Parameter Optimization (Table) Component on MLP

This workflow shows an example for the "Parameter Optimization (Table)" component (kni.me/c/dIpKMJbiO-3019eb).

The model used for parameter optimization in this case is Random Forest. The Learner and Predictor nodes are captured with Capture Workflow nodes, exported in the black Workflow Object Port and adopted in the component via a Workflow Executor node. Thus we can use this component with any classification model without making any changes to the component.

A Table Creator is used to pass the parameter range (minimum, maximum, and step size) to be used for optimization. A Variable Creator is used to send the initial set of parameters to the capture node.
The output of the component is a flow variable with the best of parameters. This output flow variable automatically configures another Learner node to train the final model.

STEPS TO FOLLOW TO ADAPT WORKFLOW ON YOUR OWN CLASSIFICATION MODEL:

1. Import your training data with a Reader node
2. Replace the Learner and Predictor nodes with the desired ones with the Capture nodes.
3. Define suitable parameters in the Variable Creator nodes with precise names (they will display in interactive view).
4. Define in Table Creator one row for each parameter. 5 columns: Name, Datatype, Start, Stop, Step Size.
5. Name of the parameter in Table Creator should match the one stated in the Variable Creator node
6. Define parameters ranges, with start, stop and step size for each parameter with either "Number (double)" or Number "(integer)" datatype.
7. Configure Learner node within Capture to use the flow variables for Variable Creator node (Flow Variable panel).
8. Configure the component with the required options.
9. Execute the component on the training set and "Open Interactive View" to inspect the results.
10. Train the model with the best parameter combination with another Learner node, using the flow variable output from the component.
11. Test the model on the test set with another Predictor node

External resources

• KNIME Verified Components - knime.com
• ML Algorithms and the Art of Parameter Selection - KNIME Blog