Automated Machine Learning with ML.NET
summary
ML.NET AutoML trains on the 8x8 EMNIST-style dataset of ~70,000 images, runs time-boxed experiments (120s produced 9 candidate models), selects a LightGBM multi-trainer hero model, and yields about 96.5% test accuracy (training ~98.3%); a 20‑minute run nudges test accuracy to ~96.8% and produced a FastTree OVA ensemble. The chosen model serializes to a compact binary (~450KB model) that deserializes into a prediction engine and correctly classifies sample 7 and 1 inputs.
AI-written summary · about
chapters (11)
- 0:00Introduction and overview
- 1:45What is AutoML
- 3:33Machine learning SDLC
- 6:20AutoML in the SDLC
- 10:53Algorithm selection challenges
- 20:03AutoML vs data scientists
- 21:06ML.NET demo setup
- 22:35Code walkthrough and training
- 32:51Model evaluation and metrics
- 37:49Other AutoML frameworks
- 40:40Resources and closing
related talks
59mMachine Learning in .NET (ML.NET)
52mML.NET - Machine Learning made for .NET
42mStep into Machine Learning .NET
57mMachine Learning Made Easy – An Introduction to ML.NET
59mMachine Learning simplified for Developers with ML.NET
56mMachine Learning for .NET developers
46mMachine Learning for .NET Developers
45mML: ML.NET: deep dive into dev AI