without writing any formal plugin
Code snippets are taken from: https://hiphive.materialsmodeling.org/tutorial/index.html.
This repository demonstrates how to wrap hiphive routines inside AiiDA's @calcfunction and @workfunction in order to preserve the result and its provenance. Such functionalities do not require writing any dedicated plugin code, although putting the routines into a Python package may still be highly desirable.
datagen.py can be run to generate and store the example fit.
load-fcp.py contains example code for loading the force constant potential from an archive export.
To create an archive, run:
verdi archive create -N <uuid1> <uuid2> -- hiphive-example.aiida
where <uuid1> and <uuid2> are the UUIDs of the two nodes printed at the end of running datagen.py.
Note that running datagen.py can be done with a temporary in-memory profile without a fully working AiiDA installation. However, at the moment, it is not possible to export data into an archive from such an in-memory profile.
The load-fcp.py can run by any aiida-core>=2.0 installation; there is no need to install any non-Python dependencies.
This script loads the archive hiphive-example.aiida and produce the provenance graph and save the force constant data included in the aiida to the disk.
Provenance graph:
Example output of run.py:
======================================================
Created data {'fcp': <SinglefileData: uuid: e3182b58-505b-489b-8097-6df005bc01fe (pk: 1556)>, 'opt': <Str: uuid: 68cf97d4-d751-4d4a-aea0-2707acbeaf3b (pk: 1557) value: ===================== Optimizer ======================
seed : 42
fit_method : least-squares
standardize : True
n_target_values : 3840
n_parameters : 119
n_nonzero_parameters : 119
parameters_norm : 0.9275019
target_values_std : 1.027693
rmse_train : 0.01667685
rmse_test : 0.01975875
R2_train : 0.9997333
R2_test : 0.9996649
AIC : -28057.89
BIC : -27326.29
train_size : 3456
test_size : 384
======================================================>}