Thanks Matt,

With "Stop and Start" I mean when I run the python script again.

I run a python script to initialize a SP and run the "compute" function,
for a given input. Lets say I get 10001000000000001 [Active columns]. If I
run the compute function several times within the same script I get the
same output. This is expected, so good.

However, When I run the script again I expect to see. the same output as
before , however I get a different SDR.

For instance
>>Run input 1001001 with 5 computes for a given input
>>10001000000000001
>>10001000000000001
>>10001000000000001
>>10001000000000001
>>10001000000000001
>>Run input 1001001 again with 5 computes
>>10000000000001001
>>10000000000001001
>>10000000000001001
>>10000000000001001
>>10000000000001001


Regards
Chandan

On Mon, Nov 24, 2014 at 10:46 AM, Matthew Taylor <[email protected]> wrote:

> Chandan,
>
> What do you mean by "stop and start the SP"?
>
> You can persist a model through the OPF with the save() function [1].
> You can resurrect a persisted model with
> ModelFactory.loadFromCheckpoint() [2].
>
> [1]
> http://numenta.org/docs/nupic/classnupic_1_1frameworks_1_1opf_1_1model_1_1_model.html#aba0970ece8740693d3b82e656500a9c0
> [2]
> http://numenta.org/docs/nupic/classnupic_1_1frameworks_1_1opf_1_1modelfactory_1_1_model_factory.html#a73b1a13824e1990bd42deb288a594583
>
> Regards,
> ---------
> Matt Taylor
> OS Community Flag-Bearer
> Numenta
>
>
> On Mon, Nov 24, 2014 at 10:35 AM, Chandan Maruthi
> <[email protected]> wrote:
> > Hi All,
> > I have a few questions on the Spatial Pooler.
> >
> > 1. Once start a SP,  it gives me the same SDR for a given input if I run
> is
> > several times. This is great and expected. However the SDRs look
> different
> > if I were to stop and start the SP . Is this normal? Does this mean that
> any
> > learning from past runs is lost when I run the system again?
> >
> > 2 . How do you persist state of a HTM? and ensure that when you run it
> again
> > it works from where it left. And get the same SDR .
> >
> > Regards
> > Chandan Maruthi
> >
> >
> >
>
>

Reply via email to