Skip to content

Configuration

Backmapping configuration

Backmapping is defined through a dedicated JSON file with the following structure:

{
  "case_json": "case.json",
  "input": {
    "file": "x01.txt",
    "format": "txt"
  },
  "reduced_space": {
    "nconf": 5
  },
  "denorm": {
    "rule": "3sigma",
    "c": 3.0
  },
  "output": {
    "file": "results/u_base.txt",
    "format": "txt",
    "layout": "col",
    "what": "Ubase",
    "write_meta": true
  }
}

Fields description

case_json

Type: string

Path to the original configuration file used to generate the model.

This is required to:

  • reconstruct normalization settings
  • recover dataset structure

Input (input)

Defines the reduced coordinates.

Fields

  • file (string)
    Path to the file containing reduced coordinates

  • format (string)
    Input format (e.g. "txt")

Reduced space (reduced_space)

Defines how many configurations are processed.

Fields

  • nconf (int)
    Number of configurations to backmap

Denormalization (denorm)

Controls how reduced variables are mapped back.

Fields

  • rule (string)

    "3sigma" → scaling based on standard deviation

  • c (float)

    scaling coefficient (typically 3.0)

Output (output)

Defines output file and format.

Fields

  • file (string)
    Output file path

  • format (string)
    Output format (e.g. "txt")

  • layout (string)

    "col" → column-wise output
    "row" → row-wise output

  • what (string)

    "Ubase" → reconstructed baseline variables

  • write_meta (bool)
    Write additional metadata

Notes

  • input reduced coordinates must match the dimensionality of the trained model
  • denormalization depends on the original dataset statistics
  • the backmapping process reconstructs design variables consistent with the training space

Summary

Backmapping is a post-processing step that maps reduced coordinates back to:

  • original design variables
  • physically interpretable configurations

using the model defined in the original case.