opss.. *and due to classifier code to handle this kind of predictions be entirely Numenta property in addition to this code is not importable (as I far I know).
Sorry for my typos, I'm a country bumpkin.. hehe On 6 October 2014 08:49, David Ragazzi <[email protected]> wrote: > Hi Guys, > > As I said in previous messages, I have been working on NuPIC encoders > integration to NuPIC Studio. The proof of concept is create a project for > HotGym example which use 2 types of encoders (DateEncoder and > RandomDistScalarEncoder). However, during this work I had some problems > that delayed (a lot) my job: > > 1. HotGym is a neural network with 2048 columns * 32 cells: this means > 65536 objects to be shown (without mention to sensor bits)! Because this > big number of objects, the simulations was very, very slow. My solution was > hiding inactive cells and show all proximal segments in order to user > visualize the limits of the network without loose performance. I don't like > this kind of visualization (I like hide inactive proximal segments and show > all cells), but this was the only soultion that I've found (at least, > temporarily). > > 2. RandomDistScalarEncoder is a delta encoder (.i.e. its min and max > values change along time) and due to classifier code to handle this is kind > of predictions be entirely Numenta code and the same code is not importable > (it's located at CLAModel.py), I can't simply replicate this code due to > copyright issues. So, I'm using the original ScalarEncoder as it can > replace RDSE and perform well the same job (thank for this, Scott!). > > The two jobs above are solved, now the next challenge is optimize the > front-end OO code which handle NuPIC in order to work faster. > > Regarding video tutorial, Matt kindly will create it on his free time, but > due to the encoders integration still is not done, I asked to him to hold > on. In meantime, anyone is able to handle pure HTM inputs without encoders, > i.e. pure arrays of 0's and 1's as shown in "Squares" project. Have fun! > > Finally, I'd like apologize you guys due to my delay to answer some > questions sent in private and public boxes, as you have seen there's a > plenty of work to a single person do. So, please be patient.. :-( > > Best wishes, > -- > David Ragazzi > MSc in Sofware Engineer (University of Liverpool) > OS Community Commiter at Numenta.org > -- > "I think James Connolly, the Irish revolutionary, is right when he says that > the only prophets are those who make their future. So we're not > anticipating, we're working for it." > -- David Ragazzi MSc in Sofware Engineer (University of Liverpool) OS Community Commiter at Numenta.org -- "I think James Connolly, the Irish revolutionary, is right when he says that the only prophets are those who make their future. So we're not anticipating , we're working for it."
