Gradient Free Deep Learning

Gradient Free Deep Learning (GFDL) is a Python package for performing machine learning classification and regression tasks using neural networks that do not require backpropagation.

It specifically provides:

• A scikit-learn compatible estimator for classification (gfdl.model.GFDLClassifier)

• A scikit-learn compatible estimator for regression (gfdl.model.GFDLRegressor)

• A scikit-learn compatible estimator for ensemble classification (gfdl.model.EnsembleGFDLClassifier)

• A small set of neural-network style activations (gfdl.activations)

Getting started

Install:

pip install gfdl

Quick example

import numpy as np
from gfdl.model import GFDLClassifier

rng = np.random.default_rng(seed=42)
X = rng.standard_normal(size=(100, 10))
y = (X[:, 0] > 0).astype(int)

clf = GFDLClassifier()
clf.fit(X, y)