Training Options - adam-nnl/Erudite GitHub Wiki

Quick Overview of the different training options available with Erudite

Training Threshold:

  • Epochs- Enable this option to select a set number of epochs to train a neural network.

  • Average Error- Enable this option to train until the average error is less that the number entered.

  • Training Examples Correct Percentage- Enable this option to train until the percentage of correctly categorized training examples exceeds the number entered.

Training Options:

  • Adaptable learning rate- Enable this option for an adapting learning rate. The learning rate is increased based on the neurode error gradient and outgoing synapses.

  • Randomize training set- Enable this option to have the training set start at a different example each epoch.

  • Auto Save- Enabled this option to automatically have the neural net saved back to ANNeML XML after training completes.