barmpy is the Python implementation of Baeysian Additive Regression Models, a generalization of BART, currently being researched [1]. We hope this library is useful for practictioners, enabling Bayesian architecture search and model ensembling.

Skeleton repo adapted from BartPy.

Check out the Tutorial

barmpy is on PyPi! Install the latest released version with pip install barmpy. barmpy also strives to be compatible with sklearn and easy-to-use. If you have arrays of target data, Y, and input data, X, you can quickly train a model and make predictions using it. barmpy currently supports ensembles of neural networks for both regression and binary classification. See below for simple examples.

from sklearn import datasets, metrics
from barmpy.barn import BARN, BARN_bin
import numpy as np

# Regression problem
db = datasets.load_diabetes()
model = BARN(num_nets=10,
          random_state=0,
          warm_start=True,
          solver='lbfgs',
		  l=1)
model.fit(db.data, db.target)
pred = model.predict(db.data)
print(metrics.r2_score(db.target, pred))

# Classification problem
bc = datasets.load_breast_cancer()
bmodel = BARN_bin(num_nets=10,
          random_state=0,
          warm_start=True,
          solver='lbfgs',
		  l=1)
bmodel.fit(bc.data, bc.target)
pred = bmodel.predict(bc.data)
print(metrics.classification_report(bc.target, np.round(pred)))

[1] https://arxiv.org/abs/2404.04425