neuronal network


I have few questions about neuronal networks in KNIME:

1.) Exist limits in terms of amount of datas in the input- and output-layers?

2.) Is it possible to enter a 4-figure number in the "class"-column?

Thanks for response.

Best regards


Hi Oliver,

As far as I know the limits are quite large, if you have less than 2^31 rows (and a lot of memory/time when you have lot of data) you will be safe.

I am not sure I understand your second question... Any string can be in the class column even numbers. You might need to make it a nominal column (compute its domain with Domain Calculator node). But if you are asking whether the NN nodes can be used for clustering not just classification, I do not know any options to do that.

(Which NN implementation were you referring to?)

Kind Regards, gabor

i want to use PNN (probabilistic neural network) for classification in KNIME. I have 59 feature or variable. But only 15 of them are continues and the others are categorical.  Can i use categorical variable in PNN? If not, what should i do?

Thanks for your help

The PNN Learner works only on numeric columns and ignores categorical attributes. You can use the One2Many node to convert those columns into numeric (bitvector) columns and use them as input for the learner.

hi gabriel

thanky you for your reply about PNN. I have anothre question: for using PNN, shoule i normalized variables? or PNN node have the algorithm for normalization? if i should normalyzed the variables, which node is apropriate?

best regard

The so-called Normailzer node has different options for normalizing numeric data. However, if you have used to One2Many node you will end up with bits (0, 1) which don't need to be normalized anymore...

hi gabriel

thank you very much for your reply. i have another question . i understand that with using one 2 many node there is not need for normalizing categorical data. should i use normalizer for numeric data for PNN? 

thank you very much





I have a question about neuronal network:

Is it possible to affect the internal structure of the network?

Thanks for your response.

Best regards,


For the PNN, the data does not need to be normalized. On the other hand, the Rprop (MLP) should have the data normalized; within the configuration you can adjust the number of hidden neurons; the number of input and output neurons is defined by the data (input dimensions and number of classes). I also suggest looking into the Weka extension for KNIME for more like state-of-the-art algorithms.

hi thomas

i want to implement PNN in KNIME. for nominal variable i used one 2 many node. in new column

missing values appear as (0) in new columns. will this create problem?

 i have another question. for performance measurement in scorer node , which columns should i use? do you have work floe example of PNN ?