traning newff with standard fmin_bfgs bug

[kobb3]

Hi, I have after an extensive investigation found out that the standard traning, fmin_bfgs, sometimes stops traning prematurely and does not report back any feedback to the user. See my Overflow answear for details about the problem. Essencially the problem is this code:

import neurolab as nl
hidden = 10
output = 1
test = [[0],[0],[0],[1],[1]]
net = nl.net.newff([[0,1]], [ 10, 1])
err = net.train(test, test, epochs = 500, show=1)

which sometimes (every third time or so) stops the traning without a reason, like:

Epoch: 1; Error: 1.4999272335669662;
Epoch: 2; Error: 1.4980705557729512;

without the desired output of  "The goal of learning is reached".

//Thanks!