globus-labs / faster-molecular-hessians / 7090659520

97%

push

github

Add internal coordinate harmonic model (#26)

* Refactor and add linear terms

Fixes #2
Fixes #4

* Make the second-order terms 1/2 the first order

* Update with the newer model

* Update the proof of concept notebooks

* use internal coordinates in linear model (linear_internals.py) (#25)

* see freqs

* Add computing Hessians using MBTR forcefield (#8)

* Add a ASE calculator for MBTR

* Move back to Py3.9 for dscribe

I get an error "from collections import Iterable" stemming
from the sparse module used by dscribe

* Add force computation

* Add ability to compute Hessian from MBTR

* Remove unused import

* Add an example notebook with MBTR

* Adjust the scale to be a unit normal

Should reduce the likelihood of numerical issues

* Update proof-of-concept with more data, better KRR

* Allow use of MOPAC in proof of concept (#13)

* Separate notebooks for exact answer, steps in fitting (#14)

* Add utility for making calculator

Avoid copying code between notebooks

* Separate exact, sampling and approximation

* Accept "None" as an option

* Sample as if computing numerical derivatives  (#17)

* Start notebook for displacing along axes

* Add code for combining perturbations into single tasks

* Add example of using this new data

* Refactor and add linear terms

Fixes #2
Fixes #4

* Make the second-order terms 1/2 the first order

* Update with the newer model

* Update the proof of concept notebooks

* see freqs

* internal coordinate model

* flake8 cleanup

* added pytest for internals

* more flake8 cleanup

* geometric package to convert to internal coordinates

---------

Co-authored-by: snelliott <s.nicole.elliott@gmail.com>
Co-authored-by: Logan Ward <WardLT@users.noreply.github.com>
Co-authored-by: Logan Ward <ward.logan.t@gmail.com>

* Remove only notebook directory

We can recreate these in the new directory structure if need be

* Move the exac... (continued)

75 of 83 new or added lines in 1 file covered. (90.36%)

370 of 381 relevant lines covered (97.11%)

0.97 hits per line