Github user ijokarumawak commented on a diff in the pull request: https://github.com/apache/nifi/pull/2686#discussion_r193654571 --- Diff: nifi-nar-bundles/nifi-deeplearning4j-bundle/nifi-deeplearning4j-processors/src/test/resources/classification_test.txt --- @@ -0,0 +1,100 @@ +1.1,0.5,0.5,0.2,0 --- End diff -- It seems this text and test are based on the [DL4J CSVExample](https://github.com/deeplearning4j/dl4j-examples/blob/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/dataexamples/CSVExample.java). IMHO, instead of mock classification, using the [iris.txt](https://github.com/deeplearning4j/dl4j-examples/blob/master/dl4j-examples/src/main/resources/iris.txt) and the same configuration with CSVExample to train the test model would be more understandable, especially for those new to Deep Learning and Neural Network topic. Because the iris.txt dataset has more background information what the data is, and what classes the model is supposed to predicate. Current mock classification can not imagine such actual objective and can be hard to understand what it does. And having the same number (4) for different things, such as inputNumber, outputNumber, numClasses makes even harder to grasp what those numbers are. How do you think?
---