v2021.09

structure optimization with on-the-fly ML
nudged elastic band with on-the-fly ML
cl.offline for offline training
examples
refactored deprecated code
math module merged with descriptor
util module split to analysis
better parallelism
faster updates
faster pickles
works with: numpy_1.21.1, scipy_1.5.2, torch_1.10, ase_3.22.0

v2021.05

faster IO (model pickles).
wildcard/specific kernels.
examples: VASP, GAUSSIAN, etc.

v2021.04

A) geometry optimization with on-the-fly ML: theforce.cl.relax
B) model instantiation by random displacements of given coordinates: theforce.cl.init_model
C) command line coupling with the Gaussian calculator: theforce.calculator.gaussian
D) Langevin dynamics from the command line: theforce.cl.md
E) steps towards metadynamics: theforce.calculator.meta

v2021.01

A) new command line utilities: theforce.cl.md, theforce.cl.train, theforce.cl.test
B) better regression algorithm by optimizing sigma (noise_e, noise_f).
C) more stable sampling algorithm by predicting the error for lce energies.
D) automatic pickling of the generated model on-the-fly
E) meta-dynamics is almost ready

v2020.12

Modules sesoap and socketcalc are created.

v2020.08

This is released before full transition to ase 1.19 and pytorch 1.7.