Neural-Network:

This repository is an ongoing work of implementing neural network in Python.

We have tried to keep the function and variable names similar to scikit-learn and keras. This repository will be updated regularly so stay tuned.

Optimizers available:

1. Stochastic Gradient Descent: use optimizer='SGD'.
   1. Vanilla Backpropogation: set momentum to 0 while creating neural network object.
   2. Classical momentum: set a value of momentum while creating neural network object.
2. Nesterov Momentum: use optimizer='Nesterov' and set a value of momentum while creating neural network object.
3. RMSProp: use optimizer='RMSProp' and set a value of decay while creating neural network object.
4. Adagrad: use optimizer='Adagrad'.
5. Adam: use optimizer='Adam'.

Layers available:

1. Sigmoid
2. Tanh
3. ReLU
4. LeakyReLU

Usage:

Divide your data into scikit-learn format of numpy arrays, i.e. have Xtrain, ytrain ,Xtest and ytest where,

1. Xtrain = trainig examples
2. ytrain = training labels
3. Xtest = test examples
4. ytest = test labels

The below example is for classification.

'Load the libraries'
from neuralnetwork.NeuralNetwork import NeuralNetwork
from neuralnetwork.Layers.layers import *
from neuralnetwork.metrics import *
from neuralnetwork.preprocessing import label_encoder

'Lets make a neural network with 2 hidden layers with sigmoid activation'
'Define your own arguments for neural netowrk, this is just an example'
nn = NeuralNetwork(alpha=0.01, epoch=300, criteria='cross_entropy', optimizer='SGD',
                 batch_size=100, verbose=False, decay=0.001, momentum=0.0, random_seed=None)
nn.add(Sigmoid_Layer(num_features,output_dim1))    #Input layer
nn.add(Sigmoid_Layer(output_dim1,output_dim2))     #Hidden layer
nn.add(Sigmoid_Layer(output_dim2,numClasses))      #Hidden layer

'Train the neural network of '
nn.train(Xtrain,ytrain)

'predict using the trained neural network'
pred = nn.predict(Xtest)

accuracy = accuracy_score(ytest, pred)

Note:

For regression the output dimension of the last layer should be 1.

Support:

If you are having issues, please let us know. We have a mailing list located at: Neural--Network@google-groups.com

Contribute:

Please feel free to get in touch for contributions. We have a huge list of things to be implemented like support for numba and PyCUDA.

License:

The project is licensed under BSD-3-Clause.