Training Speed

Training speed is a critical factor for neural network modeling. Qnet’s training algorithms have been highly optimized for speed in an effort to minimize training time. Training times are largely determined by your problem size and processor speed. Several steps can be taken to significantly improve your execution speed and reduce convergence times. These include:

• The network design. Limit the number of hidden layers and nodes, especially during model the development phase when optimal accuracy is less important.  Also, hybrid transfer function models often converge faster than the default sigmoid setup.

• The screen update rate. Training parameters are updated in the training window at the screen update rate. A value of 1 updates the screen each training iteration, a value of 2 updates the screen every second iteration and so on. Each screen update takes CPU cycles away from the solver. While this may seem insignificant, it can slow execution by 50% or more in extreme cases. To prevent screen updating from significantly retarding execution speed, limit screen updates to once every few seconds.

• The AutoSave rate. Use rates that will yield a minute or more between stores. Disk writes are extremely slow and can significantly impede execution times. By allowing extended processing time to occur between network stores, you will limit the effect this feature has on Qnet’s performance.

• Do not use animated screen savers. These will compete with Qnet for CPU time and significantly slow execution during unattended training. For long unattended training sessions consider using the monitor's power button or setting up the monitor to go into low power mode.

• Keep Qnet the foreground task. If your version of Windows allocates priority bases on whether the task is operating in foreground or background, keep Qnet as the foreground task to maximize execution priority.

• Optimize your hardware. Make sure your system memory is adequate for the model sizes being Qnet run. Qnet's execution speed will increase proportionally with with your CPU's MHZ rating. Also, CPU's with more advanced floating point units improve execution significantly.