I have some data that I want to use to develop a model, but the generated data depends on a couple parameters (moving average length). Is there a parameter optimization algorithm(s) that can be used in KNIME to final my optimal parameters. This exists in RapidMiner:
That reads as if this optimization scheme just iterates a set of parameters, learns a model and records the accuracy (not a hill climbing approach, where the accuracy response controls the next set of parameters?).
In KNIME you would use a “Variables Loop (Data)” meta-node to do that (enabled in the expert mode). You would define a table with the parameters to iterate (each row is one loop iteration, each column a parameter to set) and then use your preferred learning algorithm in the meta-node itself. Use the accuracy scorer node to determine the quality and collect it in the loop end node.
If you wanted to implement the hill-climbing approach, you would need to write a new loop end node, which determines the next set of parameters and feeds it back to the start node. Sounds complicated but should be simply as all the learning/prediction and accuracy calculation is already available (look at the LoopEndContitionNodeModel – another nice example of a loop, where you stop iterating a set of parameters if some condition is met).
Hope this helps, Bernd