Randomization of real-valued matrices for assessing the significance of data mining results