- Interpretable ML results (sensitivity analysis)
- Scikit-learn compliant
joblibfile with trained model (one file per model) - Provide GUIs to analyze the datasets and ML models whenever possible
- Classification should be flexible in resolution
- Push to the repo often!
- Python code (as PEP8 compliant as possible)
- Document your code in MD files and comments
- Git commits start with the 3 letter code of the project (PYF, STP, etc.)
- You can use Jupyter notebooks for exploration but the end result should be a
.pyset of files that can be run from the terminal - These
pyfiles will be called from abashcommand to create pipelines - Functions definitions should have their own separate file(s)
- Have separate files for: data cleaning, training, testing, evaluation
- Have paths as clearly stated variables that can be changed easily
- Datasets will be synched through Mega for the time being
- Have a file with the re-scaling constants
- Return some type of uncertainty estimate