def _initializeEncoders(self, encoderSpec):
""" Initialize the encoders"""
#Initializing scalar encoder
if self.encoderType in ['adaptiveScalar', 'scalar']:
if 'minval' in encoderSpec:
self.minval = encoderSpec.pop('minval')
else:
self.minval = None
if 'maxval' in encoderSpec:
self.maxval = encoderSpec.pop('maxval')
else:
self.maxval = None
self.encoder=adaptivescalar.AdaptiveScalarEncoder(name='AdaptiveScalarEncoder',\
w=self.w, n=self.n, minval=self.minval, maxval=self.maxval, periodic=False, forced=True)
#Initializing category encoder
elif self.encoderType == 'category':
self.encoder=sdrcategory.SDRCategoryEncoder(name='categoryEncoder', \
w=self.w, n=self.n)
#Initializing date encoder
elif self.encoderType in ['date', 'datetime']:
self.encoder = date.DateEncoder(name='dateEncoder')
else:
raise RuntimeError(
'Error in constructing class object. Either encoder type'
'or dataType must be specified')
def getStats(self, stats):
BaseStatsCollector.getStats(self, stats)
# We include subencoders for datetime field if there is a variation in encodings
# for that particular subencoding
# gym_melbourne_wed_train.csv has data only on the wednesdays, it doesn't
# make sense to include dayOfWeek in the permutations because it is constant
# in the entire dataset
# We check for variation in sub-encodings by passing the timestamp field
# through the maximal sub-encoder and checking for variation in post-encoding
# values
# Setup a datetime encoder with maximal resolution for each subencoder
encoder = DateEncoder.DateEncoder(
season=(1, 1), # width=366, resolution=1day
dayOfWeek=(1, 1), # width=7, resolution=1day
timeOfDay=(1, 1.0 / 60), # width=1440, resolution=1min
weekend=1, # width=2, binary encoding
holiday=1, # width=2, binary encoding
)
# Collect all encoder outputs
totalOrEncoderOutput = numpy.zeros(encoder.getWidth(),
dtype=numpy.uint8)
for value in self.valueList:
numpy.logical_or(totalOrEncoderOutput, encoder.encode(value),
totalOrEncoderOutput)
encoderDescription = encoder.getDescription()
numSubEncoders = len(encoderDescription)
for i in range(numSubEncoders):
subEncoderName, _ = encoderDescription[i]
beginIdx = encoderDescription[i][1]
if i == (numSubEncoders - 1):
endIdx = encoder.getWidth()
else:
endIdx = encoderDescription[i + 1][1]
stats[self.fieldname][subEncoderName] = \
(totalOrEncoderOutput[beginIdx:endIdx].sum()>1)
decodedInput = encoder.decode(totalOrEncoderOutput)[0]
if VERBOSITY > 2:
print "--"
print "Sub-encoders:"
for subEncoderName, _ in encoderDescription:
print "%s:%s" % (subEncoderName,
stats[self.fieldname][subEncoderName])
