AutoTrain Mode

Qnet's AutoTrain mode allows the user to train multiple network models in a continuous, unattended mode. This option is designed to allow users to easily maintain and/or retrain a large number of models. AutoTrain mode sequentially trains the selected network models and saves the run histories (netname.ler) and AutoSave information (netname.lod) for later processing and analysis. To initiate an AutoTrain run, simply select multiple files in the file open dialog at Qnet startup (hold down the "ctrl" key or "shift" key to make multiple selections). All Qnet network files that will be used in the AutoTrain run should be placed in the same folder so that the selection of multiple files is possible.

After selecting multiple network training files from the file open dialog, the AutoTrain Options Dialog appears with the options to run an AutoTrain session or process and optimize a previous AutoTrain session (AutoLoad). If a new training session is selected, Qnet must be told which method it should use to save the network at the end of each run. You may save the final iteration (default), save the min RMS test error iteration, or save the max test set correlation iteration. The last two options require the use of a test set.

With an AutoTrain processing/optimization session (AutoLoad session), one can perform a full analysis of the trained network and save the network to a state (iteration) that is different from that selected during training. Use the File/Save AutoSave Network... option to select a new save iteration. Note: the save iteration selected by Qnet during training will appear in the Network Definition group title field of the training dialog (if saved with the min RMS or max correlation option).

The above describes how to launch AutoTrain runs interactively. It is also possible to use the command line to automate AutoTrain runs. This is helpful when Qnet is being launched by another application and/or for scheduling or automating the training process.