Guidelines
Begin by initializing the facts client at the top of the notebook before importing any modules.
When dealing with external models, import them in custom training scripts.
Each specific use case should be labeled as an Experiment, and any training session with one or different machine learning frameworks should be referred to as a Run. Keep in mind that each experiment can encompass multiple runs.
It is recommended to use a single experiment for a given use case and notebook. This helps in avoiding the creation of multiple experiments without reason, as it can impact lineage tracking and comparison.
To ensure that facts are displayed along with stored Watson machine learning models and for proper lineage tracking, users should set custom meta attributes before storing the model using the provided utility.
For native learners in external providers (e.g., AWS Sagemaker’s Linear learner), autolog is not supported. In such cases, the manual log option should be utilized.
When employing external models, ensure that the same experiment name is used during client initiation. This aids in tracking notebook experiments along with your model asset.