Hi Kalle,

Resets are taken into account during swarms, and you should NOT add them to
the "includedFields" section of the swarming description. The error you see
is unrelated, however: https://github.com/numenta/nupic/issues/1775

Regards,

---------
Matt Taylor
OS Community Flag-Bearer
Numenta

On Wed, Feb 10, 2016 at 1:01 AM, Kalle Tammemäe <[email protected]>
wrote:

> Hi,
>
>
>
> I would also appreciate a little help.
>
> I’m experimenting with repetitive signals (rehabilitation) where it is
> possible to define start/reset of the typical exercising cycle.
>
> Sounds reasonable to this information to facilitate HTM training.
>
>
>
> As an example, the first rows of .csv file are following:
>
> sens1,sens2,sens3,sens4,sens5,sens6,sens7,sens8,sens9,reset
>
> int,int,int,int,int,int,int,int,int,bool
>
> ,,,,,,,,,R
>
> -593,-3910,-1066,55,13,33,-892,-1554,686,0
>
> <snip>
>
>
>
> It is not clear does swarming “see” and use reset field automatically.
> Should it be described in swarm_description.py file?
>
> Still, trying to do that i.e. adding section to includedFiels after input
> signal sections:
>
>     {
>
>       "fieldName": "reset",
>
>       "fieldType": "bool"
>
>     }
>
>
>
> …does give an error on swarming:
>
>     claDescriptionTemplateFile = options.claDescriptionTemplateFile)
>
>   File
> "/home/kalle/.local/lib/python2.7/site-packages/nupic/swarming/exp_generator/ExpGenerator.py",
> line 192, in _handleDescriptionOption
>
>     claDescriptionTemplateFile = claDescriptionTemplateFile)
>
>   File
> "/home/kalle/.local/lib/python2.7/site-packages/nupic/swarming/exp_generator/ExpGenerator.py",
> line 1200, in _generateExperiment
>
>     _generateEncoderStringsV2(includedFields, options)
>
>   File
> "/home/kalle/.local/lib/python2.7/site-packages/nupic/swarming/exp_generator/ExpGenerator.py",
> line 915, in _generateEncoderStringsV2
>
>     raise RuntimeError("Unsupported field type '%s'" % (fieldType))
>
> RuntimeError: Unsupported field type 'bool'
>
>
>
> I’m not using the latest nupic ( 0.3.6.dev0).
>
>
>
> Thanks,
>
> Kalle Tammemae
>
>
>

Reply via email to