Looking Into the Crystal Ball: Using Google's Prediction API
summary
Demos using Google's Prediction API with CSVs on Cloud Storage claim useful, pragmatic results: classification and regression on real datasets, with models trained from CSVs and retrained via incremental inserts; typical training takes an hour for ~16MB and scales to many hours for hundreds of megabytes, inserts cap around 2.5MB so larger sets go via storage. Examples predict NiceRide end stations, estimate NYC cab fares from lat/long, and run sentiment on the State of the Union; models produce plausible answers but remain a black box (one output only), risk overfitting, and offer hosted models for sale though the marketplace is thin.
AI-written summary · about
chapters (10)
- 0:00Prediction API introduction
- 3:14Prediction paradigms
- 8:08Prediction API features
- 14:10Training a model
- 21:17Demo: Bike destinations
- 27:59Q&A on platform
- 29:28Demo: Cab ride costs
- 36:28Demo: Speech sentiment
- 42:01Overfitting and practical applications
- 43:44Final Q&A
related talks
43mMachine Learning at Scale
43mGetting started with AI for free
37mWhich Serverless Prediction Platforms on Google Cloud Platform?
14mPredicting Football Matches Using Data
01h 05mAzure Machine Learning for the Developer
36mCloud AI Services: What they are and how to use them
59mMachine Learning and Big Data Magic with Node.js and Google Cloud
28mUse case in demand forecasting