Reproducible research using Bazel
summary
A Bazel-backed research platform that mixes Docker pragmatism with hermetic Bazel builds to make experiments reproducible and cloud-scalable. It generates project boilerplate with make, packages code with Bazel rules (including a custom extended_python_image), stamps outputs with workspace_status git hashes for traceability, and runs hyperparameter sweeps via a DOE GKE rule to launch thousands of experiments from ~15 lines of BUILD/DOE config, targeting GCP with git and Docker as prerequisites.
AI-written summary · about
chapters (8)
- 0:00Reproducible research with Bazel
- 1:05Reproducible research challenges
- 2:51Uber Elevate research platform
- 6:47Research platform architecture
- 8:55Research workflow pain points
- 12:26Pragmatism vs hermeticity
- 13:30Traceability in research outputs
- 14:13Future plans
