0.3.0

This release adds an example case for inference. The model path can be controlled by the input file.

Full Changelog: 0.1.0...0.3.0

0.2.0

This release includes multiple bug fixes. The support of inert gas with arbitrary index (but not fixed to -1) is also added.
Model trained with deepflame-kit now matches well in accuracy with the former framework.

To-do

• Parameterize more constants, e.g. artifact paths
• Add sampling - data enhancement - training - evaluation workflow
• Accelerate inferring

0.1.0

This is the initial release for deepflame-kit. It provides a structured python package to train a neural network for reacting flow.
Users are able to train a neural network predicting the change rate of each reacting species. The inference interface to deepflame is also provided.
The command-line interface and the framework are heavily based on Pytorch Lightning.

To-do

• Add inert gas processing: they should not be predicted as this might affects the accuracy significantly.
• Vanilla Pytorch inference: Currently constructing the model on infer still requires Lightning.

New Contributors

• @hanli94 made their first contribution in #1

Full Changelog: https://github.com/caic99/deepflame-kit/commits/0.1.0