Data for: Setting New Benchmarks in AI-driven Infrared Structure Elucidation

This is data accompanying the paper: Setting New Benchmarks in AI-driven Infrared Structure Elucidation. Specifically, it contains a checkpoint of the best model and the pseudo experimental spectra generated as part of the work.

Model Checkpoint:

Checkpoint file: ir_ps_75_learned_gated_linear_augment_combined_last.ckpt

Preprocessor file: preprocessor.pkl

Overrides: overides.yaml

This model was trained with a patchsize of 75, learned positional embeddings, gated linear units enabled and with all augmentation methods for 30 epochs.

Pseudo-Experimental Spectra:

pseudo_experimental.zip contains pseudo-experimental spectra mimicking real spectra for a total of 700.000 molecules. The data comes in the form of a HuggingFace dataset.