Skip to content

Latest commit

 

History

History
77 lines (53 loc) · 2.89 KB

README.md

File metadata and controls

77 lines (53 loc) · 2.89 KB

Generate Molecular Representations

This directory contains all scripts required to generate molecular representations, including wall-time measurements.

Representations for a Full Dataset

To generate molecular representations for an entire dataset, use the dataset_submit.sh script.
Arguments must be provided in the following order: Dataset name, Representation type, and Directory path containing XYZ files.
For the representation types used in the main text, see here.

bash dataset_submit.sh TM-GSspinPlus SPAHM_e ../data/TM-GSspinPlus/0-xyz

This command creates a job file named: generate_{rep}_{dataset}.job

Debug Mode (10 Molecules)

To generate representations for only 10 molecules (useful for testing and debugging), add the debug argument:

bash dataset_submit.sh TM-GSspinPlus SPAHM_e ../data/TM-GSspinPlus/0-xyz debug

This creates a job file named generate_{rep}_{dataset}_debug.job

If you are running on a SLURM-based system, submit the job using sbatch. For local execution, simply run the job file with bash.

Representations for a Subset

To generate representations for a predefined subset of molecules, use subset_submit.sh.
In addition to the dataset, representation type, and XYZ directory, provide a text file containing refcodes for the subset.

bash subset_submit.sh TM-GSspinPlus SPAHM_e ../data/TM-GSspinPlus/0-xyz \
TM-GSspinPlus/subset_refcodes/TM-GSspinPlus_sub_0.txt

Related Files and Scripts

For further details, see:

  • dataset_submit.sh
  • subset_submit.sh
  • Python scripts in each dataset directory:
    • generate_rep_spahm_qstack.py
    • generate_rep_slatm_fchl_qml2.py
    • generate_rep_soap_featomic.py
  • Example output files in examples_output/*.out

Execution timings are reported in the last line of each output file.
The job generates representation arrays (.npy files) in the corresponding dataset directory (for example, see TM-GSspinPlus).

Notes for the OctaKulik

For representations used to predict spin-splitting energies, comment out the lines that include high-spin geometries (or high-spin states).
See the Python scripts in OctaKulik for details.

Kernel computations

To measure kernel computation times for Gaussian (G) or Laplacian (L) kernels, use kernel_submit.sh.
Provide the Python script time_kernel_qstack.py, dataset name, representation type, path to the .npy representation file, and kernel type.

bash kernel_submit.sh time_kernel_qstack.py TM-GSspinPlus SPAHM_e \
TM-GSspinPlus/SPAHM_e-TM-GSspinPlus-subset.npy L

This command generates a job file named: kernel_{kernel}_{rep}_{dataset}.job

For details, see:

  • kernel_submit.sh
  • time_kernel_qstack.py
  • Example timing outputs in examples_output/*.out.

Execution timings are reported in the last line of the corresponding output file.