Fixed! http://numenta.org/docs/nupic/
---------
Matt Taylor
OS Community Flag-Bearer
Numenta


On Tue, Jul 14, 2015 at 9:23 AM, John Blackburn
<[email protected]> wrote:
> Looks like it might be time to run doxygen again! Last run in May 19.
>
> John.
>
> On Tue, Jul 14, 2015 at 5:14 PM, cogmission (David Ray)
> <[email protected]> wrote:
>> Hey John,
>>
>> Nice self-sufficient researching! I like that in ya' !!!
>>
>> Anyway, yes that last (stripNeverLearned) parameter was recently removed
>> last month. The file I gave you is older than that...
>>
>> Remember, NuPIC is ever evolving, and it is still technically "pre-release"!
>>
>> ;-)
>>
>>
>>
>> On Tue, Jul 14, 2015 at 11:09 AM, John Blackburn
>> <[email protected]> wrote:
>>>
>>> Thanks, Ralf,
>>>
>>> Actually that reminds me, David Ray kindly sent me the QuickTest.py
>>> example in Python so I just tried that. However, I ran into another
>>> problem: there seems to be some confusion about how many parameters
>>> sp.compute() takes (Spatial Pooler). In QuickTest.py the code reads
>>>
>>> sp.compute(encoding, True, output, False)
>>>
>>> However, on Github sp.compute takes only 3 parameters (apart from self):
>>>
>>>
>>> https://github.com/numenta/nupic/blob/master/nupic/research/spatial_pooler.py#L658
>>>
>>> So this causes a crash. I notice on the API docs the 4th parameter is
>>> indeed mentioned:
>>>
>>>
>>> http://numenta.org/docs/nupic/classnupic_1_1research_1_1spatial__pooler_1_1_spatial_pooler.html#aaa2084b96999fb1734fd2f330bfa01a6
>>>
>>> So I guess the 4th arg was recently removed. Pretty confusing!
>>>
>>> Can anyone shed light on this mystery?
>>>
>>> John.
>>>
>>> On Tue, Jul 14, 2015 at 12:25 PM, Ralf Seliger <[email protected]> wrote:
>>> > Hey John,
>>> >
>>> > why don't you try the QuickTest example in htm.java
>>> > (https://github.com/numenta/htm.java) or htm.JavaScript
>>> > (https://github.com/nupic-community/htm.JavaScript)? It involves the new
>>> > temporal memory, and stepping through the code with a debugger you can
>>> > easily study the inner workings of th algorithm.
>>> >
>>> > Regards, RS
>>> >
>>> >
>>> > Am 14.07.2015 um 11:39 schrieb John Blackburn:
>>> >>
>>> >> 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.
>>> >>>
>>> >>>
>>> >
>>> >
>>>
>>
>>
>>
>> --
>> With kind regards,
>>
>> David Ray
>> Java Solutions Architect
>>
>> Cortical.io
>> Sponsor of:  HTM.java
>>
>> [email protected]
>> http://cortical.io
>

Reply via email to