A question about the multistep predictions.. AFAIK the current implementation is not a top notch quality, is it something like: get best prediction for the next step, use as input for desired N repeats,right?
Better would be if the api of multi step model allowed direct bit vector as input. Thus no (context) info is lost. Thoughts? ..sorry, typing on phone On 3/30/16, Matthew Taylor <m...@numenta.org> wrote: > On Wed, Mar 30, 2016 at 7:53 AM, Chittavajhula, Phanindra (MRO) < > pchittava...@marathonoil.com> wrote: > >> Based on the plot shown in the hot gym tutorial, predictions are one to >> one to the actual value. How to get the values beyond last date or >> timestamp ? That is, I want to see prediction when no actual value exist. >> >> > I think you should NOT use the InferenceShifter on the result objects in > order to show the prediction way out in the future on the chart. > > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > -- Marek Otahal :o)
