Hi,
I was wondering if KNIME has the ability to take the predicted results and append them to a hold out set without the predicted column in that data set.
SPSS modeler has this feature where it appends a proability score to data point.
Video example
https://youtu.be/_VbQer7k6ns?t=19m1s
Hi,
in KNIME we have for most of our models a Learner and a Predictor.
The Learner generates the model. This model is fed into the predictor together with the test data (hold out set) and predicts the class. The predictor does not need/use the class information
IS there a special way to do this?
I tried removing the predicted column from the hold out set and it wont run unless the fields are there.
Keep in mind I am using a training using 60/40 test/train and am using a validation set (hold out) that the model wasn't trained on.
Ideally id like to take this set and attach a probability and or flag for each predicted data point.
I get this error message when I try to run it.
WARN Scorer Column TARGET_ATTRIBUTE not found.
Hi Data_Analyst,
of course the Scorer does need the original target_attribute for scoring. Otherwise there is nothing to compare to. However, the predictor do not use them.
Did you take a look into our example server? There are some basic examples in the 002_DataMining Category. Or maybe you post a screenshot of your workflow?
Best, Iris
Thank you for your help.
I tried doing it again and it works. Thank you so much kind sir!
I am used to just running the whole stream from the Scorer or ROC Curve node ..oops!