Building Machine Learning Pipeline Using Aerosolve
summary
Aerosolve is an open source pipeline used at Airbnb to power Price Tips and Smart Pricing, training generalized additive models (piecewise linear splines + linear terms) with heavy feature engineering and feature transformation config files to generate market×price and lat/long quantized features. Training runs on Spark (Scala), core scoring and transforms run in Java so models load from S3 for real-time Java scoring. Production A/B tests show measurable booking lift (~2% relative uplift, with cannibalization caveats); tooling exposes learned feature curves for debugging.
AI-written summary · about
chapters (9)
- 0:00Introduction to Airbnb pricing
- 3:26Smart pricing at Airbnb
- 10:49Why Aerosolve for pricing
- 15:35Feature debugging with Aerosolve
- 18:03When to use Aerosolve
- 21:58Feature representation in Aerosolve
- 25:58Aerosolve model types
- 28:28Building ML pipelines with Aerosolve
- 32:25Q&A session
related talks
40mMachine Learning
34mComplex Machine Learning Pipelines Made Easy
17mLaying the foundation for deep learning
44mMachine learning @ booking.com
33mA new way to profit: Predictive analytics and dynamic pricing
30mDriving a Higher Level of Customer Support with Machine Learning
22mBefore the Model: How Machine Learning Products Start...
11mAWS On Air ft. BMW & BCG