Each time I optimize the neural network I get slightly different results without changing any settings. Why?
Neural networks are not deterministic systems, but are initiated from a random initial state and are trained from that state to make conclusions about the training sequence( price history). It is possible to make a deterministic system out of a neural network, but it is not advisable. This is because the training sometimes fails and you would always get the same results when training the network again if it was initiated from the same initial state. If you select Model Settings/ANN and check Deterministic Training, the network will always be trained from the same initial state and you will always get the same results. It is possible to obtain more consistent results at a cost of longer processing time by adjusting the following parameters: 1. Increase the Number of Networks – Because every single neural network produces slightly different results, the results of several networks are combined to produce a more stable result. The value of this parameter sets the number of networks used f
Related Questions
- The date and time formats associated with "Customer Info" are different than the current system settings. Do you wish to use the system settings for data entry?
- I use my wireless notebook on multiple networks. How can I connect to them without having to keep changing my network settings each time I switch networks?
- Each time I optimize the neural network I get slightly different results without changing any settings. Why?
