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
