In the recent scripts I used Matlab’s rand() function to seed the random number generator. I aimed at generating reproducible results. However, if the data can be modeled with an MLP, then it should most of the time converge towards the same solution, given that the initialization is not too different (weights from -1 to 1).

I verified this on the agriculture data with an unseeded rand(). Basically, I ran the same script that generated the error-mean-stddev plots without seeding. I checked the graphs to see whether the established network configuration (n-8-1) is still justified or if it was just an outlier that led to choosing eight neurons in the hidden layer. As it turns out, the choice of eight is justified after examining the following five graphs.