How to Track your Development Process with DVC
summary
A live fruit classifier project uses 982 images to show concrete results: an overfitting convolutional model, a reproducible pipeline built with DVC pipelines that stores binaries to an S3-style remote via Git metadata, and deployment to TF Serving for REST inference. Automated reproduction retrains and redeploys when a config adds walnuts; the live demo sometimes misclassifies walnuts as bananas but the full workflow — data, model, hashes, artifacts — is built and runnable.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:02How did we start
- 5:23Data Engineers
- 5:59Data Scientist
- 9:32Data Engineering
- 10:42DVC Stack
- 16:55Pipelines
- 17:59TF Serving
- 18:59Fruit Classifier
- 19:30Version Control
- 21:10Data Pipeline
- 21:39Preprocessing
- 23:26Train
- 26:34Reproduction
- 28:59My use case
- 29:55Recap
related talks
33mFlexible ML Experiment Tracking System for Python Coders with DVC and St...
16mMake your data science workflows reproducible
31mBecoming a Pokémon Master with DVC: reproducible machine learning experiments
35mData versioning in machine learning projects
01h 32mHow to easily set up and version control your Machine Learning Pipelines
02h 10mBuild End-To-End MLOps Platform with Open Source DVC Ecosystem
42mApplications in ML Drug Discovery pipelines
30mHow to make continuous integration work with machine learning