Exploring Effects of Modified Machine Learning Pipelines of Astrochemical Inventories

Abstract Machine learning pipelines for astrochemical inventories have been introduced as a useful addition to the astrochemist toolbox, having first been used to model and predict column densities in the Taurus Molecular Cloud (TMC-1). Rapid changes in the field of machine learning have provided new tools in optimizing this pipeline, including improved vector representations. Furthermore, the addition of new detections since the original model allows for a retrospective analysis of model performance, in addition to new data for the model. This study revisits TMC-1, investigating both effects of an increased detection inventory on the model and changes to various portions of the pipeline, yielding a significant improvement in column density predictions. Through these comparisons, we attempt to derive insight into the ultimate effectiveness of these models, as well as their current limitations and words of caution in their use and interpretation. Finally, we provide suggestions for future machine learning of interstellar sources.