Skip to main content

Auto-Tune (Special)

Description​

The Auto-Tune policy provides automated model selection and hyperparameter optimization. When configured, it evaluates multiple underlying policy types (specified in the policies list) and/or searches for optimal hyperparameters for a given policy using techniques like cross-validation. It selects the best configuration based on a specified performance metric (e.g., NDCG).

Policy Type: auto-tune Supports: embedding_policy, scoring_policy

Configuration Example​

scoring_policy_auto_tune.yaml
policy_configs:
  scoring_policy: # Or embedding_policy
    policy_type: auto-tune
    # List the candidate policy types (as strings) to consider
    policies:
     - "lightgbm"
     - "als"
     - "bert4rec"

References​

  • Bergstra, J., & Bengio, Y. (2012). Random Search for Hyper-Parameter Optimization. JMLR.
  • Snoek, J., Larochelle, H., & Adams, R. P. (2012). Practical Bayesian Optimization of Machine Learning Algorithms. NeurIPS.