On-disk format

A PLAID Dataset is a collection of physics configurations, stored on disk in a human-readable, tool-friendly structure:

folder
├── dataset
│   ├── samples
│   │   ├── sample_000000000
│   │   │   ├── meshes/
│   │   │   │   ├── mesh_000000000.cgns
│   │   │   │   ├── mesh_000000001.cgns
│   │   │   └── scalars.csv
│   │   └── sample_yyyyyyyyy
│   └── infos.yaml
└── problem_definition
    ├── problem_infos.yaml
    └── split.json (or split.csv for <=0.1.7)

  • dataset/samples/: one directory per Sample.

  • meshes/: CGNS files for time steps; can be explored in ParaView.

  • scalars.csv: constant scalars for the Sample.

  • problem_definition/: learning task definition and splits.