shamohamin / Neural_Network

Neural Network From Scratch Using Numpy And Numba.

Neural Network

In this repository, I have implemented the Neural Network using Numpy, Numba packages.

Using Nerual Network

For using implemented Neural Network your input shape must be (number_of_features, number_of_inputs) and then make your desired layer with a specific activation function and then use SGD or BGD Network for training your data.

Supported Activation Functions(activation package):

• sigmoid
• linear
• tanh
• relu

Supported Metrics Functions(metrics package):

• mean squared error(mse)

Supported Optimization Functions:

• momentum
• SGD
• BDG
• mini batch gradient decent

Supported Preprocessing Functions:

• MinMaxScaler
• Standard Scalar

start

In main.py file i have made 500 dummy data for predicting humps function.

    python3 -m venv env
    source env/bin/activate
    (env) pip install -r requirements.txt
    python main.py

screen shots

training on humps function:

• last 1000 epochs of training on Batch Network using 500 data
• output after training

todo list

• implementing regularization (l1(ridge), l2(lasso), dropout)
• implementing more loss functions such as binary cross entropy, ... .