Heterogeneous Data Training

MACE supports loading multiple files for training, validation, and testing, with flexible support for different file formats within a single training run. This capability is particularly useful when working with data from multiple sources or when your dataset is split across several files.

Supported File Formats

MACE can handle a mix of the following file formats:

ASE files: Standard ASE-readable atomic structure files
HDF5 files: Pre-processed data in .h5 or .hdf5 format
LMDB databases: Efficient storage and access for large datasets

Configuration

Basic Usage

You can provide multiple files as a list in your configuration:

heads:
  DFT:
    train_file: ["path/to/training1.xyz", "path/to/training2.xyz"]
    valid_file: ["path/to/validation1.xyz", "path/to/validation2/"]
    test_file: "path/to/test.xyz"

Mixing Different File Formats

MACE automatically detects file types and uses the appropriate loader:

heads:
  QM7:
    train_file: ["data/qm7/train.xyz", "data/qm7/train.h5", "data/qm7/database/"]
    valid_file: ["data/qm7/valid.xyz", "data/qm7_h5/valid/"]
    test_file:  ["data/qm7/test.xyz", "data/qm7/test.h5"]