ProductizeML: Assisting Your Team to Better Build ML Products // Adrià Romero // MLOps Meetup #47

Model Updates

Updating machine learning models in production is crucial for maintaining their accuracy and effectiveness. Adrià Romero explains that the need for updates depends on the domain and the data collected from users. By sampling and labeling new data, models can be retrained to better solve real-world cases.

As soon as you label this new data that you are sampling and collecting, that's the point you should be retraining your model.

--- Adrià Romero

Demetrios Brinkmann emphasizes the importance of understanding KPIs and the model's objectives to ensure its continued relevance and performance 1.

   

Deployment Practices

Effective deployment practices are essential for ML models to be production-ready. Adrià Romero shares that models were trained using Keras, converted to TensorFlow, and deployed using Docker and Kubernetes. This approach ensures models are accessible via APIs for frontend integration.

We were using the TensorFlow serving to create the docker images that we were hosting in the cloud.

--- Adrià Romero

Demetrios Brinkmann highlights the balance between high-level explanations and detailed technical insights, aiming to inspire new product creation through hands-on experiences 2 1.