Thanks very much, Matthew. This will help many people. John
On Tue, Jul 14, 2015 at 5:35 PM, Matthew Taylor <[email protected]> wrote: > 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 >> >
