Experiment Tracking

Created: 2023-02-13 14:41
#note

Each ML training run is considered and experiment. Experiment tracking helps manage the evolution of an ML model toward stated performance goals.

We can track model (different version and hyperparameters), inputs, outputs, and results.

Benefits:

• measure impact of changing model parameters;
• identify model behavior changes due to new training data;
• measure project progress toward stated requirements;
• decide on model promotion to production;
• automated model analysis and promotion with experiment-tracking tools.

Tags

#mlops #ml