The prediction capabilities of the neural network that was coded in the last post do not seem to be as good as expected, at least not in the standard configuration. When I fed the data set (which I will not publish here) through the network and the cross validation, the results are as follows:


The legend is clear: target values and predicted values are shown at the top of the chart and the calculated error is at the bottom. The scale for the competition results is in LEN points which range from 1 to 1000, where 1000 is the best result (aka world record) — thus, the points for achievements change when a new world record is set. The error in the above plot is by far too high and I will generate a thorough evaluation with different network parameters. I should be able to tell which parameters have to be set and how.

In the paper that I linked to in this entry, the average error was at least an order of magnitude smaller than the one I obtained. There still is quite a way to go. I assume that the data set is suitable for this model and that it is just the configuration of the neural network that needs tuning.