Carlos Guestrin, CEO of Dato Inc. @ MLconf SEA
summary
Shows that a columnar SFrame compresses real datasets far smaller than gzip and lets 931 million-row, 950-column biology tables run aggregations in ~2 seconds and composite feature creation in ~8 seconds on one desktop. Demonstrates graph and image benchmarks (PageRank on Common Crawl: ~9 minutes/iteration on one AWS machine) and a deployed real-time predictive service that copies a model to S3 and boots it on 3 EC2 nodes. Uses Python APIs, REST, and open-source SFrame/SGraph; the deployed recommender works and supports zero-downtime updates.
AI-written summary · about
chapters (5)
related talks
27mData @scale Seattle - Yucheng Low
30mPython based predictive analytics with GraphLab Create
29mTed Willke, Senior Principal Engineer & GM, Datacenter Group, Intel
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
01h 12mGPU-Accelerated Data Analytics in Python
41mBig Data Serving: The Last Frontier. Processing and Inference at Scale in Real Time
39mAccelerating Data Science with RAPIDS
33mBuilding smarter ML and AI models and making them more accurate