- Development of features and code works in different branches, using Git.
- Everything should be Merged into a
masterbranch, with the appropriate code review.- To make life simple for everyone, keep Merge Requests brief and do them often.
- This will ensure code quality and that it won't take much time of the reviewer.
- When making experiments, you can either do locally or in a server by pulling the git repo.
- Remember to always make experiments in a different branch, so you can keep your code organized
- Don't commit data or binaries to git, as they can be really heavy.
- When the experiment is good enough, you can open an MR for review.
- You can put on the description the workspace.
- All the experiments should be saved in the
workspace/folderworkspace/data/keeps the train and test data for your experimentworkspace/models/keeps the trained models, preprocessing objects and any pickles- You can save this files using
make save VERSION=xand everything inside the workspace folder will be saved onhttps://ufc.lerax.me/models/asd/x/
- By using
make release VERSION=xyou'll:- test your code
- save the files inside
workspace/tohttps://ufc.lerax.me/models/asd/x/ - generate a git tag
xfor the current commit - push the tag
xto gitlab - then the Continuous Integration will build a Docker image with the trained binaries and publish the Docker to the registry, ready for being deployed