README-1.md

STAGE LEARNING WORKFLOW

We want to generate molecules given a prior (reinvent, libinvent, linkinvent, or mol2mol). This process can be subject to transfer learning, where, starting from a final model in the first stage of training, we want to introduce new scoring functions in a second stage (stage 1: final model → stage 2), focusing on a particular region of the chemical space. Another common practice is to select a checkpoint (.chkpt) from the best model (the model with the highest score) of a stage and continue to the second stage using this best model (stage 1: best model → stage 2).

This is the workflow followed by the stage learning environment in REINVENT4:|

flowchart TD
    A[run_staged_learning.py] --> B[create_adapter.py]
    B --> C[run_staged_learning.py]
    A --> D[learning.py]
    B --> D
    C --> D
    D --> A

Loading