It's only for the call to anomalyLikelihood.anomalyProbability() that you need to fixup the input. I forget the reason, exactly, but it didn't like 2-tuple consisting of the coords as numpy array and a scalar value for radius, as is passed to model.run(). You may be able to do as Subutai suggest and pass a random value, too.
On Sat, Jun 6, 2015 at 11:35 AM, Pascal Weinberger < [email protected]> wrote: > But dont we mess with the scale then? that way you input the coordinate > array with the radius *added*... ? So each point gets shifted up and right > according to its radius... :S > > > On Sat, Jun 6, 2015 at 5:36 PM, Austin Marshall <[email protected]> wrote: > >> I ran into a similar problem. It had to do with how the input >> to anomalyLikelihood.anomalyProbability() was constructed. I had it as a >> 2-tuple consisting of coords + radius, and fixed it by using a single numpy >> array of length 3. See >> https://github.com/oxtopus/shakehack/commit/1f3d06f98cfde106e8d2cfc726ef47bbdbb18754#diff-6e38f16215ae91c11fc5c54b74c66d54L72 >> for >> concrete example. >> >> On Jun 6, 2015, at 12:57 AM, Pascal Weinberger <[email protected]> >> wrote: >> >> Hey all, >> >> I just ran into a problem with anomalylikelihood using the Coordinate >> Encoder. >> I get valueErrors when trying to calculate the anomalyLikelihood from >> iteration 600 onwards, which suggests there is a problem in the >> AnomalyLikelihood... >> I found the issue prevously here: >> >> http://lists.numenta.org/pipermail/nupic_lists.numenta.org/2015-January/010244.html >> And wanted to know if that solves it ? I'd try on my own but it takes me >> a night of runtime to get to the point of error so help would be cool :) >> >> --Pascal >> >> >> >
