Just a quick correction: the HTM is always using the time series of the variable you are predicting. If the swarm did not pick up any additional fields, it just means these additional fields did not contribute in lowering the error score. Your data could still be very predictable based on the temporal behavior of the original predicted field.
--Subutai On Mon, Sep 14, 2015 at 6:26 PM, Matthew Taylor <[email protected]> wrote: > I ran the swarm in your project myself, and I get the same results as > you. The swarm results tells you basically that your data is not > predictable by NuPIC. > > It looks like you are trying to predict a string category "action". > Your data fields all look rather random when plotted. Can you talk > about what they actually represent? If they are indeed random, it > makes sense that NuPIC cannot use it to predict it. > > Also, there is a problem with your min/max values in your swarm > description. Looks like the max 10, but your data is more than 10. > However, even after I updated those max values in the swarm > description, the results were the same. > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > > > On Sun, Sep 13, 2015 at 10:05 AM, Jevgenijs K <[email protected]> wrote: > > from codedhard > > > > My code sample is based on Hot Gym. > > > > The HTM must make the output based on the data1, data2, ... > > > > Could be something wrong as how "string" type is handled. > > > > > > After swarming > > At the end it states Field Contributions: 0.00 ... > > The encoders data are None. > > Tried inferenceType TemporalMultiStep and TemporalClassification. > > > > > > > > nupic version 0.3.1.dev py 2.7 > > https://github.com/codedhard/nupic_experiments- > >
