Input parameter optimization

Features
1

Hello everyone! We’re excited to announce the release of a new feature for optimizing model input parameters.

2 Likes
2

Feature

The platform provides solution of an inverse problem: find input conditions leading to specified output constraints. It computes inputs minimizing deviation measured by loss function.

To run fitting

  • Click the “Fit inputs” icon on the top panel
  • Specify inputs to be found
  • Set output constraints
  • Click the “Run” icon. You will get a grid containing
    • fitted inputs
    • viewers visualizing the goodness of fit
    • line chart showing the loss function minimization

fitting-run

Learn more

2 Likes
3

Annotation

You can apply parameter optimization to any function with the RichFunctionView editor. Add meta.features: {"fitting": true} to enable it:

//name: Test
//language: javascript
//input: double x
//output: double y
//editor: Compute:RichFunctionViewEditor
//meta.features: {"fitting": true}

let y = x * x;
1 Like
4

Diff Studio

Datagrok Diff Studio solves ordinary differential equations right within your web browser, and parameters optimization is a built-in feature:

  • Click Run fitting inputs icon
  • In the Fit block, use switchers to specify inputs to be found:
    • Set min and max values for each selected item
    • Set values of all other inputs
  • In the Target block, specify output constraints:
    • Set dataframe with expected output values (in the table input)
    • Set column with values of the independent variable (in the argument field)
  • Specify settings of fitting:
    • Choose numerical optimization method (in the method field), and set loss function type (in the loss field)
    • Specify number of points to be found (in the samples field)
    • Set the maximum scaled deviation between similar fitted points (in the similarity field): the higher the value, the fewer points will be found
  • Click Run fitting icon. You will get a gridcontaining
    • loss function values
    • fitted inputs
    • line charts visualizing the goodness of fit and showing the loss function minimization
  • Open Context panel (F4). You will get the simulation run corresponding to the selected grid row

DiffStudio-run-fitting

1 Like
5

We’ve improved Parameter optimization. A new version accelerates computations and provides the reproducibility and early stopping features.

Check the updated Scientific Computing tutorials:

1 Like