I’m trying to use TMultiLayerPerceptron, but when I ask it to train, it merely says that it fails the line search, and gives nans for the weights of the nodes and the traing and test scores. I suspect there is a problem with the data I’m feeding it, as I have changed one of the input nodes to be completely correlated with the output node (the same value) and th line search still fails.
Is there any call I can make which will test the validity of the data going in to the TMLP? Is there any other cause of the 'nan' phenomenon? I tell it to run for 1000 epochs, but get no output during the training, but then again, I don't know what I should expect to see. I have asked it to draw the output as text and graphical, and as far as I know this is not happening.