Swarming Nupic Data

Engineering NuPIC
1

After swarming the hot gym why do we get u’timestamp_dayOfWeek’: None even when i swarm my data some feature are depicted as none can anyone please tell what could be the issue.

2

Swarming finds that there is a weekly pattern, so it is using a date encoder that differentiates the days of the week differently. Does that make sense? Can you ask your question another way?

3

Hi Rhyolight,

Actually I swarm the input hot gym data the model params file contains

u’kw_energy_consumption’: { ‘clipInput’: True,
‘fieldname’: ‘kw_energy_consumption’,
‘maxval’: 53.0,
‘minval’: 0.0,
‘n’: 29,
‘name’: ‘kw_energy_consumption’,
‘type’: ‘ScalarEncoder’,
‘w’: 21},
u’timestamp_dayOfWeek’: None,
u’timestamp_timeOfDay’: { ‘fieldname’: ‘timestamp’,
‘name’: ‘timestamp’,
‘timeOfDay’: ( 21,
6.090344152692538),
‘type’: ‘DateEncoder’},
u’timestamp_weekend’: { ‘fieldname’: ‘timestamp’,
‘name’: ‘timestamp’,
‘type’: ‘DateEncoder’,
‘weekend’: ( 21,
1)}

As u can see that timestamp_dayOfWeek has None so i just wanted to confirm what could be the problem due to which the feature timestamp_dayOfWeek is having None as encoding

4

Hi Rhyo,

In the similar fashion when i swarm my server data.I get

u’cpu_user_pct’: None,
u’load_1min’: { ‘clipInput’: True,
‘fieldname’: ‘load_1min’,
‘maxval’: 0.0525,
‘minval’: 0.0025,
‘n’: 380,
‘name’: ‘load_1min’,
‘type’: ‘ScalarEncoder’,
‘w’: 21},
u’memory_actual_used_pct’: { ‘clipInput’: True,
‘fieldname’: ‘memory_actual_used_pct’,
‘maxval’: 0.649,
‘minval’: 0.0847,
‘n’: 22,
‘name’: ‘memory_actual_used_pct’,
‘type’: ‘ScalarEncoder’,
‘w’: 21},
u’process_summary_total’: None
As you can see process_summary_total feature and cpu_user_pct have None as their encoding value so just wanted to confirm what could be problem because of which None is coming

5

Sorry I think I misunderstood your question. When one of these encoder fields is None that means the swarm found that encoding that input did not help improve prediction accuracy, so the field is simply not encoded.

Swarming is a brute-force solution to generate model Paramus. Look at your data, and if you see it has weekly patterns, add in the proper encoder.

If you are doing anomaly detection, we suggest you use the getScalarMetricWithTimeOfDayAnomalyParams() function to generate model parameters. I think this includes day of week encoding.

6

@rhyolight , mee too I used getScalarMetricWithTimeOfDayAnomalyParams() but this function work only on numeric data isn’t ?
It’s ture this function generate a model param but not the perfect model, I mean after you get your model you must make some updates ( spParams and tmParams ) and seen the best result isn’t.

7

Yes, it only works on a single numeric field. But you can modify the model parameters it creates to your liking.

1 Like
8

Thank You for the help

1 Like