A Keras-Like Deep Neural Network Abstraction

An OOP Deep Neural Network using a similar syntax as Keras with many hyper-parameters, optimizers and activation functions available.

Installation

Just clone the repository to your computer to get started:

git clone [email protected]:kinoute/Elyane.git
cd Elyane

The only dependencies to make this neural network work are "numpy" and "pdoc3". They can be installed directly from the main directory if you don't have them already:

pip3 install -r requirements.txt

Three examples are available: mnist.py, fashion_mnist.py and xor.py.

Features

Batches

• Batch Gradient Descent
• Mini-batch gradient descent
• SGD

Layers

• Full-Connected Layer
• Softmax Layer
• Dropout Layer

Optimizers

• Momentum
• RMSprop
• Adam Optimization
• Amsgrad optimization

Losses

• Mean Squared Error (MSE)
• Mean Absolute Error (MAE)
• Cross Entropy
• Multi-Class Cross Entropy

Regularizers

• L1 Regularization
• L2 Regularization
• Dropout

Activation functions

• Sigmoid
• TanH
• Relu
• Leaky Relu
• Softmax

Tools

• One hot encoding
• Normalization of images