This operator learns the mean and the standard deviation of a single value and writes it into the output. On port 0, the learned values (mean and standard deviation) are put out, on port 1 the tuples which are used to learn the deviation. The operator puts out the original tuples with the groupId the operator chose for the tuple. This is necessary for the following operator to be able to map the deviation values from port 0 to the correct tuples on port 1.
In this example, the online-algorithm is used to learn the deviation. Therefore, no window is needed. The deviation is learned for the parameter "temp". The following operator uses the original tuples with the group of the DeviationLearn operator (1:deviationLearner) and the learned deviation (0:deviationLearner).
#PARSER PQL #ADDQUERY /// Use the online training mode to learn the mean and the standard deviation deviationLearner = DEVIATIONLEARN({ trainingmode = 'ONLINE', attribute = 'temp' }, System.manual ) /// Compare the current tuple with the learned values deviationAnalysis = DEVIATIONANOMALYDETECTION({ interval = 3.0, attribute = 'temp' }, 1:deviationLearner, 0:deviationLearner ) |
The operator can save the learned values info a database. The learned information is saved, not the tuples which learned the operator. When the query starts, the learned information is read from the database and written into the operator. This only happens once. The backup happens every time the operator updates the values. The PQL code belows shows, how to use this option.
#PARSER PQL #DEFINE BACKUPSCHEMA [['sum1', 'Double'],['sum2', 'Double'],['sumWindowSqr', 'Double'],['sumWindow', 'Double'],['m2', 'Double'],['mean', 'Double'],['backupId', 'String'],['k', 'Double'],['standardDeviation', 'Double'],['n', 'Double'],['group', 'Double'],['backupId', 'String']] #ADDQUERY /// Read backup data from the database backupMongo = MONGODBSOURCE({ database = 'odysseus', port = 27017, host = 'localhost', collectionname = 'condition' } ) /// Convert backup data to tuples backupTuple = KEYVALUETOTUPLE({ schema=${BACKUPSCHEMA}, TYPE = 'Backup', KEEPINPUT = 'false' }, backupMongo ) /// Run DeviationLearn and use backup data, if it exists ("Recovery") deviationLearner = DEVIATIONLEARN({ trainingmode = 'ONLINE', attribute = 'vibration', uniquebackupid = 'dev1' }, System.fridgeVibration, backupTuple ) /// Convert backup data from port 2 to a key value object keyValueOp = TUPLETOKEYVALUE({ type='KEYVALUEOBJECT' }, 2:deviationLearner ) /// Save the backup data to the database mongoSink = MONGODBSINK({ database = 'odysseus', port = 27017, host = 'localhost', collectionname = 'condition', batchsize = 1, deletebeforeinsert = 'true', deleteequalattribute = 'backupId' }, keyValueOp ) |