Loading Lagun...

LAGUN

LAGUN is a R/Shiny platform providing a user-friendly interface to methods and algorithms dedicated to the exploration and the analysis of datasets.
Guided workflows are provided to help non-expert users to apply safely the proposed methodologies.



These tools are commonly used in the numerical uncertainty community (GDR Mascot-Num for example) but are also widely applicable to experimental problems. The main functionalities are the following:

  1. Optimized design of experiments
    If you have control on the inputs/parameters on the system which will generate the dataset (numerical simulations, settings of the experiments, …), you can benefit for a better spatial repartition of the experiments.

  2. Visual exploration tools
    When the complete dataset with inputs/parameters and outputs/responses is available, you can load it to perform insightful visual analyses and identify the main trends and the most influential parameters.

  3. Going further with surrogate models
    A next common step is to use the dataset to infer a predictive relationship between the inputs/parameters and the outputs. This estimated relationship, the surrogate model, can help push forward the analysis with its ability to predict the responses for any new combination of the inputs. In particular it can be extensively used for uncertainty quantification, sensitivity analysis, deterministic optimization, optimization under uncertainty (robust and reliability based) or more intensive graphical studies.

  4. Numerical simulations
    In the special case of numerical simulations, you can benefit from a direct connection between LAGUN and your simulation scripts to perform automatic and sequential optimizations with the surrogate models.



For a general introduction, please see the following tutorial:


History

The first version of LAGUN was initiated at Safran Tech (under a different name), the corporate research center of Safran. Its goal was to give an easy access to methods and algorithms to all Safran engineers with a user-friendly interface. A collaboration was later launched with IFPEN in 2019 to share algorithms and developments in a common platform now named LAGUN ("Assistance" in Basque language).

The platform is organized in tabs, each one of them corresponding to a step above.

Click to expand the panels below to learn more through tutorials and test cases (links to Gitlab).

In each tab, the primary tasks to be performed in the workflow will be indicated by blue buttons:

Tasks that are not explictly in the workflow but are highly recommended will appear in red:

Tasks that may bring relevant additional information are colored in lightblue:

Finally, information on what is loaded in memory and which surrogate model is active is available at all times in the 'Info' panel (found in the navigation bar).









Advanced Settings

You will find here an access to some parameters that are used internally in the platform for some of the tasks. You are free to change them, however it is recommended that you are familiar with the concepts. Indeed increasing one or some of these parameters can have a large impact on the computation time of the tasks.

About

Information on the releases and versions




Please contact saf.lagun@safrangroup.com for information.


Warning: Updating inputs resets DOE.

Current Inputs













Switching to expert mode enables more complex surrogate combinations to handle difficult problems.
It is highly recommended to try first this linear model. For some outputs it may be predictive enough and for the others you can check how nonlinear the output may be.


Click on a cell in the Q2 table below to display the prediction quality for each output and each metamodel.

Here you can choose the criterion used to automatically assign the best surrogate model to each output.
If you make a manual change in the list of selected metamodels, it will not be saved unless you click on the save button below.



If you change cutoffs, you need to reactive display to update the table

Loading...


Predict



Export Predictions

Import Uncertainty Parameters




If you do not import a file (or the imported file is not valid), by default each input will follow a uniform distribution on the domain given by the imported bounds (for continuous ones) or the levels (for categorical ones).

Current Inputs



Export UQ propagation



It is recommended to choose a sample size at least equal to 100/p if the probability to be estimated is of the order p. If you do not have an estimate of the probability, you can use the UQ propagation above as a rough estimate.



Refine Surrogate Model Near Threshold



Export Additional Simulations

For mono-objective optimization, the local BFGS algorithm is used.
For multi-objective optimization, the problem is solved with the genetic algorithm NSGAII.

Current Inputs



Export Optimization Results



Current Inputs



Please select what to display & export if no optimum respecting all constraints was found.

Export Optimization Results


Current Inputs













Estimation of Problem Complexity

(% of points in the domain which satisfy the constraints with a given probability)





Export Optimization Results

This is a beta version of Lagun Computer Code Exploration Platform.

Version history

Version 0.9.10 (December 2020)

  • First release

Please contact saf.lagun@safrangroup.com for suggestions, comments or bugs.