2025-26-MEng-Research-Project-Team12

----- Setup -----

1) Install pytorch on GPU (windows)

• ref: https://www.youtube.com/watch?v=d_jBX7OrptI

2) You can also install all libraries I had ('binary_test.yml')

• anaconda prompt -> 'conda env create -f binary_test.yml'

3) Files you can run

• 'preprocessing/data_custom.ipynb': data preprocessing
• 'train_model.ipynb': model training
• 'test_model.ipynb': predict Gibbs E and enthalpy at 298K -> calculate Gibbs E at different T

4) Download datasets

• Train dataset: BinarySolvGH-QM.csv
• • ref: https://zenodo.org/records/14238055
• Final test dataset: Data and predictions of solvation free energies in binary solvents (BinarySolv-Exp).csv
• • ref: https://doi.org/10.1016/j.cej.2025.162232

5) Create 'data' folder

• After creating 'data' folder, move the all datasets to this path

----- To-do list -----

1) Data preprocessing

• Run 'data_custom.ipynb' to preprocess the train and test datasets
• The preprocessed datasets should be saved in the same path with 'train_model.ipynb' and 'test_model.ipynb'

2) Complete the 'test_model.ipynb' to calculate Gibbs E in different temperature

• You can predict Gibbs E and enthalpy at 298K via 'test_model.ipynb'
• Complete the final code to predict Gibbs E at different temperature

3) Check the pipline

• Run 'train_model.ipynb' using the preprocessed 'BinarySolvGH-QM.csv' with default hyperparameters
• Run 'test_model.ipynb' using the preprocessed 'Data and predictions of solvation free energies in binary solvents (BinarySolv-Exp).csv'

4) Hyperparameter optimization (HPO)

• Optimize the hyperparameters of the models to minimize its loss
• You can find the main hyperparameters by searching 'hyperparameter' in 'train_model.ipynb'