Recently,I want to establish a neural networks to forecast runoff. But now I tested a example as folloes,but I don't understand the results that you can see from the pictures.
You need to be looking at MSE, during training that is your measure of how well it is progressing. Basically it should drop during training. It will reach a plateau at which point you have the best network for that training run. You are, of course, trying to minimise MSE (maximise classification accuracy). Early experiments can establish a reasonable estimate of the number of epochs to train for (and suitable values for learning rat eand momentum). After that the fun starts, as you experiment with input pre-processing, choice of transfer function and network architecture!
- Badges
- Science topic