How I do a ROC curve of two decision tree?

KNIME Analytics Platform
1

Hi,

Knime is great. I have been trying to make a Roc curve of several models (decision tree, regression, etc.), but I can’t, I don’t know how to… Could Anybody help me? Thanks a lot.

Best

2

There is a ROC curve node in the standard distribution. However, it needs class probabilities and not all predictors output them. But if they do there is an option in the dialog to switch them on.

3

hi dear thor,
Thanks, I tried it, in dialog box I activated “Append columns with normalize class distribution”, run it, and the console shows this error messages

WARN DecisionTreeNodeSplitPMML Decision Tree Prediction failed. Could not find branch for row 'Row2376: (Es dueño de una Empresa., 62, Masculino, 54, Si, SLP, MÉXICO, ?, ?, ?, X, ?, Usted produce algo., 11 - 50, ?, ?, ?, ?, ?, ?, ?, ?, X, ?, INDUSTRIA, PEQUEÑA, No, PR, ?, ?, X, ?, ?, ?, ?, X, ?, ?, ?, ?, ?, X, ?, ?, 1, PREREGISTRO, 0, 51 - 60, Rango31a90)
ERROR Decision Tree Predictor Execute failed: (“NullPointerException”): null

Now, it is the problem. :).

Best,

4

This error has nothing to do with the “Append columns with normalize class distribution” option. It will most likely also occur if you run the predictor runs without it. If you provide me to your data and flow, I can investigate the problem in detail.

5

Hi,

1) to compare two classifiers with ROC Curve node, e.g. output of Multilayer Perceptron Predictor and output of Weka Predictor. How can I do? I must joining the two output tables?


2) Starting only from Weka Predictor: in ROC Curve configuration I set "class column" with my nominal class that is classified as YES or NO, then I set "Positive class value"= YES and then "Include columns"= YES and NO columns. Is it ok?