Adding on-device recommendations to your app using TensorFlow and Firebase
summary
Builds an on-device recommender from Firebase Analytics events exported to BigQuery, transforms events into per-user timelines (skips sequences under three IDs, timeline padded to 100 IDs), trains a convolutional encoder in Colab and runs roughly 100 epochs for the demo, exports a TensorFlow model to TensorFlow Lite, uploads it via Firebase ML, and downloads it to the app where a TFLite interpreter returns movie IDs with confidence scores for inference.
AI-written summary · about
chapters (14)
- 0:00On-device recommendations intro
- 2:10Firebase Analytics setup
- 4:00Export Firebase to BigQuery
- 5:26Import dataset to BigQuery
- 6:55Colab service accounts
- 7:16Clone and prep TensorFlow model
- 8:21Process BigQuery data
- 9:06Pre-processing data for model training
- 10:13Train model in Colab
- 10:55Export model to TFLite
- 11:16Deploy with Firebase ML
- 11:39Integrate into mobile app
- 12:44Running inference and generating recommendations
- 13:28Summary and next steps
related talks
29mPushing Machine Learning into your app When TensorFlow and MLKit rule the world
42mCreating Smarter Applications
19mOn-device machine learning: TensorFlow on Android
47mMachine Learning on Android Demystified
36mMachine Learning for Mobile Developers
11mPersonalizing your app in real-time: Powered by Firebase and ML
14mDo you want an easy way to add Machine Learning into your app?
17mDeploy a custom Machine Learning model to mobile