MPNN is a small library for Minima-preserving neural network construction. See Katrín Blöndal, Khachik Sargsyan, David H. Bross, Branko Ruscic, and C. Franklin Goldsmith, "Adsorbate Partition Functions via Phase Space Integration: Quantifying the Effect of Translational Anharmonicity on Thermodynamic Properties", The Journal of Physical Chemistry C 2021 125 (37), 20249-20260. DOI: 10.1021/acs.jpcc.1c04009.
Build the library
./build.sh
Load data
prep/prep_xytrain.x <caseid>
prep/mpnn_load.py <caseid>
# where <caseid> = 1 for H PBE-D3(ABC), 2 for H BEEF_vdW, 3 for CO, 4 for CH3OH, 5 for CH3OH_2d, 6 for CH3
Exploratory data analysis
prep/plot_yx.py
prep/plot_xdata.py
prep/plot_ydata.py
Training
fit/mp_fit.py
Postprocess
postp/plot_parity.py
postp/plot_datacl.py
postp/plot_zslice.py
postp/plot_2dslice_tile.py
postp/plot_2dslice_xtrain.py
postp/plot_2dslice_grid.py
postp/mpnn_eval.py
postp/mpnn_integrate.py <T> <N> <seed> # only for CO for now, seed is optional