I’m training a model for anomaly detection. The data is as follows
neuron_id,value
int,float
S,
6921,0.300356
6921,0.069099
6921,0.316716
6921,0.167633
6921,0.100218
…
The neuron_id field serves as a sequence identifier hence the S flag on the second line. The neuron_id value changes once every 50 lines to mark the beginning of a new sequence. How can I check that sequence reset is actually happening when the neuron_id value changes?
I was thinking that by printing the model at each step I would see the sequenceReset field going from 0.0 to 1.0 but this not the case. Here is the model output when the transition is supposed to happen (neuron_id going from 6921 to 53):
ModelResult( predictionNumber=49
rawInput={'value': 0.004258, 'neuron_id': 53}
sensorInput=SensorInput( dataRow=(0.004258,)
dataDict={'neuron_id': 53, 'value': 0.004258}
dataEncodings=[array([ 1., 1., 1., ..., 0., 0., 0.], dtype=float32)]
sequenceReset=0.0
category=-1
)
inferences={'multiStepPredictions': {1: {0.08826500000000001: 0.0026059335204494951, 0.1321598: 0.0028300311198308369, 1.564562: 0.0026570105870779375, 0.779939: 0.0028050158965931883, 0.261934: 0.0027851742030226227, 0.076092: 0.0026463728372261518, 0.004258: 0.90575743768592765, 0.756751: 0.002573166054106916}}, 'multiStepBucketLikelihoods': {1: {0: 0.90575743768592765, 8: 0.0026463728372261518, 9: 0.0026059335204494951, 76: 0.002573166054106916, 13: 0.0028300311198308369, 78: 0.0028050158965931883, 26: 0.0027851742030226227, 156: 0.0026570105870779375}}, 'multiStepBestPredictions': {1: 0.004258}, 'anomalyLabel': '[]', 'anomalyScore': 1.0}
metrics=None
predictedFieldIdx=0
predictedFieldName=value
classifierInput=ClassifierInput( dataRow=0.004258
bucketIndex=0
)
)
ModelResult( predictionNumber=50
rawInput={'value': 0.005851, 'neuron_id': 53}
sensorInput=SensorInput( dataRow=(0.005851,)
dataDict={'neuron_id': 53, 'value': 0.005851}
dataEncodings=[array([ 0., 1., 1., ..., 0., 0., 0.], dtype=float32)]
sequenceReset=0.0
category=-1
)
inferences={'multiStepPredictions': {1: {0.0627962: 0.0055714714736493946, 0.005851: 0.86610401636359924, 0.261934: 0.011780366887778429, 3.217478: 0.008001018185725078, 0.076092: 0.0090889321537235071, 0.336359: 0.004160104491713979, 0.0412563: 0.010323617261193156, 0.316716: 0.0042960821749932062}}, 'multiStepBucketLikelihoods': {1: {32: 0.0042960821749932062, 1: 0.86610401636359924, 34: 0.004160104491713979, 4: 0.010323617261193156, 6: 0.0055714714736493946, 8: 0.0090889321537235071, 322: 0.008001018185725078, 26: 0.011780366887778429}}, 'multiStepBestPredictions': {1: 0.005851}, 'anomalyLabel': '[]', 'anomalyScore': 0.1}
metrics=None
predictedFieldIdx=0
predictedFieldName=value
classifierInput=ClassifierInput( dataRow=0.005851
bucketIndex=1
)
)