Choose the right algorithm to help you build better ML models
summary
Amazon SageMaker demonstrates choosing from 17 built‑in algorithms and shows k‑NN production techniques using the Faiss library for indexing, Faiss IVF flat clustering and Faiss IVF PQ product quantization to shrink memory. The demo geocodes addresses with GeoPy, trains a k‑NN model with k=11 in 171 billable seconds and returns neighbors and distances, yielding a mean absolute error around $500K on California house prices. The notebook highlights subsampling, random projection (FJLT) and quantization tradeoffs.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:50Overview
- 6:05Linear regression
- 7:09Gradient boosting
- 8:17K nearest neighbors
- 14:09Basic operation
- 15:31Subsampling
- 21:05Product Quantization
- 23:15Demo
- 30:02Conclusion
related talks
30mChoose the right machine learning algorithm in Amazon SageMaker
39mAnd Then There Are Algorithms
19mDeveloping large-scale machine learning algorithms on Amazon SageMaker
44mPragmatic Machine Learning in the Cloud
26mGeospatial Machine Learning at Scale
40mTools and Tricks from a Pragmatic Data Scientist
31mTrain and tune ML models to the highest accuracy using Amazon SageMaker
47mBuilding ML practices to address the top four use cases