2021.03.15 (Mon)

By setting AutoTune during model training, hyperparameters are automatically explored and optimized.

Optimization methods supported by VARISTA

The following methods are supported by VARISTA.

  • Grid Search
  • Radamized Search
  • Hyperopt
  • Optuna

Configure AutoTune

Modeling › Templates (tab) › Any Template or New Template › Model Settings section

VARISTA Document AutoTune 001
Set the parameter setting item AutoTune as desired.
(In this example, TPE Optimization with Optuna is selected.)

VARISTA Document AutoTune 002

Set a range for any hyperparameter you wish to explore.
The format for input is as follows.

  • The range is separated by commas (e.g. 0.000, 1.0)
    It is also possible to enter the range in E notation (e.g. 1e-8,1.0).
  • If you want to fix the value during the search, enter an arbitrary number.
  • Leave blank for default value

When you have completed the settings, save the template under an arbitrary name.

Training with AutoTune

Modeling › Models (tab) › New Model

VARISTA Document AutoTune 003
Start creating a new model using the template you have just created.

Check the optimization history

Select a model that has completed training to view the model report.
To check the optimization history, do the following
Algorithm › Select any algorithm › AutoTune (tab)

The best trial rounds will be displayed in green text color.

Made with
by VARISTA Team.
© COLLESTA, Inc. 2021. All rights reserved.