Once a model has been trained using AOCL-DA, it can be saved to disk for later reuse without needing to retrain. This is particularly useful for computationally expensive models or when deploying models in production environments.
The recommended way to save and load AOCL-DA models is using joblib. Under the hood, joblib uses Python’s pickle protocol for the Python layer objects, which wraps AOCL-DA’s custom binary serialization format for the underlying C/C++ algorithm implementations. This ensures efficient storage while preserving all trained parameters and internal state.