Based on the Create_Model_Demo notebook from the ReinventCommunity repo. If you are planning to train a new model from scratch, this app is the first step of the process. This module creates an initial generative model of prior/agents for the next step (Reinvent: Train Initial Generative Model for Prior/Agent ) by generating a vocabulary based on your SMILES training data. The vocabulary defines what tokens the model can propose, which directly controls the possible atom types in the SMILES output. The output of this app (empty_model.ckpt) will be the input for the next app (Reinvent: Train Initial Generative Model for Prior/Agent ) that trains the generative model. Finally, the output file of the second app is what you can use in Reinvent: Reinforcement Learning. For more information, please see the tutorial page of Reinvent Apps.

Example Use Case: Generate generative model of prior/agents

Limitation: At least 100.000 SMILES are recommended for creating a generative model