abil 25.2.17

Aquatic Biogeochemical Interpolation Library

Abil.py ·

Overview

Abil.py provides functions to interpolate distributions of biogeochemical observations using Machine Learning algorithms in Python. The library is optimized to interpolate many predictions in parallel and is thus particularly suited for distribution models of species, genes and transcripts. The library relies on scikit-learn.

Current support (v0.10):

• Random Forest, XGBoost, Bagged KNN

• Continuous data

• 2-phase zero-inflated models

• Hyperparameter tuning and cross-validation

• Automatic feature scaling and one-hot-encoding

• Example SLURM and Singularity scripts

Generally the workflow is as follows:

1. Define the model setup in a model_config.yml file (for an example see /examples/configuration/)
2. Tune the model for the species of interest using tune.py
3. Predict the distribution of each species using predict.py
4. Merge the predictions into a single netcdf and do post processing using post.py

Examples for each step are provided in the respective Jupyter notebooks which can be found in /examples/notebooks.

Directory structure

The recommended directory structure is:

Abil
├── abil
|   └── __init__.py
|   └── functions.py
|   └── post.py
|   └── predict.py
|   └── tune.py
├── dist
|   └── abil-0.0.9.tar.gz
|   └── abil-0.0.9-py3-none-any.whl
├── docs
├── examples
|   └── conda
|       └── environment.yml
|   └── configuration
|       └── 2-phase.yml
|       └── classifier.yml
|       └── regressor.yml
|   └── data
|       └── prediction.csv
|       └── targets.csv
|       └── training.csv
|   └── hpc_example
|       └── hpc_post.py
|       └── hpc_predict.py
|       └── hpc_tune.py
|       └── post.sh
|       └── predict.sh
|       └── README.md
|       └── tune_KNN.sh
|       └── tune_RF.sh
|       └── tune_XGB.sh
|   └── notebooks
|       └── tune.ipynb
|       └── predict.ipynb
|       └── post.ipynb
|   └── singularity
|       └── singularity.sif
├── studies
|   └── devries2024
├── tests
├── README.md
├── pyproject.toml
└── README.md

Installing the package:

Install the dependencies in a new environment:

conda env create -f package_save_path/examples/conda/environment.yml

Activate the new environment and install Abil:

conda activate abil-env

python -m pip install package_save_path/Abil/dist/abil-0.0.9.tar.gz

Updating the package:

If you have changed the scripts and want to update the package, a new version can be build.

CD to the planktonSDM directory, then run:

python3 -m build

Note: if you want to change the version name of the package, this can be changed in:

pyproject.toml

Running the model on a hpc cluster

See: /examples/hpc_example/README.md