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. > >
