An empirical comparison of the performance so single-layer algorithms for training threshold logic units