Hi all,
My project is to schedule job shop using neural network. I got problems that I solved with a heuristic that I want my netywork to learn.
*I used MLP 12(input)-6(hidden units)-6(output)
*Levenberg Marquadrt
*conjugate
*momentum 0.7 ,0.8
*I tried with random initial weights , I changed the initial weights like mean=0, variance 0.5
mean = 0 ,variance 0.1
mean = 0, variance =1
*I tried learning rates 0.1,0.2,0.5,1 ,0.01,0.02
*weight decay but gives me bad results
with all this the correlation r didn't come better than 0.5
the MSE didn't come better than 0.27
the learning curve becomes flat (I trapped in a local minima) and when this happens I jog the weights or randomized it but it returns again to the same level.
What to do? please, help.
Celebrate Earth Day everyday! Discover 10 things you can do to help slow climate change. Yahoo! Earth Day