I'll just throw out a thought that I had while reading through this thread. For the 2D visualization, you could make use of transparency for the nodes that are not active. Or better still, the most recently active nodes are the most opaque, but then they begin to fade to transparent the longer they remain idle. Predictions vs. activations could continue to be done as they currently are using different colors. The fading effect might give you enough information to see the time evolution of the network, but if that's insufficient you could also implement a kind of thumb-dial GUI widget that lets you rapidly scroll through the temporal sequence of activations like thumbing through a stack of cards.
FWIW, Eric On Wed, Jun 18, 2014 at 3:13 PM, Jim Bridgewater <[email protected]> wrote: > Jumping on the 2D band wagon. It would be nice to see how permanences > evolve over time for 2D inputs in visual applications. > > On Wed, Jun 18, 2014 at 1:03 AM, Ruaridh O'Donnell > <[email protected]> wrote: > > Hello Yuwei, > > > > Thanks for the input. > > > > I agree that visualising 2D sdrs is hard. Be able to see 2D SDRs is > > something I'd like to do. I have some ideas but it'll take a bit more > work. > > > > The sequence memory (or temporal memory as i think it's called now) is > > definitely something I'd like to see visualised. Its really hard to see > > what's gong on and it's one of the most important bits. > > Part of the challenge is that there's so much data. I think it'll be a > > matter of choosing carefully what to cut out. And then having different > > visualisations to show different parts of the data. > > > > Ruaridh > > > > Hi Ruaridh, > > > > The visualization will be extremely helpful to understand what's going > on. > > The SDR viewer is very intuitive. Are you planning also to add topology? > It > > can be challenging to visualize the evolution of 2D SDR over time. > > > > In addition to the spatial pooler output. It will also be useful to > figure > > out how to visualize the synaptic connectivity pattern between cells in > the > > temporal pooler during sequence learning. This can be challenging if you > > have multiple cells per column. > > > > Also, matplotlib seems to be a good library for scientific > visualizations. > > > > Yuwei > > > > > > On Tue, Jun 17, 2014 at 2:23 PM, Ruaridh O'Donnell > > <[email protected]> wrote: > >> > >> Hello everyone, > >> I'm one of the students in this year's summer of nupic. I'm creating > >> something that will hopefully be useful for the community so I thought > I'd > >> get some input on it from everyone. > >> > >> I'm creating a library of plots and visualisations for use with nupic. > The > >> idea is to write a collection of functions that you can call on a model > to > >> easily see whats going on inside it. Each will output some kind of plot > >> describing the state of the model over time. As a very basic example, a > >> simple line plot of SDR sparsity. > >> I'd like these plots to be useful to the community so I'm asking if > anyone > >> has ideas for visualisation they think would be useful. > >> > >> I made a mock up of a simple SDR viewer which you can see here. It might > >> give a better idea of the kinds of things I'm thinking about. > >> > >> > http://nbviewer.ipython.org/github/rhuairahrighairidh/SoN/blob/master/Basic%20SDR%20Plot.ipynb > >> > >> Anything else visualisation related let me know as well. I'd like to > hear > >> about people's ideas / experience in this area. > >> > >> Ruaridh > >> > >> > >> > >> _______________________________________________ > >> nupic mailing list > >> [email protected] > >> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > >> > > > > > > > > -- > > -- > > Yuwei Cui > > > > Neuroscience and Cognitive Science (NACS) > > 1207 Bioscience Research Building > > University of Maryland, College Park, MD, 20742 > > > > Tel: +01-(301)-405-1588 > > Email: [email protected] > > Homepage: http://terpconnect.umd.edu/~ywcui/ > > > > _______________________________________________ > > nupic mailing list > > [email protected] > > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > > > > > _______________________________________________ > > nupic mailing list > > [email protected] > > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > > > > > -- > Jim Bridgewater, PhD > Arizona State University > 480-227-9592 > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
