Push straight to prod: API development with R and Tensorflow
summary
Built a containerized API pipeline that keeps models in R using keras and plumber, Docker and TensorFlow, with a Rocker base image, Miniconda to replace Anaconda, and Apache2 as an HTTPS proxy. Initial images hit about 5 GB so dependencies are pruned; deployed model survives 50x load testing and runs in production powering Apple Business Chat routing decisions. Model is a shallow CNN (one conv layer + 1–2 dense layers) that outperforms an RNN on classification; code and Docker repo are open sourced.
AI-written summary · about
chapters (4)
related talks
46mDev Ops meets Data Science Taking models from prototype to production with Docker
01h 18mPutting Tensor Flow Models Into Production
33mDeploying and Scaling Spark ML and Tensorflow AI Models
15mA Fiery approach to R web services
32mDeploying Python models to production
33mHow you really get your data science models into production the cool way!
17mR & Python playing nice, in Production
01h 18mPutting Tensorflow Models into Production