How to use MACE

The script routine.py gives the flow of training & storing a MACE architecture, and immediately applies to the specified test dataset once training is finished. As such, it returns an averaged error on the MACE model compared to the classical model. More info on the training routine can be found in the paper.

The script routine.py takes an input file with the needed (hyper)parameter setup. An example of such an input file can be found in input/.

python routine.py example

Disclaimer:

In order to train MACE with a certain chemical dataset, the Dataset class should be made compatible with that data. Currently, the script src/mace/CSE_0D/dataset.py works only for the specific dataset used here, i.e. models from Maes et al. (2023), using the Rate22-CSE code.

Tutorial of the routine:

• Training a MACE model

Note

Required packages:

• torch

• torchode

• numpy

• matplotlib

• natsort

• tqdm