Experiment tracking, model registries, feature stores, drift monitoring, CI/CD for ML, serving and batch versus online inference.
35 cards · basic cards · AI-written, checked twice. Edit anything.
- What is experiment tracking in MLOps?
Show answer
Recording and comparing ML experiments, parameters, metrics, and results for reproducibility and analysis- What is a model registry?
Show answer
Centralized repository for storing, versioning, and managing trained models across their lifecycle- What is a feature store?
Show answer
Centralized system for storing, managing, and serving features for training and inference- Define data drift
Show answer
When the distribution of input features changes over time in production- Define model drift
Show answer
When a model's performance degrades because the underlying data distribution has changed- What is ML CI/CD?
Show answer
Continuous integration and deployment practices adapted for ML pipelines, models, and data- What is batch inference?
Show answer
Processing many prediction requests together in a single computation for efficiency- What is online (real-time) inference?
Show answer
Generating predictions for individual requests immediately as they arrive- Why is model versioning critical in production?
Show answer
Enables reproducibility, rollback to previous versions, and comparison between iterations- What should experiment tracking record?
Show answer
Hyperparameters, metrics, code versions, datasets, timestamps, and environment configuration- Why use artifact storage in ML pipelines?
Show answer
To preserve models, datasets, and transformations needed for reproducibility and auditing- What is model validation?
Show answer
Testing a model to verify it meets performance and business requirements before production deployment- What is A/B testing in ML?
Show answer
Comparing two model versions in production by serving them to different user groups- What is shadow deployment?
Show answer
Running a new model in parallel with production without affecting real user predictions- What is canary deployment?
Show answer
Gradually routing small percentages of traffic to a new model while monitoring its performance
FlashKeepers