Tag: mlops
Concepts
- Automated Retraining Pipelines
- Batch vs Online Inference
- Canary Releases for Models
- Continuous Training Triggers
- Data Drift vs Concept Drift
- Data Validation and Schema Checks
- Dataloader Throughput and GPU Utilisation
- Distributed Training on Panel Data
- Experiment Tracking and Model Registries
- Feature Freshness and Staleness SLAs
- Designing Feature Pipeline DAGs
- Feature Stores
- Feedback Loops and Self-Fulfilling Predictions
- Hot-Reloading Models Without Downtime
- Running Hyperparameter Sweeps at Scale
- Immutable Training-Set Snapshots
- Incremental vs Full Feature Recomputation
- Inference Batching and Throughput
- Inference Latency Budgeting
- Label Latency and Delayed Feedback
- LLM Serving Cost and Latency Tradeoffs
- Maximum Mean Discrepancy Drift Tests
- ML Technical Debt and Hidden Coupling
- Packaging and Serialising Model Artefacts
- Model Packaging and ONNX
- Model Quantization for Inference
- Model Registry and Versioning
- Model Rollback and Freeze Procedures
- Model Serialization and Pickle Risks
- Model Serving Architectures
- Offline/Online Feature Parity
- Online Experimentation for Model Rollout
- ONNX and Cross-Language Model Export
- Estimating Live Performance Without Labels
- Pipeline Orchestration and DAGs
- Point-in-Time Correctness in Feature Pipelines
- Point-in-Time Feature Computation Engines
- The Population Stability Index
- Prediction Drift Monitoring
- Prediction Lineage and Auditability
- Pruning and Model Compression
- Rollback and Model Kill Criteria
- Schema-Constrained JSON Output
- Streaming Feature Computation
- Sweep Result Databases and Model Leaderboards
- Table Extraction From Financial PDFs
- Training-Serving Skew
- Unit Tests for Feature Transforms