From the course: Advanced Data Processing: Batch, Real-Time, and Cloud Architectures for AI

Unlock this course with a free trial

Join today to access over 24,500 courses taught by industry experts.

Real-time model training

Real-time model training

From the course: Advanced Data Processing: Batch, Real-Time, and Cloud Architectures for AI

Start my 1-month free trial Buy for my team

Real-time model training

“

- [Instructor] What does model training in real time look like? What are the architectural considerations? Let's understand the constraints and considerations for real-time model training. Model training cannot be done one record at a time. It needs a significant sample size big enough to address all expected variations in input data. This is what makes it hard to do model training in real time as sufficient sample sizes may not be available in a short period of time. So model training can only be done in batches or micro-batches in real time. We can collect data in real time for smaller intervals, like every 30 minutes, and then use that data to build a new model. We can also use windowing to keep a longer interval of data to retrain the model. The training pipeline should have automatic validation, review, approval, and promotion based on predefined criteria so manual intervention is not needed. In real time, it may be a better option to only fine tune the base model so the original…

Contents