DeepNeuralBranchNet 0.0.14

Neural network module with branching output

Neural network with branching output

This package can be used to train neural network for classification with option of multiple outputs by branching at different layers.

Features available

• Customizable layers with option of selecting activation function
• No limits on adding number of layers and number of neurons in each layers
• By default gradient descent for optimization but in further version, other optimization methods are to be included
• Multiple outputs can be selected at different layers with just one method call

Requirements

This package is developed at very low-level python coding so there is no much requirement other than numpy library. Numpy

Installation

To install a stable version, use the following command

pip install DeepNeuralBranchNet

Example of how to use

Import module

from DeepNeuralBranchNet.neuralnet import NeuralNet

Initiate the neural network object class

example=NeuralNet()

Add input by passing number of nput features as parameter

example.add_input(input_length=16)

Add layer sequentially by passing number of neurons and activation function

example.add_layer(10,activation_function="relu")

If you want to add multiple output(or opting for branching), then trigger branching by following line of code and then keep adding layers

example.do_branching()

Run model to update parameters such as weights and bias and for training.

• input_array : Input matrix in the form of array
• output_array : Output matrix in the form of array
• number_of_iterations : Total number of times you want to run back propagations to update weights
• multiple_output : If you have branches then set it to True

example.run_model(input_array,output_array, number_of_iterations=1000, learning_rate=0.001, multiple_output=False)

Finally predict by calling predict method

example.predict()