Automated Machine Learning

Models for automated Machine Learning are composed of 2 types of parameters:

  • Hyperparameter

  • Model parameter

Hyperparameters are all parameters which can be set individually by the user before the model training starts (e.g. number of layers or learning rate). Model parameters are instead learned during the model training (e. g. weights in neural networks). Finding good machine learning models is a type of an optimization problem. With a series of hyperparameters it is targeted to find the right combination of the values that provides the best model delivers.

The different modeling methods RNN and NARX have different hyperparameters.

RNN, e.g.:

  • Cell type

  • No. of layers

  • Memory size

  • Optimizer settings

NARX, e.g.:

  • No. of basis functions

  • Used features

Define a tree structure how the algorithm finds a good model. See the following tree structure for details of the differences.

These settings can be made in Model Configurations: Recurrent Neural Network (RNN) or Model Configurations: NARX Structure(Model > Properties). If you wantASCMO-DYNAMIC to find the best parameters automatically without setting them youself, use the Automated Machine Learning feature (Model > Automated Machine Learning), see also Automated Machine Learning.