Fig. 4: Accuracy and density (fraction of predictions) of the Transformer for MaxFrag top-1 retrosynthesis accuracy.
From: State-of-the-art augmented NLP transformer models for direct and single-step retrosynthesis
The accuracy and density are shown as a function of the frequency of appearance of the top-1 SMILES in the output of the Transformer for the respective test sets of the models.
