Hey Ralf!
CONGRATULATIONS!!! I didn't even realize you had finished htm.JavaScript?
Nice going buddy!
On Tue, Jul 14, 2015 at 11:23 AM, John Blackburn
<[email protected] <mailto:[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] <mailto:[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] <mailto:[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]
<mailto:[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] <mailto:[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]
<mailto:[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] <mailto:[email protected]>
> http://cortical.io
--
/With kind regards,/
David Ray
Java Solutions Architect
*Cortical.io <http://cortical.io/>*
Sponsor of: HTM.java <https://github.com/numenta/htm.java>
[email protected] <mailto:[email protected]>
http://cortical.io <http://cortical.io/>
