John, The CPU example isn't going to perform well when run against the CPU activity of a laptop because individual human behavior (what usually drives laptop CPU) is hard to predict. But it could be very useful to make predictions against the CPU usages of application servers. Grok [1] is doing quite well flagging server anomalies by taking into account predictions of CPU usage on AWS instances.
[1] https://www.groksolutions.com/product.html --------- Matt Taylor OS Community Flag-Bearer Numenta On Mon, Mar 10, 2014 at 4:53 AM, John Blackburn <[email protected]> wrote: > I've also run the CPU example that comes with NuPIC. My concern with this > example is that there is no way the HTM can predict CPU usage! It does not > "see" when a user is about to launch a new program for example, so as far as > it is concerned the CPU use is completely random. Sure enough, the > predictions I saw in the CPU example where not very good, probably no better > than random guesses (I may be wrong on this, that was just my first > impression, please correct me!) > > I think you may encounter the same problem with your game. If the HTM does > not "see" baddies, collectables, walls etc, there is no way it can > realistically predict left, right, up, down key presses. > > We cannot expect HTM to be any smarter than a human. I don't think a human > could predict CPU usage in 10 seconds time without being able to see the > computer screen or be given any other information. > > What I think would be far more useful is for the HTM to PLAY the game rather > than observing it. Then there would be a sensori-motor loop and the HTM > would come to understand its world like an animal does and learn > sensory-motor skills like seeking food or avoiding walls. > > Just my $0.02. I admit to being a novice at all this! > > John. > > > On Sat, Mar 8, 2014 at 5:24 AM, Matt Roesener <[email protected]> wrote: >> >> Hi everyone, >> >> I've been following the Nupic community for sometime, this is my first >> post. >> >> I wanted to share some of my work. I modified the cpu example and created >> a simple game that takes in user input such as up, down, right and left keys >> and feeds these events into nupic. >> >> The idea is for nupic to learn and predict future events. I want to build >> on top of this overtime to create more sophisticated game events, bad guy >> moves, boundaries, etc.. >> >> I wanted to make sure my code is working correctly, specifically in terms >> of the model input and inference, as well as the correct encoders. Below is >> my code. >> >> Thank you! I've learned so much from just reading and experimenting with >> everyones code and ideas! >> >> https://github.com/roesenerm/nupicGame.git >> >> Matt >> >> >> >> >> _______________________________________________ >> 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
