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
