You the Man! Sabutai.
It looks great. I am able to get predictions with values close to what I am
expecting. :-)


On Wed, Dec 3, 2014 at 5:12 PM, Subutai Ahmad <[email protected]> wrote:

>
> You might be able to use:
>
> sp.getInputData('bottomUpIn')
>
> I'm not sure if the input is saved reliably though.
>
> --Subutai
>
>
> On Wed, Dec 3, 2014 at 5:03 PM, Chandan Maruthi <[email protected]
> > wrote:
>
>> Sabutai,
>> I am using the Network Api.
>>
>> This code sp = net.regions.get('SP', None)
>>
>> Lets me get access to the Sp object i guess. From there i am unable to
>> find a function to provide Sp input given an sp output. Does that make
>> sense
>>
>>
>> Chandan
>>
>>
>> On Wednesday, December 3, 2014, Subutai Ahmad <[email protected]>
>> wrote:
>>
>>> Hi Chandan,
>>>
>>> Are you using the OPF? If so, you should be able to do something like
>>> the following:
>>>
>>> sensor = model._getSensorRegion()
>>> sensor.getOutputData('dataOut')  # numpy array with the current sensor
>>> output
>>>
>>> sp = model._getSPRegion()
>>> sp.getOutputData('bottomUpOut') # current SP output
>>>
>>> Otherwise if you are using the Network API directly, you can get the SP,
>>> sensor, etc. with calls something like:
>>>
>>> sp = net.regions.get('SP', None)
>>>
>>> --Subutai
>>>
>>> On Wed, Dec 3, 2014 at 4:36 PM, Chandan Maruthi <
>>> [email protected]> wrote:
>>>
>>>> Hi,
>>>> I have a question on Spatial Pooler Reconstruction /Decoding/
>>>> Classification question
>>>>
>>>> I am using a HTM network with 1 sensor, 1 spatial pooler and 1 temporal
>>>> pooler.
>>>> I see that I can get the predicted temporal columns, which are the same
>>>> as the spatial pooler output.
>>>> Now I want the corresponding spatial pooler input which essentially is
>>>> the encoder output
>>>> so i can decode it and get the actual raw value.
>>>>
>>>> I am unable to find a function to get the SP input from a given SP
>>>> output. what am I missing?
>>>> I am tending to write some code to create a mapper to store SP input
>>>> and outputs is that the way to go?
>>>>
>>>> Regards
>>>> Chandan
>>>>
>>>>
>>>
>

Reply via email to