Neural Network from scratch

IPython notebook illustrates experiments with a 4 layer network.

Notebook setup-

# Create and activate new virtual environment (optional)
virtualenv venv
source venv/bin/activate
# Run notebook
jupyter notebook mlnn.ipynb

  Please refer mlnn.py for model code.

Methods

• fprop(train_data_X) : Forward propagation using train_data_X.

• bprop(train_data_X, train_data_y, regularization_lambda, epsilon) : Backward propagation using (train_data_X, train_data_y).

• predict(train_data_X) : Make predictions using train_data_X.

• train(train_data_X, train_data_y, regularization_lambda, epsilon, num_pass, print_loss) : Train network using (train_data_X, train_data_y).

• visualize_preds(train_data_X, train_data_y) : Plot classification graph.

• animate_preds(train_data_X, train_data_y, regularization_lambda, epsilon, num_pass=35000) : Animate training process.

 

Requirements:

• numpy
• sklearn
• matplolib

Please feel free to contact me at ishank269[at]gmail[dot]com to report any issues or you can open an issue here. Fork or clone for your use :)