Skip to content

README.md

Latest commit

 

History

History
30 lines (23 loc) · 1.57 KB

README.md

File metadata and controls

30 lines (23 loc) · 1.57 KB
Raw

bac

Fit bond additivity corrections.

  • Type 1: Linear regression with bond (or atom) types as parameters
  • Type 2: Non-linear regression based on BACs described by Anantharaman and Melius, JPCA (2005)

Input files

Required: Two space-separated .csv files with SMILES or InChI in the first column and enthalpies of formation in the second column. The first file contains calculated data and the second file contains experimental data.

Optional: A .xyz file with molecular geometries in the same order as the identifiers in the calculated data file.

How to run

python fit_bac.py <calc_data> <expt_data> <out_dir>

Optional arguments

General:

  • --weighted: Run a weighted regression if experimental uncertainties are available as a third column in expt_data
  • --val_split <float>: Separate out the specified fraction of data as a validation set
  • --folds <int>: Run cross-validation (overrides val_split)

Specific to type 1:

  • --atom_features: Fit atom additivity corrections instead of bond additivity corrections

Specific to type 2:

  • --geos <xyz>: Fit type 2 BACs using the provided xyz file with geometries
  • --geo_exceptions <file>: Override the geometry check for the identifiers in this file (e.g., when bonds aren't recognized in the 3D geometry)
  • --with_mult: Also fit molecular corrections if given the multiplicities as a third column in calc_data (destroys size consistency)
  • --global_min: Use a basin hopping algorithm to find the global minimum
  • --global_min_iter <int>: Number of basin hopping iterations
  • --lam <float>: Regularization parameter