Sergey, thanks for your questions. I am in the process of updating these
examples, so getting your questions now is useful. I'm working on the
prediction example, and will move to the anomaly detection example
afterwards.

Unfortunately, I think there are some error with that example. It doesn't
look like the "aggregationInfo" section of the model_params is being used
at all. In fact, if you simple remove that block of configuration, the
example runs with exactly the same results.

I can at least answer a couple of your questions regarding aggregation. The
aggregationInfo section specifics how input data will be aggregated, and
this requires an aggregation function. Aggregating data is kindof like
summarizing it and making it smaller, so in order to do this there must be
a function applying to a set of data to turn it into one value. That's what
the 'sum' and 'first' mean. 'First' means to simply take the first value in
the set of input data to use as the representative within the aggregate
data set. This works best with temporal values, for example if you have
values every 5 minutes, but you just want 15 minute aggregate data, you
want to use the first time value in the data set to signify that time
period. The 'sum' is just the sum of all data fields within the data. I
imagine there is also a 'mean' or 'avg'.

I'll be getting deeper into this once I'm finished with
https://github.com/numenta/nupic/issues/548 and provide a better tutorial
for anomaly detection with the hotgym data set.

Regards,

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


On Wed, Apr 16, 2014 at 2:50 AM, Sergey Bryukov <[email protected]> wrote:

> Hello,
>
> Im extending  hotgym_anomaly example for multi denominational records and
> have question on how to fill MODEL_PARAMS correctly
>
> Let's assume we have fields:  timestamp, (float)P0, (float)P1
>
> 1. what is the place for P0 and P1 in below structure? What means 'sum'
> and 'first'
>
>  'aggregationInfo': {  'days': 0,
>         'fields': [(u'P0', 'sum'), (u'P0', 'first')],
>         'hours': 0,
>         'microseconds': 0,
>         'milliseconds': 0,
>         'minutes': 0,
>         'months': 0,
>         'seconds':1,
>         'weeks': 0,
>         'years': 0
> }
>
>
> 2. I need encode only seconds, how to set encoder properly for timestamp?
>    And what fields mean for P0:  'clipInput' 'maxval', 'minval', 'n', 'w'
>
> 'encoders': {   u'timestamp_timeOfDay':    {   'fieldname': u'timestamp',
>     'name': u'timestamp_timeOfDay',
>     'timeOfDay': (21, 0.5),
>     'type': 'DateEncoder'
> },
>                        u'timestamp_dayOfWeek': None,
>                        u'timestamp_weekend': None,
>
>                         u'P0':    {  'clipInput': True,
>                                         ' fieldname': u'P0',
>                                         'maxval': 100.0,
>                                         'minval': 0.0,
>                                         'n': 50,
>                                         'name': u'P0',
>                                         'type': 'ScalarEncoder',
>                                         'w': 21
>                                        }
>                         u'P1': {}
>
>                     }
>
>
>
> _______________________________________________
> 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

Reply via email to