Biolith: Bayesian Ecological Modeling in Python

Biolith is a Python package designed for Bayesian ecological modeling and analysis with a focus on occupancy modeling. It has similar goals to Unmarked and spOccupancy, but is written in Python and uses NumPyro and JAX to enable rapid model fitting and iteration.

Features

• Hackable: Models are easy to understand and implement, no likelihood derivations needed.
• Fast: Models can be fit on GPUs, which is fast.
• Familiar: Everything is written in Python, making it easy to integrate into existing pipelines.

Installation

Install Biolith using pip:

pip install biolith

Accelerator support

For NVIDIA GPU support, check your CUDA version with nvidia-smi and install the corresponding JAX variant:

# CUDA 13.x
pip install biolith[cuda13]

For AMD GPUs:

pip install biolith[rocm]

For Google Cloud TPU VMs:

pip install biolith[tpu]

Usage

Here is a simple example using simulated data to get you started:

from biolith.models import occu, simulate
from biolith.utils import fit

# Simulate dataset
data, true_params = simulate()

# Fit model to simulated data
results = fit(occu, **data)

# Compare estimated occupancy probability to the true mean occupancy
print(f"Mean estimated psi: {results.samples['psi'].mean():.2f}")
print(f"Mean true occupancy: {true_params['z'].mean():.2f}")

Real-world Example

To see a real-world example on camera trap data, see this Jupyter Notebook from the EFI Statistical Methods Seminar Series or

Documentation

API Documentation and examples are available here.

Development

Make sure to install development dependencies, e.g. using:

pip install -e ".[dev]"

Run pytest to run unit tests.

Run scripts/format.sh to format the codebase. Execute scripts/check.sh to run isort, docformatter and black in check mode along with pylint and pyright.

To install the pre-commit hook for formatting and code linting, run:

./scripts/install_precommit.sh

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For questions or feedback, please open an issue or email haucke@mit.edu.

Acknowledgements

This work was supported by the AI and Biodiversity Change (ABC) Global Center, which is funded by the US National Science Foundation under Award No. 2330423 and Natural Sciences and Engineering Research Council of Canada under Award No. 585136. This work draws on research supported in part by the Social Sciences and Humanities Research Council.