qml2

Code repo for convenient usage of methods developed in the chemspacelab group. Successor to qmlcode/qml repo that also draws a lot of inspiration from qmllib and qml-lightning repos.

🔧 Installation

Install with:

pip install .

or, if makefile is installed,

make install

NOTE: Using FJK representation or constructing adaptive basis sets additionally requires installing pyscf package via

pip install pyscf

📋 Testing

To check that the installed repo works correctly run

make test

NOTE: The command assumes that python environmental variable points towards a valid executable. If you use an environment alias change definition of the python variable in the beginning of the Makefile.

Additional scripts

Apart from the source the code contains the following folders:

sanity_checks - scripts used during development to check implemented procedures work correctly.

failure_cases - scripts demonstrating problems encountered in previously implemented procedures after they've passed peer-review publication stage. (Hopefully inspiring others to fix them.)

💻 Environmental variables

Calculation management

NUMBA_NUM_THREADS - since the code is written in Numba OpenMP parallelization is mostly controlled via this variable.

QML2_NUM_PROCS - number of processes spawned by parts of the code parallelized via python.multiprocessing (training set representations in model-related classes, pyscf calculations made by OML_Compound_list attribute calls). For limiting number of OpenMP threads spawned in turn by these processes use suppress options (such as KRRModel class's training_reps_suppress_openmp option). Also see parallelization.set_default_num_procs.

Experimental

QML2_DEFAULT_JIT - setting to NUMBA (default) or TORCH (both are case insensitive) determines whether Numba or TorchScript JIT compilation is used. Also see jit_interfaces.set_default_jit.

Debugging

QML2_DEBUG - if 1 add debug=True to all Numba @njit instances.

QML2_SKIP_JIT - if 1 do not use JIT.

QML2_SKIP_JIT_FAILURES - if 1 TorchScript does not terminate when encountering uncompilable part of the code.

🤝 Contributing

We use several packages that maximize code readability, listed in requirements-dev.txt; hence should you decide to commit make sure you have a conda environment you are prepared to modify. Having, for example, created a fresh conda environment named qml2dev with

conda create --name qml2dev
conda activate qml2dev

one prepares the environment and the pre-commit scripts in the repository with

make dev-setup

This allows automatic formatting/readability checks for the committed code. It is also possible to enforce adherence to Conventional Commits format of commit messages inside your fork with

make conventional-commits