Sample Design Space

Modified on Fri, 15 Sep 2023 at 01:40 PM

Description

Based the ranges of an existing dataset or defined by the user, this function creates a sampling of the design space using various sampling methods and granularities.


Application

You can create a new sampling of your dataset based on the min- and max-values of the columns in a dataset or based on user-defined ranges . Latin hypercube, random and grid methods are used to create the sampling of the dataset.You can use this manipulator for example to create a DoE plan based on the range of values of the input columns in your dataset. Alternatively, you could create the parameters of your design space and their respective ranges from scratch.


How to use

  • First of all, you will have to decide if you want to define the design space from scratch or if you prefer to Use a dataset as a starting point. Tick or untick the first option depending on your preference.

    ⚠️ When this tickbox is updated, all existing parameters below in the step will be deleted.

    • If the option is ticked, you will be asked to select the Data set that you want to use as the starting point.
  • You can then add parameters to your design space by clicking on Add parameter. Depending on your option for the starting point, the parameter will be defined in a different way:
    • If you selected a dataset as a starting point, you will be able to select a column from that dataset. Once the column is selected, the range (or fixed value) is automatically updated based on the data. However, you can still modify these values if you want.
    • If you haven’t selected a dataset as a starting point, you will be able to directly type the name of the parameter and the values that you want to use to define the range (or fixed value) of the parameter.
      Note that in both cases, if the min value is higher or equal to the max value, the step will return an error.
  • Once you have added all the desired parameters, the design space is fully defined. Now, you will define how the sampling is performed.
  • First, specify the Method that you want to use for the sampling. The different methods as well as their benefits are listed below:
Random Sampling

Method
Data points are randomly selected from the parameter space.

Benefits
Simplicity, exploration of a wide range of parameter combinations, serves as a baseline for comparison.

Grid Sampling

Method
Parameter space is divided into a regular grid, and data points are selected at each intersection.

Benefits
Uniform coverage, systematic representation, reproducibility.

Latin Hypercube Sampling

Method
Stratified subsets of the parameter space are created, ensuring one sample per parameter with a random distribution.

Benefits
Efficient exploration, reduced correlation between parameters, flexibility in sample density.

  • Secondly, specify the number of samples the new dataset should have (Granularity). You can choose from three predefined options or specify a custom size.

Coarse100 Samples
Medium1,000 Samples
Fine10,000 Samples
CustomSpecify any integer > 0 (If the number is too high, you will receive an error saying that the platform ran out of memory).
  • You can then choose the Output name and click Apply.

Note that if you selected the Grid Sampling method, the step will produce the biggest possible uniform grid that fits within the Granularity specified. An info message will appear to show how many points were actually used to create the grid.


Examples

Consider the following table with four inputs. The table contains just two row with the min and max values for each column. The ID column is just a label to increase clarity. 

If you use that table for Sample Design Space on all four inputs and create a Latin Hypercube Sampling with ten points the result would look like this:

Alternatively, you could either (i) edit the min and max values of the parameters if the values picked from the table are not suitable, or (ii) create the new parameters from scratch, without using a table.


More on this step

Keep in mind that this function simply considers the design space as a hypercube. There's currently no option to add constraints.

Consider the example below with two inputs. If there was a constraint which excluded the space highlighted in red, the Sample Design Space function would nevertheless fill the entire cube as indicated by the blue dots. A workaround would be to first sample the entire design space and in a second step apply filters to remove the unwanted points from the dataset.



Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select atleast one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article