how is the node cross validation work?

I have some question about the cross validation node, did it reset the model in every loop or not, or just train the model based in the result in last loop? in other word , is the model in each loop independent?
thank you!

Hi @674153814 and welcome to the Forum

As far as I know (from experience) is each loop independent. Every loop (fold) has always the same settings in the Learner. There is no data going back from the Predictor to the X-partioner node. All the results are collected in the X-aggregator node, so the scores per fold can be compared.
gr. Hans


Related with @HansS 's response, the cross-validation node makes a loop using X-partitioner and X-aggregator. In each iteration, the models are reset and trained with other folds of the X-partitioner node. However, it is possible to configure the loops to modify the variables of other nodes using the variable ports (I did not do that before but I know that it is possible). Once all the iterations have finished you have the result of all the nodes. It is that you have the prediction of all the instances of your database. As @HansS told you you can select in the X-aggregator node the options to add a new column with the number of the fold that did the predictions of these data.

I hope I helped you in addition of @HansS 's answer


This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.