Added the KS Hamiltonian mapping with machine-learning workflow to workflows/nbra:
Added the data generation workflow for generating guess and reference data for ML:
Generating data with CP2K at different levels of theory
Added the do_more option for additional reference calculations
Automated the generation of sample files used in the ML workflow
Added the ML workflow for training a KRR model with guess and reference model
KRR models with different kernels and hyperparameters
Two different partitioning: equal and atomic partitioning
Added a computational procedure for computing the energies, overlaps, and time-overlap matrices for molecular orbitals obtained from ML models
Added new functionalities to CP2K methods:
Reading and resorting atomic orbital matrices data
Solving generalized KS equation
Some other changes to the code:
Moving pdos files as soon as they are completed in step2 to all_pdos files for the non-ML step2 workflow
Added loading of npy and npz files reading to data_read
Added the installation of scikit-learn package to the installation instruction in the main README file
Added the atomic mass dictionary to md_align module for easier alignment of the MD trajectory
This workflow is tested using C60 and benzene molecule. The tutorials for benzene molecule are uploaded to the Tutorials_Libra repository in CompChemCyberTraining.
Added the KS Hamiltonian mapping with machine-learning workflow to workflows/nbra: