I'm happy to work on a patch for this issue if it is indeed a bug - can
anyone explain what is meant by the error?  Has anyone encountered this?


On Wed, Nov 27, 2013 at 7:02 AM, Aaron Zimmerman <
[email protected]> wrote:

> All,
>
> Not sure if this is a bug or user error.  I can persist a model to a local
> file (running on mac os x mavericks).  When I load that model and persist
> it again (to the same place), I can no longer load from the checkpoint.  I
> dont see any obvious "version" flags changing in the files, anyone
> experienced this?
>
> Repro:
> model = ModelFactory.create(MODEL_PARAMS)
> ..do stuff ..
> model.save(SAVE_LOCATION)
>
> ..next run..
> model = ModelFactory.loadFromCheckpoint(SAVE_LOCATION)
> ..do stuff ..
> model.save(SAVE_LOCATION)
>
> ..next run..
> model ModelFactory.loadFromCheckpoint(SAVE_LOCATION) <-- fails
>
> I see the check in spatial_pooler.cpp:1325, but it isn't clear to me what
> this is checking, or why it would pass on the first re-inflation but fail
> on subsequent.
>
> Is there a better way to persist a model for multiple uses/training
> sessions?
>
> Thanks,
>
> Aaron Zimmerman
>
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to