A general framework for building and training constructive feed-forward neural networks. Provides an implementation of sibling-descendant CCNN (Cascade-Correlation) [1,2] with extendable wrappers to tensorflow, keras, scipy, and scikit-learn. Also supports custom topologies, training algorithms, and loss functions [3, 4].
The simplest way to install this package currently is to clone the repository and use pip. First, clone the repository:
git clone https://github.com/mike-gimelfarb/cascade-correlation-neural-networks.gitNext, navigate to the folder and use pip
cd cascade-correlation-neural-networks
pip install .We are currently in the process of hosting this project from PyPI, please stay tuned.
The package has been tested using:
- Python 3.7
- Tensorflow 2.3.1
- scikit-learn 0.23.2
- pandas 1.1.3
- scipy 1.5.2
Regression
Classification
Unsupervised Learning
- Fahlman, Scott E., and Christian Lebiere. "The Cascade-Correlation Learning Architecture." NIPS. 1989.
- Baluja, Shumeet, and Scott E. Fahlman. Reducing network depth in the cascade-correlation learning architecture. CARNEGIE-MELLON UNIV PITTSBURGH PA SCHOOL OF COMPUTER SCIENCE, 1994.
- Kwok, Tin-Yau, and Dit-Yan Yeung. "Bayesian regularization in constructive neural networks." International Conference on Artificial Neural Networks. Springer, Berlin, Heidelberg, 1996.
- Kwok, Tin-Yau, and Dit-Yan Yeung. "Objective functions for training new hidden units in constructive neural networks." IEEE Transactions on neural networks 8.5 (1997): 1131-1148.