Skip to main content

Monitor Drift

Synopsis​

Calculates a summary of input data and submits them for model monitoring for drift calculations.

Description​

This operator takes a data set as input and calculates an aggregated summary of all input columns. This summary can then be sent to Model Ops to a model with the specified Id. You need to first register a new model in Model Ops which will provide the required Id to you. The calculated input statistics can either be sent to Model Ops for the training data or for scoring data. If you provide both, Model Ops will be able to calculate input drifts for your models. Additionally, you can also provide a set of column weights which allows Model Ops to display drift vs. importance for all input columns.

Input​

data​

This input port expects a data set with the regular attributes for which drift should be calculated.

weights​

Optional attribute weights which allow to show drift vs. importance. Will override existing weights if provided.

Output​

drift summary​

This object summarizes the input drift information for monitoring purposes.

data​

The input data.

weights​

The input weights.

Parameters​

model id​

The model id under which this measurement should be stored.

mode​

The mode of this drift monitoring operator: updates the drifts either for training or for scoring data.

clear before update​

If this is the true, all historic statistics about input data drift will be deleted as part of this update. ATTENTION: This can not be reversed.