A bare-bones implementation of deep learning neural networks using numpy. toynn is not comparable to powerful deep learning libraries
such as as Tensorflow, Keras, PyTorch and many others. It's goal is not to train production quality deep learning
neural networks. It is a personal project born out of a desire to, understand deep learning neural networks by coding them up and gain insights into various tricks that make them so powerful. I also used it to gain
some practice in prototyping and shipping machine learning algorithms utilzing python/git/CI ecosystem.
- Sigmoid and ReLU layers
- Batch Gradient Descent
- Constant, He and Xavier weight initializations
- Dropout
- L1 and L2 Regularization
- Mini-batch gradient descent
- MIT license
The implementation of deep learning neural networks is this package is inspired from lectures and codes in Andrew Ng's Deep Learning Specialization on Coursera.