Upper left-hand honeycomb: Quality

The Upper left-hand honeycomb shows what quality the neural net has just reached on your computer. The honey-coloured curve runs continually from left to right and displays the current quality development. The white curve also saves the past and always keeps new data coming from the right, this is the learning curve. For the white curve, the axis below is therefore defined as follows: at the far right the point in time 0 (beginning of computing period), to the far right is now. The time axis therefore comprises the entire training period.

Occasionally, one can see how the net remains "helpless" before the quality starts to climb rapidly (the proverbial penny has dropped) and the net starts to refine its results slowly. If the learning curve flattens out at a higher level, the computing cycle is interrupted - one cannot count on any better results after this phase. If the quality is very high, the actual and the computed stock prices that the neural net has processed must be very congruent as is recognisable in the centre honeycomb.

The quality shown in the screensaver is computed in a different manner to the forecast probability next to the forecasts on the Internet Information Page. In principle, the screensaver always computes the distance (the smaller the better) between the forecast curve and the actual curve (which always refers to the past). This, in turn, displays the quality of the neural net and is simple to depict graphically. The exact formula for this computation can be found on our Explanation Page of the Exact Computation.

The website, on the other hand, shows whether the forecast applies in the boundaries of a previously delineated variation or not. This method is better suited for measuring the quality of the forecast and is user friendly: many radio transmitters use the same method to predict rainfall probability in percentages.