• Download Chromophore dataset from https://figshare.com/articles/dataset/DB_for_chromophore/12045567/2, and leave only Absorption max (nm), Emission max (nm), and Lifetime (ns) column.
  • Make separate csv file for each column, and erase the NaN values for each column.
  • We log normalize the target value for Lifetime (ns) data due to its high skewness.
• Download Solvation Free Energy datasets from https://www.sciencedirect.com/science/article/pii/S1385894721008925#appSB, and create the dataset based on the Source_all column in the excel file.
  • Make separate csv file for each data source.
• Put each datasets into data/raw and run data.py file.
• Then, the python file will create {}.pt file in data/processed.

Following Options can be passed to main.py

--dataset: Name of the dataset. Supported names are: chr_abs, chr_emi, chr_emi, mnsol, freesol, compsol, abraham, and combisolv.
usage example :--dataset chr_abs

--lr: Learning rate for training the model.
usage example :--lr 0.001

--epochs: Number of epochs for training the model.
usage example :--epochs 500

--intervention: Decision on whether model performs intervention or not. usage example :--intervention True

--conditional: Decision on whether model performs conditional intervention or naive intervention. usage example :--conditional True

--lam1: Hyperparameters for weight coefficient for KL Loss.
usage example :--lam1 1.0

--lam2: Hyperparameters for weight coefficient for intervention Loss.
usage example :--lam2 1.0