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]
> 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
>
>


-- 
*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]
http://cortical.io

Reply via email to