git clone https://github.com/mjwen/rxnrep.git
cd rxnrep
git checkout bondnet 
conda env create -f environment.yml
conda activate rxnrep
pip install -r requirements.txt
pip install -e . 

First, update electrolyte.yaml by setting the trainset_filename, valset_filename, and testset_filename to the absolute paths to the data files on your machine. Example data files are located at dataset/electrolyte.

Next, train the model by

python run.py model/decoder=regressor.yaml datamodule=regression/electrolyte.yaml

Instead of using mrnet reaction, reactions can be provided as SMILES strings. See dataset/nrel for example datasets.

To train, first update nrel.yaml
by setting the trainset_filename, valset_filename, and testset_filename to the absolute paths to the data files on your machine (e.g. on in dataset/nrel).

Then, train the model by

python run.py model/decoder=regressor.yaml datamodule=regression/nrel.yaml

bondnet <path_to_data_file> 

where path_to_data_file is a json file containing a list of mrnet Reaction you want to predict; of course, subclasses of Reaction is OK. An example data file can be found at examples/reactions_mrnet.json. Note, depending on how the mrnet reactions are constructed, atom mapping may or may not be there. If not, you can use get_reaction_atom_mapping() to generate the atom mappings.

Running this command will generate a file named results.json, which is a copy of the input data file, but with an additional key predicted_reaction_energy inserted to each entry, the value of which gives the BDE if the reaction is a one-bond breaking one.

If you've trained your own model and want to use it for prediction, you can do

bondnet <path_to_data_file>  --model <path_to_model>

Again, path_to_data_file is a file containing the reactions you want to make predictions for. path_to_model is the path to the model you've trained. More specifically, after training, models will be saved to a directories looking like

- outputs
  ... 
  - 2022-08-09
    - 14-22-51
    - 16-44-02
  - 2022-08-10
    - 10-10-02
    - 15-24-01

The latest model will always be the last one in the list. So, for example, you can then set <path_to_model> to ./outputs/2022-08-10/15-24-01 to use the latest model for prediction.