Sergey, I haven't worked much with Metrics, so I can't answer your question. But hopefully you've seen these docs? https://github.com/numenta/nupic/wiki/Online-Prediction-Framework#metrics
I'm hoping to add another tutorial expanding the hot gym one that includes deeper use of Metrics. --------- Matt Taylor OS Community Flag-Bearer Numenta On Mon, Apr 14, 2014 at 5:57 AM, Sergey Bryukov <[email protected]> wrote: > Hello! > > Im analysing signal and need to evaluate window of 2 000 000 ticks. > > Object also has "silence" phase when there is no activity but some noises. > > Watching to this huge window human can recognize object behaviour changing > related to "silence". > > So I have an idea to train system to predict "silence", then I would stop > training and start watching for 2 000 000 window to recognize anomaly. > > Is hotgym good example to use? How to set up better MetricSpec() and > model_param.py in terms of described problem. > > MetricSpec(field='P0', metric='multiStep', > inferenceElement='multiStepBestPredictions', > params={'errorMetric': 'altMAPE', 'window': 2000000, 'steps': > 5}) > > > > > > > > > _______________________________________________ > 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
