ML_iterative_algorithm: A Python repository from ssthurai

ML_iterative_algorithm

This repository includes the python scripts to generate the surface/molecule descriptors and to run the developed iterative ML algorithms. Each folder has the main codes and an example.

re-ordered Coulomb Matrix

Prepare the molecule structures. Here we recommend to use the ASE to build the structure with the .traj format, other formats also work, such as .cif, .xyz.
See the example to get the reordered Coulomb Matrix. ($python run_test.py ). The descriptors will be saved in a .csv file.

Extract the DOS of active site

The extractDos.py file is in the path of /data/scripts_py.zip, this file is used to extract the DOS of active site.

DFT calculated DOS was saved in the .pickle file.
The site name and atom index on clean surface are required. ($python run_get_data.py) The descriptors are also saved in a .csv file.

Iterative algorithm for looking for the valuable training set

The Iteration*.py files are in the folder of Iteration_I_AdsorptionE , all these scripts work for training process. We also show an example in this folder.

Max predicted error: a parameter to control the accuracy and number of DFT calculations.
$python run_test.py

Iterative algorithm for finding the low-energy structure

The example is constructing the phase diagrams including ML phase diagrams and DFT phase diagrams. Please see the folder of Iteration_II_PhaseDiagram. Each folder have the main scripts and an example. The algorithm can construct the N dimensional phase diagrams (1<= N <=4).

atomType_chemPot : a dictionary to specify the atomic type and the range of chemical potential.
N_int : Resolution of phase diagram
$python run_pd.py

Note

The data folder includes the structure file (.cif) and python script files.

Paper: Scalable approach to high coverages on oxides via iterative training of a machine-learning algorithm (ChemCatChem)