The new DL4J learner on 3.3 does not recognize images as the input. I keep getting "No input columns selected" and don't have them as an option in the feature column selection.
Things run fine on 3.2.1 with exception of when I enable GPU. Then I get an error "Could not initialize class org.nd4j.linalg.factory.Nd4j". I've tried with both CUDA 7.5 and CUDA 8.0 installed.
are you working with the nightly build? Did you install the KNIME Image Processing - DL4J Extension, too? See: https://tech.knime.org/deeplearning4j-imageprocessing for more infos about this.
sorry fou the trouble. Could you give me some more information about the problem? Which OS are you running? Which kind of GPU are you using? What's the output of ncvv --V?
Also, could you maybe attach the KNIME log file here?
thank you very much for the information. It would be great if you additionally could provide me with the full knime.log file and attach it to this thread. It is located in your workspace folder under '\.metadata\knime'.
I am trying to get any of the 04_Analytics 14_Deep_Learning examples to work.
Through "Install KNIME Extensions", I have installed everything found with "deeplearning4J".
But when I try to execute 02_Basic_Learner_View Tutorial, I get an error "ERROR DL4J Feedforward Learner (Classification) 3:14 Execute failed: Could not initialize class org.nd4j.linalg.factory.Nd4j"
What can I do?
The knime log file is not attached as it is. I have deleted the log from the earlier days.
thanks for the log file. Are you also trying to run on GPU? Furthermore, what Windows version are you running on? You could try to run KNIME in administrator mode because it seems some dll can't be accessed on your machine.
now, after I increased Off-Memory to 4000MB, I am getting the error
ERROR Word Vector Model Reader 0:70:142 Execute failed: Cannot allocate new IntPointer(8): totalBytes = 3G, physicalBytes = 4G
Also, if I am loading the german.model https://tubcloud.tu-berlin.de/s/dc4f9d207bcaf4d4fae99ab3fbb1af16/download from https://devmount.github.io/GermanWordEmbeddings/ the "Word Vector Model Reader" Node takes a long time to load. Is that normal? It is loading for a couple of minutes now and isn't finished yet.
I increased to 8000MB, but the german.model still gives the same error. The google model is loading for 40 Minutes now. I am stopping the process. I think there is a bug somewhere.
the DL4J library only supports a limited number of formats which currently include in KNIME: the models saved by the Word Vector Writer node, models in plain text format and some binary models. I don't know the exact format of the model you want to read so it could be that it is not supported. However, the google model should work. Unfortunately, it can take a long time to read because it is so big (maybe even an hour).