Model drift controller - Classification

This component is useful to track model performance of a classification model.

The component needs as input the true label column, the prediction made by the model and the estimated probability of the reference class to occur.
Finally, the component controls for predictors' shape and a measure of distrubutional divergence is computed. An high value for a significant number of predictors could cause a model drift, potentially leading to a downgrade of the predictive power.

The following columns must be correctly idetified in the configuration dialog:
-Target varable
-Target prediction
-Probability column
-List of predictors used

Options

Columns on which compute the Kullback-Leibler Divergenge

The list of predictors used by the model to produce estimates.%%00010Continous predictors are binned into classes to make computation easier.

Probability column

Point out the probability column

Target prediction column

Point out the prediction column

Select the target variable and the corresponding target class

Select the target class value and the true values column

Nodes

• Constant Value Column17 ×
• GroupBy16 ×
• Math Formula13 ×
• Table Row to Variable9 ×
• Column Filter8 ×
• Show all 44 nodes

Extensions

• FeatureKNIME Base nodes
• FeatureKNIME JavaScript Views
• FeatureKNIME JavaScript Views (Labs)
• FeatureKNIME Javasnippet
• FeatureKNIME Math Expression (JEP)
• Show all 6 modules

Links

• No links available