These are the steps for releasing a new version of the system:
- Create a new conda environment and clone the latest version of the system:
conda create --yes -n label-sleuth python=3.9; conda activate label-sleuth; git clone [email protected]:label-sleuth/label-sleuth.git; cd label-sleuth; pip install -r requirements.txt
- Run the tests:
python -m unittest
For testing the system, either clear your browser cache or add --port <PORT_NUMBER> to the command with a different port number from what you usually use.
- Assuming you have some existing workspaces/categories in the default output directory:
start the system with
python -m label_sleuth.start_label_sleuth --port <PORT_NUMBER>, open the system, label some elements in an existing category from different views (document, Label next, etc.) - Delete or backup the output directory (usually
~/label-sleuth) - Start Label Sleuth with
python -m label_sleuth.start_label_sleuth --load_sample_corpus wiki_animals_2000_pages --port <PORT_NUMBER>to test loading the sample corpus - Upload an additional corpus
- Create a workspace
- Create a category, label 2-3 elements.
- Use the search and label the search results to reach >20 labeled elements (continue playing with the system while waiting for a model)
- Go to "label-next", label an element, jump to a document
- Go to "positive predictions", label an element, jump to a document
- Label an additional ~20 elements to train another new model
- Download the labeled data
- Create a new workspace
- Import the downloaded model to the new workspace and make sure the same elements appear on both workspaces
Only if all the above steps are in order, create a new PyPi release by pushing a new version tag: create a new GitHub release, create a new tag with the new version number. Describe the changes and publish.
Within a few minutes, you should see that the new version appears in https://pypi.org/project/label-sleuth/