SJNET 0.1.1

A neural network Framework

SJNET

This repository includes SJNET.py, which contains the Network and Layer classes.

These can be used to create and train custom neural network

Import SJNET

# download SJNET.py then continue
from SJNET import Network ,Layer

A new Network can be initialized by

# provide the dataset X and Y
# each element is Y must be an array [] , Y = [[y1..y1n],[y2..y2n]...[yN..yNn]]

network = Network(X=X,Y=Y,learningRate=0.0002,epoch=1000,errorThresh=3)

Declare or initialize layers with its coresponding arguments

• neuronCount : number of neurons required in the layer

• position : position of layer in network

  • Note : (1 for input layer and -1 for output layer)

• activation : required activation function , avilable("linear","relu")

#Note: position must be (1 for input layer and -1 for output layer)
inputLayer = Layer(neuronCount=2,position=1)
hidden = Layer(neuronCount=10,position=2,activation="linear")
hidden2 = Layer(neuronCount=4,position=3,activation="linear")
output = Layer(neuronCount=1,position=-1,activation="linear")

Add declared layers into the network

#add them in order (inputLayer->first , outputLayer->last)
network.add(layer=inputLayer)
network.add(layer=hidden)
network.add(layer=hidden2)
network.add(layer=output)

Compile or initilise the network setup

network.compile()

Train the model

network.Train()

Save the model

# saved as json
network.save(name="testModel")

Load the model

Model = {}
with open('./savedmodels/testModel.json', 'r') as file:
    Model = json.load(file)

#Loads the  network topology weights and biases 
network.loadNetwork(network=Model)

predict with model

pred_val = network.predict(inputvals=[3.1, 2.5])

How to train and save a model -> Training_and_Saving_Model.py

How to load and use the model -> Loading_Pre_Trained_Model.py

It uses stochastic gradient descent, just in case you are curious.