Automatic retraining of TensorFlow models
summary
An automated retraining pipeline uses an Airflow DAG to trigger AWS SageMaker jobs that read feature CSVs from S3, run TensorFlow Docker images stored in ECR, and write trained artifacts back to S3. Recurrent neural networks replace moving averages to predict daily cost and revenue for paid-search bids, with CloudWatch-extracted metrics, parallel hyperparameter sweeps, an automated fetch/evaluate step, and a manual gating step before production deployment to avoid regressions.
AI-written summary · about
chapters (5)
- 0:00Start
- 2:16Machine learning at Oui.sncf
- 3:51Positioning against competitors
- 6:33Machine learning product lifecycle
- 21:11Demo data
related talks
18mAutomate the boring ML stuff with pipelines
12mLeveraging AWS and Kubernetes for Scalable AI/ML Workloads in the Cloud
42mDeep learning beyond the learning
49mOrchestrating data and ML workflows with Apache Airflow
19mDrive transformation through machine learning
25mDesigning an AWS Well Architected Framework for AI/ML
33mDeploying and Scaling Spark ML and Tensorflow AI Models
43mMachine Learning Ops: DevOps for Data Science