Hi Neal, net.initialize() can be called after the network structure has been set and before run(). However if you don’t call it, run() will do it for you so you usually don’t have to worry about it (I think this is actually mentioned in the wiki docs). I believe initialize sets up various memory buffers, etc. once the network structure has been finalized.
net.run(n) performs n iterations of compute for each Region in the correct order. Each time the ImageSensorRegion's compute is called, and assuming some Explorer is initialized, it updates the image position. The image code has not been used in many years and I barely remember some of the details. I hope it still works! It would be great to have a super-simple working example of this as many people have wanted to setup vision networks. If you create one we could perhaps stick it in our examples directory. Examples: there is some test code, such as tests/integration/py2/nupic/engine/network_serialization_test.py. Hope this helps! Please keep us updated on your progress. —Subutai On Thu, Jan 2, 2014 at 10:17 AM, Matthew Taylor <[email protected]> wrote: > Neal, > > I'm just starting to get the Network (and Region, Link, etc) API > documented, so there is a minimal doc: > http://numenta.org/docs/classnta_1_1_network.html. Not sure how > helpful this is to you at this point. I can't answer your other > questions, but it does point out holes in our docs. I've created a > ticket to address this at least from an API docs standpoint: > https://github.com/numenta/nupic/issues/525. > > I'm not sure how much time Subutai or Scott have to further explain > the Network API, but I'd have to defer your other questions to them. > Have you seen this video of us discussing the API a couple months ago? > http://www.youtube.com/watch?v=uVgIcwEwKHg > > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > > > On Sun, Dec 29, 2013 at 9:05 PM, Neal Donnelly <[email protected]> wrote: > > Hey everyone, > > > > I'm writing again to the listserv for pointers on my ongoing senior > thesis > > using NuPIC. I've scaled back my ambitions quite a bit after some sorely > > needed and well written advice from Fergal. I'm now trying to simply use > the > > spatial pooler as a sparse auto-encoder to do feature detection in > images. > > To start out with a simple case, I'm trying to build a network with an > > ImageSensor at the bottom, several SPRegions, then an SVM ClassifierNode > at > > the top. In general, I've found the in-code documentation for the > regions/ > > directory to be fantastic and the nodeSpecs are very helpful. I think I > have > > a good understanding of how to add regions and links, but I have a few > > questions about running the network. > > > > when do I call net.initialize() ? What does that do? > > what is the relationship between calling net.run(), which I assume it > > advances everything one step, and the ImageSensor region reading in > images > > and moving across them with the explorer? > > are there more examples of using the Network Engine? I've been working > off > > the CLA implementation but it would be cool if there was something else > that > > uses it. > > is there more documentation about using the Network Engine besides the > stuff > > at https://github.com/numenta/nupic/wiki/Nupic-architecture? > > > > Of course I'd love to hear any comments or tips anyone thinks is helpful. > > Thanks so much to everyone who's responded to any of my messages. > > > > Happy holidays, > > Neal > > > > _______________________________________________ > > 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 >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
