" (key id: 54BA8735)
------------------------------------------------------------------------------
Keep Your Developer Skills Current with LearnDevNow!
The most comprehensive online learning library for Microsoft developers
is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3,
Metro Style Apps, more. Free future releases when you subscribe now!
http://p.sf.net/sfu/learndevnow-d2d
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
