ANN_GEN — General utility functions
To provide for some functions wich are usefull for a larger number of architectures. This is their general man page.
In alphabetical order, not all functions require all parameters: x Matrix containing input patterns, one per column. There shall be a unique column corespondence with the matrix of targets t and/or y. y,z Matrix containing actual output patterns, one per column. There shall be a unique column corespondence with the matrix of targets t and/or x. t Matrix containing target patterns, one per column. There shall be a unique column corespondence with the matrix of targets x and/or y. y = ann_log_activ(x) Logistic activation function. Here y and x have to be column vectors. z = ann_d_log_activ(y) Derivative of logistic activation function, expressed in terms of actual output (not in terms of total input !). E = ann_sum_of_sqr(y,t) Sum-of-squares error function. err_d = ann_d_sum_of_sqr(y,t) Derivative, with respect to network outputs of sum-of-squares error function. Here y and t have to be column vectors. [x,t] = ann_pat_shuffle(x,t) Shuffles patterns randomly for better network training, the corespondence between x and t is preserved