Skip to content

Non nested training data approach to multifidelity machine learning. Includes scripts run to test this on the CheMFi dataset.

License

Notifications You must be signed in to change notification settings

SPP2363/NonNestedMFML

 
 

Repository files navigation

Non-Nested Configuration of MFML and o-MFML

This code repository accompanies the manuscript titled 'Assessing non-nested configurations of multifidelity machine learning for quantum-chemical properties' available at [https://iopscience.iop.org/article/10.1088/2632-2153/ad7f25] and contains the scripts used to generate the various results and plots of therein. The CheMFi dataset was used in this work and can be found at [https://zenodo.org/records/11636903] with a preprint of the data descriptor at [https://arxiv.org/abs/2406.14149] for reference. The scripts included in this repository and their corresponding use are listed below for ready reference.

  • The script RepComp.py compares the different representations for a single fidelity KRR model on excitation energies and ground state energies from CheMFi.
  • Model_MFML.py is the module that was developed in this previous work and contains both both MFML and o-MFML implementations
  • TrueNonnested_Model_MFML.py is the development of MFML and o-MFML with a non-nested configuration of these approaches as documented in the preprint.
  • NestedCheMFiAllLCs.py produces the output corresponding to the nested configurations of MFML and o-MFML.
  • NonNestedCheMFiAllLCs.py produces the output required to test the non-nested configuration of MFML and o-MFML.
  • The jupyter notebook PlottingRoutines.ipynb contains the functions to produce the plots that result from the outputs of the various scripts.

About

Non nested training data approach to multifidelity machine learning. Includes scripts run to test this on the CheMFi dataset.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 94.0%
  • Python 6.0%