The properties in the Attribution Modeler | partitions | partition[n] | AdvancedOptions category specify values that affect how
Attribution Modeler evaluates data, and that apply across all partitions.
The sampleSize property defines the percentage of available records that are used for training. This value needs to be set to a number that is greater than
0 but less than
100 (percent).
The randomSeed property represents the starting point that
Attribution Modeler uses to select records randomly.
The maxTrainingTime property specifies the maximum time, in minutes, that
Attribution Modeler spends training itself. It sets a time limit on the training process as it iterates over the data to reach the goal that is set by the
converganceThreshold property. This time limit helps you limit the resources that
Attribution Modeler uses. A warning is logged in the log file if SIRA exceeds this training time limit.
The convergenceThreshold property is used to set a limit for how much difference is allowed between the results of one training iteration and the next. This difference is expressed as a percentage of responses for which the results (winning offers) are allowed to change from one iteration to the next.
If you set the property to 0 (zero), then you are not allowing any changes to the results from one training iteration to the next; this is the most rigorous standard. If you set this property to a value higher than
0, then you are allowing the training results to be more flexible; the standard is less rigorous and training might be completed sooner.
The noiseEliminationThreshold property is used to eliminate the SIRA credits below the threshold value.
Copyright IBM Corporation 2012. All Rights Reserved.
|