VIP20
本课程属于：005-Artificial Intelligence and Machine Learning Engineering English Technical Courses · 共 20 门相关课程，组合学习更完整。
查看所属VIP全部课程 →
← 返回所属目录 咨询 / 购买

MLOps Pipelines for Model Training and Production Monitoring

30 Chapters
View Course Samples Online
01
MLOps pipeline fundamentals
lifecycle stages, reproducibility requirements, and the training-to-production gap.
MLOpspipeline
02
Pipeline orchestration frameworks
Airflow, Kubeflow Pipelines, and Metaflow for reproducible ML workflows.
Pipelineorchestration
03
Data versioning and lineage
DVC, LakeFS, and Delta Lake for traceable training datasets.
versioningDVC
04
Feature stores in training pipelines
Feast and Tecton for consistent offline and online feature computation.
trainingpipelines
05
Containerized training environments
Docker, Conda, and reproducible dependency management for model training.
Containerizedtraining
06
Experiment tracking and metadata management
MLflow and Weights & Biases for run comparison and artifact logging.
Experimenttracking
07
Distributed training orchestration
Horovod, PyTorch DDP, and Ray Train for scalable model training pipelines.
Distributedtraining
08
Hyperparameter tuning pipelines
Optuna, Ray Tune, and Katib integration with training workflows.
Hyperparameterpipelines
09
Model registry and artifact management
MLflow Model Registry and SageMaker Model Registry for versioned model promotion.
registryartifact
10
CI/CD for machine learning
automated testing, model validation gates, and continuous training triggers.
CI/CDautomated
11
Automated model retraining pipelines
drift detection triggers, scheduled retraining, and incremental learning workflows.
Automatedretraining
12
Model validation and evaluation pipelines
holdout testing, cross-validation, and fairness/bias checks in CI.
validationevaluation
13
Model packaging and serialization
ONNX, TorchScript, and Pickle for portable production artifacts.
packagingserialization
14
Model deployment patterns
batch inference, real-time REST endpoints, and streaming prediction pipelines.
deploymentpatterns
15
Canary deployments and shadow mode
safe model rollout strategies with traffic splitting and A/B testing.
deploymentsstrategies
16
Production monitoring architecture
metrics collection, logging, and observability stacks for ML services.
Productionmonitoring
17
Data drift and concept drift detection
statistical tests, windowed comparisons, and alerting thresholds.
detectionstatistical
18
Model performance monitoring
prediction quality metrics, delayed label handling, and automated degradation alerts.
performancemonitoring
19
Feature monitoring in production
feature distribution shifts, missing value rates, and schema validation.
monitoringproduction
20
Explainability and debugging in production
SHAP, LIME, and error analysis for live model diagnostics.
Explainabilitydebugging
21
Incident response for ML systems
rollback procedures, root cause analysis, and postmortem practices.
Incidentresponse
22
Pipeline observability and alerting
Prometheus, Grafana, and PagerDuty integration for ML workflows.
Pipelineobservability
23
Governance and compliance in MLOps
audit trails, model cards, and regulatory requirements for production models.
Governancecompliance
24
Cost optimization in training and monitoring pipelines
spot instances, autoscaling, and resource quotas.
optimizationtraining
25
Security in MLOps pipelines
secrets management, model artifact signing, and access control for training data.
SecurityMLOps
26
End-to-end case study
building a training pipeline with Kubeflow, MLflow, and Feast for a recommendation model.
End-to-endbuilding
27
End-to-end case study
deploying and monitoring a real-time fraud detection model with drift alerts.
End-to-enddeploying
28
End-to-end case study
automated retraining pipeline triggered by data drift for a demand forecasting model.
End-to-endautomated
29
Scaling MLOps pipelines
multi-team workflows, shared feature stores, and platform engineering practices.
MLOpspipelines
30
Future directions in MLOps
LLMOps, foundation model monitoring, and continuous evaluation at scale.
directionsMLOps