Learn Rate Coefficient

The learn rate, eta, controls the rate at which Qnet’s training algorithms attempt to learn. This factor determines the size of the node weights adjustments during training. Eta’s valid range is between 0.0 and 1.0. While a higher eta results in faster learning, it can also lead to training instabilities and divergence. A small, infinitesimal eta will offer improved numerical convergence, however, training time is greatly increased. When initiating training on a new network, the user must provide a starting eta value. It is better to start conservatively by using a low number. Using a value in the 0.001 to 0.1 range will normally start the training process safely. If the initial guess is too high and the network diverges, reset the network by selecting the "Options/ Randomize Weights" from the training window menu. Select to try a lower value. Qnet’s Learn Rate Control (LRC) can help keep eta in its optimal range during training when active.

The learn rate is set in the Training Setup/Training Parameters dialog and during training.