I would like to know how MultilayerPerceptron node treats missing values (symbol '?') found in some attributes of training set. Does it ignore them? It is not specified in any documentation...
- how does Weka Predictor node treat missing values (symbol '?') found in some attributes of test set?
Thank you for your interest!
regarding you second question: the weka predictor node actually converts the KNIME data table into a Weka instance set. If now a KNIME data cell is missing, the according attribute of the weka instance is also marked as missing (directly supported by the weka API). Hence, how these values are treated depends on the weka classifier/clusterer you use. Each weka classifier/clusterer defines its own capabilities which provide the information whether missing values can be handled by a specific classifier/clusterer (use the "Capabilities"-button in the configuration dialog of the corresponding weka classifier/clusterer node to get this information).