neonet 0.1.1

Neonet is a deep learning tool for building simple to medium sized neural networks, it supports multiple configurations and arguments for training

Neonet

Neonet is a lightweight deep learning library for building small to medium scale neural networks, with small to medium sized datasets .it supports multiple training configurations, mutiple optimization methods, regularization techniques, and different acttivation functions

Installation

pip install neonet

Getting started

from neonet.nn import NeuralNetwork, TrainArg

Creating a neural network

nn = NeuralNetwork(4, [(16, "LeakyReLU"), (8, "LeakyReLU"), (3, "Softmax")])

Training arguments

• batch_size
• learning_rate
• optimizer
• regularizer
• alpha
• lasso
• b1 coefficient
• b2 coefficient
• epochs
• use_decay

training_args = TrainArg( 
batch_size=16, 
learning_rate=0.001, 
optimizer="adams_loss", 
loss="MSE", epochs=500, 
logging_steps=100, 
use_decay=True )

to train:

nn.train(X_train, y_train, training_args=training_args, eval_dataset=[X_test, y_test], check_loss=True)

Features

-	Dense neural network architecture
-	Multiple activation functions
-	Mini-batch training
-	Learning rate decay
-	Model evaluation during training
-	Configurable optimizers and regularizers