TensorFlow Wide & Deep: Advanced Classification the easy way
summary
A concrete demo builds a TensorFlow wide and deep classifier with canned Estimators, using feature_columns (bucketed age, crossed categories, and embedding dim 8) on the 1994 Adult census (about 32k/16k split), trains in seconds (demo shows ~45s) and reaches roughly 83% accuracy. The model exports an exported SavedModel and runs under TensorFlow Serving or Google Cloud ML Engine, so the pipeline actually trains, exports, and serves.
AI-written summary · about
chapters (28)
- 0:00TensorFlow Wide & Deep intro
- 1:05Classification with structured data
- 2:11Training vs prediction
- 3:15TensorFlow overview
- 5:42Higher-level abstractions
- 6:46Canned estimators
- 7:50Wide & Deep model
- 13:28Why Wide & Deep simplifies
- 13:50Census income dataset
- 16:01Setting up the Jupyter notebook environment
- 17:03Import & preprocess data
- 20:10Explore dataset
- 20:54Feature columns
- 23:20Transform features
- 24:42Wide & deep feature groups
- 25:45Build Wide & Deep model
- 27:51Train on census data
- 29:37Monitor training loss
- 30:42Evaluate accuracy
- 31:44Iterate & experiment
- 32:31Export model
- 33:55Deploy with TF Serving
- 35:02Cloud ML Engine serving
- 36:27Export to Cloud Storage
- 37:31Cloud prediction endpoint
- 38:12REST API predictions
- 38:54Flexible train/serve
- 39:15Resources & next steps
related talks
51mTensorFlow Wide & Deep: Advanced Classification the easy way
43mMachine Learning at Scale
01h 06mDeep Learning with TensorFlow
37mSimplifying Training Deep & Serving Learning Models...using Tensorflow
01h 07mDeep Learning with TensorFlow
57mTensorflow, deep learning and modern convolutional neural nets, without a PhD
55mTensorflow: Learning Machines
59mTensorFlow in Three Sentences