Hi John,

Check out the TutorialTemporalMemoryTest [1]. You can increase the verbosity 
level for more information.

[1] 
https://github.com/numenta/nupic/blob/master/tests/integration/nupic/algorithms/tutorial_temporal_memory_test.py
 
<https://github.com/numenta/nupic/blob/master/tests/integration/nupic/algorithms/tutorial_temporal_memory_test.py>

- Chetan

> On Jul 14, 2015, at 2:39 AM, John Blackburn <[email protected]> 
> wrote:
> 
> Thanks, Chetan,
> 
> Any tutorials, examples of how to use temporal_memory.py? The nice
> thing about old TP is it has an example: hello_tp.py.
> 
> John.
> 
> On Mon, Jul 13, 2015 at 7:55 PM, Chetan Surpur <[email protected]> wrote:
>> Hi John,
>> 
>> The TP is now called "Temporal Memory", and there's a new implementation of
>> it in NuPIC [1]. Please use this latest version instead, and let us know if
>> you still find issues with the results.
>> 
>> [1]
>> https://github.com/numenta/nupic/blob/master/nupic/research/temporal_memory.py
>> 
>> Thanks,
>> Chetan
>> 
>> On Jul 13, 2015, at 4:44 AM, John Blackburn <[email protected]>
>> wrote:
>> 
>> Dear All
>> 
>> I'm trying to use the temporal pooler (TP) directly as I want to get
>> into the details of how Nupic works (rather than high level OPF etc)
>> 
>> Having trained the TP I used this code to get some predictions:
>> 
>> for j in range(10):
>>   x=2*math.pi/100*j
>>   y=math.sin(x)
>> 
>>   print "Time step:",j
>> 
>>   for k in range(nIntervals):
>>       if y>=ybot[k] and y<ytop[k]:
>>           print "input=",x,y,k,rep[k,:]
>>           tp.compute(rep[k,:],enableLearn=False,computeInfOutput=True)
>>           tp.printStates(printPrevious = False, printLearnState = False)
>>           break
>> 
>> 
>> Here is the result I got:
>> 
>> Time step: 0
>> input= 0.0 0.0 9 [0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0]
>> 
>> Inference Active state
>> 0000000001 0000000000
>> 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000
>> 0000000001 0000000000
>> Time step: 1
>> input= 0.0628318530718 0.0627905195293 10 [0 0 0 0 0 0 0 0 0 0 1 0 0 0
>> 0 0 0 0 0 0]
>> 
>> Inference Active state
>> 0000000000 1000000000
>> 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000
>> 0000000001 0000000000
>> Time step: 2
>> input= 0.125663706144 0.125333233564 11 [0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
>> 0 0 0 0 0]
>> 
>> Inference Active state
>> 0000000000 0100000000
>> 0000000000 0100000000
>> Inference Predicted state
>> 0000000000 0000000000
>> 0000000000 1110000000
>> Time step: 3
>> input= 0.188495559215 0.187381314586 11 [0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
>> 0 0 0 0 0]
>> 
>> Inference Active state
>> 0000000000 0000000000
>> 0000000000 0100000000
>> Inference Predicted state
>> 0000000000 0000000000
>> 0000000000 1110000000
>> 
>> You can see that in time step 3, one cell (12th column) is shown as
>> being both in the active and predictive state, which I though was
>> impossible. (its inference active state is 1 and its inference
>> predicated state is 1)
>> 
>> Also if you look at time step 0, only 1 cell is in the predictive
>> state. However, the input that comes in at time step 1 activates the
>> colum to the right of this cell (the 11th slot is "1") so I would
>> expect the 11th column to have both cells active, the "unexpected
>> input state" but this does not happen.
>> 
>> Can anyone explain this?
>> 
>> John.
>> 
>> 
> 

Reply via email to