v0.1.0 - Multifidelity Learning

A major set of changes as I introduced multi-fidelity learning algorithms

What's Changed

• Introduce a "Store" class by @WardLT in #119
• Readying for ExaRedox by @WardLT in #121
• Update Selector to use DB by @WardLT in #123
• Support multi-fidelity learning in RDKit-based models by @WardLT in #124
• Refactor multi-fidelity learning by @WardLT in #125
• Support multi-fidelity active learning by @WardLT in #122
• Allow pre-computing fingerprints by @WardLT in #126
• Add an example for multi-fidelity optimization by @WardLT in #127

Full Changelog: v0.0.5...v0.1.0

v0.0.5 - Refactor Thinkers

Makes the thinkers more flexible:

• Separate out configuration for different solutions methods
• Implement multiple types of thinkers
• Accept different types of data for search spaces

What's Changed

• Launch more tasks during startup phase by @WardLT in #112
• Tweak the performance of redox, gather more data by @WardLT in #115
• Support using pre-computed geometries for molecules by @WardLT in #114
• Refactor specifications and thinkers by @WardLT in #118

Full Changelog: v0.0.4...v0.0.5

v0.0.4 - Improved simulation pipelines, added multi-fidelity learning

This update marks some major changes in stability and a few in API:

• The redox property calculation has been made more robust, enough to evaluate 100k+ redoxmer energies at PM7 level
• The Scorer classes now support using multiple fidelities of data for training and inference

What's Changed

• Add BLYP//DZVP to checks by @WardLT in #93
• Fix adiabatic redox energy being larger than vertical by @WardLT in #96
• Screen conformers by whether optimized by @WardLT in #98
• Ensure that failed relaxations throw errors by @WardLT in #100
• Add code for creating initial datasets via brute force by @WardLT in #101
• Add ability to recover recipe from name by @WardLT in #103
• Initial models using updated datasets by @WardLT in #104
• Introduce multi fidelity learning by @WardLT in #105

Full Changelog: v0.0.3...v0.0.4

v0.0.3 - Support multiobjective optimization, hybrid GGA XC, OSX

What's Changed

• Add BOTorch Support by @WardLT in #75
• Skip appropriate sections from coverage by @WardLT in #76
• Add support for multi-objective optimization by @WardLT in #77
• Add EHVI, our first multi-objective selector by @WardLT in #78
• Add MOPAC by @WardLT in #83
• Add support for OSX by @WardLT in #84
• Add a starter which picks structurally-distinct molecules by @WardLT in #85
• Add support for B3LYP and other XC by @WardLT in #81

New Contributor

• @mbae26: Added the KMeansStarter class

Full Changelog: v0.0.2...v0.0.3

v0.0.2 - Bebop and Polaris-Ready Version

This version was tested on both Polaris (using CP2K) and Bebop (using Gaussian)

I'm using the release process to track what is happening with the project. We're not yet at a point where "major or minor versions" or milestones makes sense, but I'm doing enough that I wanted to provide a view for others to see which parts are seeing the most improvements.

What's Changed

• Remove buffer size as class attribute by @WardLT in #47
• Initial support for Gaussian by @WardLT in #48
• Make testing more stable by @WardLT in #51
• Add a "start" function to the applications by @WardLT in #50
• Add an example which uses Bebop by @WardLT in #52
• Use the GaussianOptimizer for optimizing structures by @WardLT in #54
• Add missing descriptors, test multiprocessing by @WardLT in #56
• Label calculation directories according to molecule, calculation by @WardLT in #58
• Add a caching capability to thinker by @WardLT in #61
• Add options and use proxies by @WardLT in #64
• Add support for running on Polaris by @WardLT in #73
• Refactor recipes by @WardLT in #74

Full Changelog: v0.0.1...v0.0.2

v0.0.1 - First Documented Release!

ExaMol is now documented at a point where we should start versioning it!