A guide on how to set up the project on a local machine. This section should include steps to install any dependencies required to run the project.

# create a new conda environment
conda env create -f environment.yml
conda activate rl-opt

# install the packages
pip install -r requirements.txt

Explain the organization of this project's folders and their contents:

.
├── Evaluation
│   ├── compare.py
│   ├── pysisyphus
│   ├── test-configs
│   └── worker.py
└── Train
    ├── configs
    ├── pysisyphus
    └── run.py

• configs: Contains the configuration files for training the model.
• pysisyphus: this package is modified from the original pysisyphus. Optimizer.py is modified to the generator. It is also added RDKit.py as a new calcaulator.
• run.py: This file is the entry point for starting the model training process.

In configs

• test-configs: you can determine the dataset, the type of coordinates(e.g. cartesian(cart) or internal(redund) coordinates), the type of calculator(e.g. mmff or psi4) and the type of optimizer(e.g. bfgs or rl).
• worker.py: This file is to collect the data(.json) which the optimizer optimize the molecular geometry.
• compare.py: This file is to compare the performance of different optimizers after collecting the data.

cd Train
python run.py \
    --name NAME \ 
    --version VERSION \
    --seed SEED

We select the configuration file at many510/v3.ini and set the name as many510 and the version as 3. The seed is set as 283484033. The command is as follows:

cd Train
python run.py \
    --name many510 \ 
    --version 3 \
    --seed 283484033

After training, the model will be stored in Train/saves/NAME/VERSION/. You need to copy or move the model to the Evaluation folder if you want to evaluate the model.

cd Train
python runWithPretrain.py \
    --name many510 \
    --version 3 \
    -r 708700354 \
    --pre_trained saves/RL-many510/v3/seed-283484033/best_model.pt \

Before we compare the performance of different optimizers, we need to collect the data(.json) which the optimizer optimize the molecular geometry. The data will be stored in Result-GROUP_NAME/OPTIMIZER_NAME/.

To collect the data, we can run the evaluation script first.

python worker.py --name GROUP_NAME --opt_key OPTIMIZER_NAME -p

In our setup, we have chosen BFGS (denoted as bfgs) as our optimizer. To utilize a dataset with perturbations, add the -p flag. If you prefer to use a dataset without perturbations, simply omit the -p flag. The script fetches its settings from the configuration file located at test-configs/bfgs.yml, which specifies which dataset to use.

The command to execute the script is:

python worker.py --name group-C --opt_key bfgs -p

To run the dataset without perturbations, use the following command. The script fetches its settings from the configuration file located at test-configs/rl.yml, which specifies which dataset to employ.

The command is as follows:

python worker.py --name group-C --opt_key rl

After collecting the data, we can compare the performance of different optimizers by running the following script.

python compare.py

cd e-Baker-Evaluation
python main.py