CourseAI / ML Engineer: The Complete Course · Module 4: Deep Learning · part 27 of 93
Part 27 · Module 4: Deep Learning

Lesson 3: Architectures: matching the network to the data

33 min read·
Module 4 · Lesson 3 · About 55 minutes · Needs: lessons 1 and 2 (a training step, losses, normalisation), and the Kestrel data (course-guide/06). GPU optional: everything runs on a CPU.

The situation

Week 4 at Kestrel, day four. Kestrel's data comes in several shapes: a parcel is a table row, a depot's daily volume is a sequence, a customer message is text, and Hannah asks whether photos of damaged parcels could be sorted one day. Adaeze wants one page: "Which network for which data, and what does each cost?"

Running examples: message M0001, "hiya you said tomorrow 4 my parcel. it is now late! Regards" (simulated, intent delivery_late), and test picture 0 of Fashion-MNIST, an ankle boot, standing in for a parcel photo.

WHERE THIS FITS The Kestrel system this lesson SHARED SERVICES Depot scans parcels, routes Ingestion nightly batch Validation Pandera checks Data lake Parquet, DuckDB Experiments MLflow runs Training LightGBM, Optuna Online store Feast, Redis Features point in time Gates vs the champion Registry staged versions Serving API FastAPI, ONNX Consumers vans, customers Retraining trigger, budget Alerting on call Monitoring drift, Evidently Late labels a day later Orchestrator Prefect flows Versions DVC, Git CI/CD GitHub Actions Containers Docker images
This lesson stays inside training (blue): which network suits which kind of Kestrel data; the rest of the system is grey for now.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.

Buy this course · ₹399See what is inside

Or open everything with a plan