A user on Reddit is seeking advice on implementing a machine learning approach for detecting performance regressions, specifically using anomaly detection techniques. They are working with a limited dataset of approximately 10 healthy samples for training and are unsure about the best practices for data splitting, threshold selection, and evaluation metrics. The user is questioning whether a traditional train/validation/test split is necessary for one-class anomaly detection and if their regression samples can serve as the test set. AI
IMPACT This discussion highlights common challenges in applying machine learning to performance monitoring and the need for robust evaluation strategies.
RANK_REASON User is asking for advice on a technical implementation, not announcing a new development.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →