Automated Machine Learning

Model menu > Automated Machine Learning

The Automated Machine Learning (AutoML) feature provides the opportunity to use machine learning models and techniques without expert knowledge in machine learning. It automatically finds the network architecture and other hyperparameters.

The Automated Machine Learning window contains the following elements:

Output

Shows selectable outputs one below the other. You can choose outputs used for the AutoML run.

Inputs

Select (all) inputs that are used for the model training in each AutoML run. This button also indicates how many inputs are selected or if all inputs are selected.

Model RMSE

Shows the RMSE of the selected model from an AutoML run. The RMSE value is calculated using validation data if available, otherwise using the training data.

Results

Each button refers to the respective output.

Show AutoML results.

Clear AutoML results.

Use selected AutoML results as new model for the selected output.

(De)select All

Selects or deselects all outputs.

Select Inputs

Select inputs for all outputs simultaneously. If you want to configure this individually for each output, use the button Select of the output. See Inputs above.

Clear Selected

Clears the intermediate Automated Machine Learning results of selected outputs (checkbox). You can also use the Clear button of each output individually.

Use Selected

Uses the selected AutoML results as new model for all selected outputs (checkbox). You can also use the Use button of each output individually.

Iterations/Duration [h]

Select iterations or duration and enter a value for all/selected output(s).

Parameter Range

Opens the Parameter Range window to specify the range of hyperparameters. See, Parameter Range Automated Machine Learning (ASCMO-DYNAMIC).

Parameter Probabilities

Opens the Parameter Probabilities window to show the expected improvement for each hyperparameter. The probabilities can only be displayed after a model has been trained.

Export Job to M Script

Exports the current model settings to a MATLAB Script file (*.m) to outsource the training (e.g. in MATLAB® on a server).

Use for further options:

Export Job to Docker: Exports the single result optimization information as *.docker.ascmo file. Use the file to perform the optimization in a Docker container, e.g., in the cloud.

Configuration of Parallelization

Select the parallelization mode. You can choose if you want to use one or more ETAS ASCMO instances for automated machine learning.

Start/Continue

Starts/continues the automated machine learning with entered values.
You can stop the process by clicking the Stop button in the bottom bar.

Results

Shows the Pareto Front of found models with corresponding RMSE (model complexity versus model quality). To view its metadata, click on a model in the plot. To delete a model, right-click it and click Delete Selected Model. The complexity corresponds to the size of the model respectively the required memory. Only the memory consumption is measured, but this also means that the model is more complex to calculate.

Additionally, you can show all results outside the Pareto Front and the results of previous automated machine learning runs with the checkboxes Show all Results and Show old results. These results are indicated as blue or gray circles outside of the Pareto Front. See Non-pareto optimal and Outside Parameter Range in the legend.

When you close the window, your input is saved and the existing results are kept.