ok. First I tested with KNNClassifier. I didnt get good results.
Then I tested Subutai's reply (actually I think Matt was talking about this
classifier too) and it seems it worked good :)
I just want someone to confirm I'm doing it right. This is the piece of
code:
# Code Start
1 - classifier_region = network.addRegion('classifier',
'py.CLAClassifierRegion', json.dumps({'steps': '1', 'implementation':
'cpp'}))
2 - classifier_region.setParameter('learningMode', True)
3 - classifier_region.setParameter('inferenceMode', True)
4 - params = {'bucketIdx': encoder.getBucketIndices({'consumption':
consumption, 'timestamp': timestamp}), 'actValue': consumption})
5 - predicted_columns = sp_region.getOutputData('topDownOut').nonzero()[0]
6 - result = classifier_region.getSelf().customCompute(i,
predicted_columns, params)
7 - inference = result['actualValues'][numpy.argmax(result[1])]
# Code End
I'm not sure if the line 5 is correct. I mean if I should use topDownOut as
the input for the Classifier.
And maybe the line 4 is wrong too.
2015-01-19 20:54 GMT-03:00 Subutai Ahmad <[email protected]>:
> Hi Ricardo,
>
> The CLAClassifier is responsible for doing this. It is in:
>
> nupic/algorithms/CLAClassifier.py
>
> There is also a faster C++ version. If you are using the Network API,
> there is a CLAClassifierRegion. The OPF uses this to map SDR's back to the
> input values.
>
> --Subutai
>
>
> On Mon, Jan 19, 2015 at 9:24 AM, Ricardo Franco <[email protected]>
> wrote:
>
>> Lets consider a SpatialEncoder that encode the value 99 to [0 0 0 0 1 1]
>>
>> The SpatialPooler will receive this [0 0 0 0 1 1] and output a totally
>> different thing to 'bottomUpOut' output (this new array/matrix is the SDR,
>> right?)
>>
>> Lets the SDR is [0 0 1 0 0 0 1 0 1 0 0 1 1 0].
>>
>> Now how to get the SDR back to 99? Is this possible?
>>
>> --
>>
>> Ricardo Franco Andrade
>>
>> *Web Developer*
>>
>> email: [email protected]
>> skype: ricardo.krieg
>> phone: +55 (86) 9569 8521
>> linkedin: http://br.linkedin.com/in/ricardokrieg/
>> github: https://github.com/ricardokrieg
>>
>>
>>
>
--
Ricardo Franco Andrade
*Web Developer*
email: [email protected]
skype: ricardo.krieg
phone: +55 (86) 9569 8521
linkedin: http://br.linkedin.com/in/ricardokrieg/
github: https://github.com/ricardokrieg
