I would like to compare the quality of predictions performed by the Knime (native) MultilayerPerceptron and the Weka (3.7) version of it.
To test this, I have created a simple dataset containing 100 values of a sine function such that there are 3 periods in this dataset. For both the Knime and the Weka multilayerperceptron, I set the number of hidden layers to 4 and the number of nodes per layer to 20. The dataset is such that all values (x and f[x]) are in the region [0,1], therefore no additional normalisation is needed.
For the Knime node, I get a very nice R^2 of (almost) 1. The Weka node gives very bad results independent of the settings. Does anyone have an idea what could be the issue? Could I overlook something crucial?