Appendices A Dataset information

The USPTO-50K dataset is annotated with 10 reaction types, the distribution of reaction types is displayed in Table 4. The distribution is extremely unbalanced. Table 7: Atom Features used in EGA T. All features are one-hot encoding, except the atomic mass is a real number scaled to be on the same order of magnitude. Semi-templates size is 654 for the USPTO-50K dataset.Feature Description Size Atom type Type of atom (ex. The decoder is similar to the encoder.