Backpropagation - Stopping Training

Top 

Backpropagation - Stopping Training

 

To stop training, check the appropriate box and specify a number that corresponds to the stop training criteria you selected for the test set and/or the training set.  If more than one is selected, the network will stop training as soon as any one of the criteria is reached:

 

Training Set Criteria - Computed at the end of each epoch

Average error below a predefined level.

Epochs since min. average error exceed a specified number.

Largest error below a predefined level.

Learning epochs exceed a predefined number.

           

Test Set Criteria - Computed at the end of the Calibration interval # of events

Average error below a predefined level.

Number of events since the minimum average error exceed a specified number.

Largest error below a predefined level.

 

Refer to Backpropagation - Calibration for more information on how these should be set.