Question on swarm results in hotgym example

When I run on the hotgym prediction example I get a model that seems to work quite well. However when I look at the model parameters in that the swarm produces in I am surprised that it does not seem to encode the date in a way which includes whether it is a weekend or not. What I get is:

'sensorParams': { 'encoders': 
  { '_classifierInput': 
    { 'classifierOnly': True,
      'clipInput': True,
      'fieldname': 'kw_energy_consumption',
      'maxval': 53.0,
      'minval': 0.0,
      'n': 362,
      'name': '_classifierInput',
      'type': 'ScalarEncoder',
      'w': 21},
     { 'clipInput': True,
       'fieldname': 'kw_energy_consumption',
       'maxval': 53.0,
       'minval': 0.0,
       'n': 271,
       'name': 'kw_energy_consumption',
       'type': 'ScalarEncoder',
       'w': 21},
    u'timestamp_dayOfWeek': None,
     { 'fieldname': 'timestamp',
       'name': 'timestamp',
       'timeOfDay': ( 21, 8.421340604599116),
       'type': 'DateEncoder'},
    u'timestamp_weekend': None
   'sensorAutoReset': None,
   'verbosity': 0},

Am I misinterpreting this – is this what you expect?

You are interpreting this right. The swarm finds the best parameters it can find within limits. We have found it’s a good tool to use to help narrow down the search space, but sometimes it missing things. A “medium” swarm for example, will use many less permutations of encoder parameters and might not find that a field should be encoded because the parameter search space is too small.

My advice to you is if you see a weekend vs weekday pattern, include the weekend encoder. Your brains knows better than the swarm.