To install from pypi, run
pip install aind-data-schema-models
To install from source, in the root directory, run
pip install -e .
To develop the code, run
pip install -e .[dev]
Add new classes to the _generators/models/*.csv
files.
Run ./run_all.sh
in the top-level folder.
The model class files, brain_atlas.py
etc, are auto-generated. You should never need to modify the class files directly.
Instead, take a look at the jinja2
templates in the folder _generators/templates
. The filename of the template is used to pull the corresponding .csv
file and populate the data
DataFrame. In the template you can pull data from the various columns and use them to populate each of the fields in your class.
To re-build all the models, run the run_all.sh
bash script in the root folder, which loops through the template files and runs them through the generate_code
function.
There are a few special cases, e.g. if data are missing in columns they will show up as float: nan
. See the organizations.txt
template for examples of how to handle this.