Fig. 4
From: Molecular generation by Fast Assembly of (Deep)SMILES fragments
Model training (left) and sampling speed (right). RNN: Recurrent Neural Network (numbers cited from literature [8]); Frag (train): molecular fragmentation (Python/RDKit); Tag: tagging cleaved bond (proposed method, Python/RDKit); Frag (sampling): assembly of molecular fragments using molecular graph operations (Python/RDKit); Smi: fast assembly of SMILES fragments (proposed method, OCaml). Dsmi: fast assembly of DeepSMILES fragments (proposed method, OCaml); Tag is the model training prerequisite of Smi and Dsmi sampling. All methods use a single CPU, except RNN which uses four CPUs and one GPU