Quoting Subutai from http://lists.numenta.org/pipermail/nupic_lists.numenta.org/2014-December/010021.html :
"I would not recommend using NuPIC for image recognition. Today's implementation is not well suited for it. We need to make progress with temporal pooling and hierarchies before it will do well on end to end image recognition tasks." --------- Matt Taylor OS Community Flag-Bearer Numenta On Thu, Feb 5, 2015 at 7:18 PM, 天朗气清 <[email protected]> wrote: > hi > I'm having a research on the encoding of image in NuPIC. I have read the > code such as scalar.py, base.py in > https://github.com/numenta/nupic/tree/master/nupic/encoders，and I think I > have known how does ordinary data be encoded. But how to encode image data? > If there is a piece of image with 30*30 pixels, will Nupic encode it as 1d > 900 pixels? if not, how does it encode. > I'm trying to use NuPIC to study disease data. The data include 1 disease > and 90 symptoms which are intgers from 0 to 10. I need to judge what kind of > disease it is according 90 symptions. I think the symptoms just like the > pixels and the disease just like the image. So I use the code on > https://github.com/numenta/nupic.research/tree/master/image_test to have a > test. The accuracy is about 70%.I don't know how image data be encoded, so I > don't know wether I can regard the disease data as image data either, or do > you have any suggestion about how to process multivalue prediction and > catogery? > Hoping for your reply. > > > --yajingfu
