I have ten columns of data, and want to predict one of them using the others. First tried to point to my data, starting with the Example for Learning a Neural Network, but got error ": no entry data.bin in file" when I used the Table Reader.
When I changed to a File Reader, now get "Column 'Col36' for stratified sampling does not exist", but my data only has ten columns. I see that the Normalizer is only showing six of my data columns.
But it looks easy to build my own, so found and dragged the nodes, and built my own workspace very similar to the example (however for example I am using File reader instead of Table Reader), but now I get "only double columns for input' error in the RProp MLP Learner.
Looking in the configurations for the nodes, I see:
Stratefied Sampling dropdown in Partitioning node only shows four of my columns.
Maybe there is a different example I should be starting from, or I need to deal with these errors ?
the table reader can only be used for the KNIME internal data format *.table. To read text based files you need to use the File Reader.
Which node gives the error for Col36?
The stratified sampling can only be performed for columns with a domain attached. Columns with a lot of different values don't have a domain attached on default. You need the Domain Calculator inbetween to add the domain to the columns.
I have File Reader, Normalizer, and Partitioning. Then a learner, predictor, and scorer. Where should I place the Domain Calculator ? I tried both between File Reader and Normalizer, and between Normalizer and Partitioning, but still get the error "only double columns for input' in the RProp MLP Learner.
Every column has a type. Double stands for double precision numerical columns. You can see the type of a column in the DataTableSpec. Doubles for example are marked with a star.