cli-anything-unimol-tools

You can automate molecular property prediction workflows using a Python-based CLI that orchestrates Uni-Mol Tools for end-to-end training and inference. It provides interactive mode, named projects, and JSON output to reduce manual setup and orchestration.

Uni-Mol Tools support five task types for molecular property inference and training: classification, regression, multiclass, multilabel_classification, and multilabel_regression. You can select the appropriate task type within the interactive CLI environment.

Yes, you can get machine-readable JSON outputs from Uni-Mol Tools training runs by using the --json flag. This feature enables easy automation and integration of model tracking and experiment results into local sandboxes.

You can manage multiple molecular property experiments using the CLI's project management features, which organize experiments with named projects and persistent state. This allows researchers to maintain cross-task data management across local sandboxes.

No specific external dependencies are required to run the Uni-Mol Tools CLI itself, as it operates as a standalone Python module. You simply launch the interactive REPL by running the python3 command to start your project-scoped experiments.

The best way to start an interactive molecular property training session is to launch the Python CLI by running the python3 -m cli_anything.unimol_tools command. This opens an interactive REPL for immediate model tracking and inference.