Irwan, Swarming gets you the best parameters for a data set to create an OPF Model. You can swarm over as little or as much data as you wish. It's usually better to swarm over more data, but there is a point of diminishing returns. Once you have model params, you can create an OPF Model and immediately start feeding it data. A new Model is just like a section of blank neocortex, it has no memories. Just because it was created with model params created by swarming over the data doesn't mean this new model knows any of the patterns within the data, so there will still be a period of time where the model returns high anomaly scores and bad predictions while it learns the patterns in the data.
But all the time it sees rows of data, the model is learning. At the beginning, it knows nothing, it was just configured to have decent parameters for the data you pass it because you swarmed to get those best params. There's really not a "training set" and a real data set for NuPIC. You just feed it data and it gets better. The sine wave example is not the best one for this, because the data is generated -- not real world data. You might want to look at a more realistic example I just finished: https://github.com/numenta/nupic/wiki/Using-NuPIC#one-hot-gym-prediction-tutorial. --------- Matt Taylor OS Community Flag-Bearer Numenta On Mon, Apr 28, 2014 at 9:57 PM, Irwan Ramli <[email protected]> wrote: > Dear All, > > I have successfully ran the example of sine wave prediction. My > understanding is that opf create a model from the input data, but I dont > understand how to use that model to predict the new input data. For example > if I have 1000 rows data, and I want to use 700 rows data for creating > prediction model, and then the rest 300 rows data will be used for testing, > could anyone help me how to do that? > > Kind Regards, > Irwan > > _______________________________________________ > 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
