RProp MLP Learner Domain range for regression

Hi,

 

I have this simple workflow:

XLS reader -----> Normalizer -----> RProp MLP Learner

XLS reader and Normalizer run OK but when I connect the MLP node, it gets the following:

ARN      RProp MLP Learner     Input data not normalized. Please consider using the Normalizer Node first.

I had used the normalizer first!


WARN      RProp MLP Learner     Domain range for regression in column RESPIRAT not in range [0,1]

I didn't use a regression, but a MLP. What does this mean and how can I resolve it?

 

Thanks in advance,

Marta

Hi Marta,

Did you select all columns for normalization that you later use in the RProp node? Can you please also check the output data on the Normalizer node, if it is in the [0,1] range; right-click the node and select the second last item of the context menu. In the DataTableSpec tab, the lower and upper bounds' values should be 0 and 1. If not, can you please check the settings of the Normalizer and post them here. Thanks.

Regards, Thomas

Did you select all columns for normalization that you later use in the RProp node?

Yes.

Can you please also check the output data on the Normalizer node, if it is in the [0,1] range;

The values were out of that range. I had used tha z-score normalization. I changed to the Min-Max with the range 0.0 to 1.0 and it's fine now.

Must I connect this node to the Normalizer (apply)? Is it mandatory?

Best regards,

Marta

Hi Marta, The Normalizer Apply node can optionally be connected to the second out-port of the Normalizer when you want to normalize a second (test) dataset with the same normalization method. The advantage is that you don't need to readjust the normalization method in both nodes. Regards, Thomas

Thank you Thomas.

Best regards,

Marta

After I applied the MLP learner I generated PMML code with the PMML writer node.

Later, I used the PMML reader with the generated code and applied a MLP predictor and I get this:

ERROR     MultiLayerPerceptron Predictor     Execute failed: org.knime.core.data.def.StringCell cannot be cast to org.knime.core.data.DoubleValue
 

What might be the problem?

Thanks,

Marta

Hello Gabriel,

I am new at KNIME and I am trying to use it to predict the market price of lands used as warrants in credit operations. That said, I have a data set with the respective lands and their attributes as numbers. Since my output is not nominal, I will have to use a regression, and Neural Networks are great for that purpose. However, I have never used the Rprop model before. Do I have to use a X-Partitioner and X-Aggregator? Since I have to normalize by set before applying the model, how can I denormalize it (only one column)? Thanks in advance!

All the best, Pedro

 Hi Marta, 

did you solve the problem?

I have exactly the same issue :(

Thank you in advance

Hey Amy,

Which of the issues are you referring to? Maybe I can help you with that.

Best,
Ferry