ASK java levenberg-marquardt algorithm -

i try implement levenberg algorithm in java code, met difficulties, dont know whats problems, number of mse doesnt going down. made network 1 hidden layer, 18 input's neurons, 7 hidden's neurons, , 3 output's neuron. these formula use calculate jacobian matrix:

• b2=f'(yin)
• wjk=b2*zj
• din=sum(b2*wjk)
• b1=din*f'(zin)
• vij=b1*xi

all of it, save in jacobian matriks

j[iteration][]={vij,b1,wjk,b2} 

to network error, multiply jacobian transpose find it's gradient used code:

public double err(int[]t,double[]y){     double err=0;     for(int z=0;z<out;z++){         err+=t[z]-y[z];     }return err; } 

and calculate mse of network used code:

public double mse(int[]t,double[]y){     double mse=0;     for(int z=0;z<out;z++){         mse+=math.pow(t[z]-y[z],2);     }return mse;  } 

both error , mse function called when iterate data input, , sum of mse divide sum of data input.