Simple Multilayer Perceptron from scratch in Python
The Relu activation function is used in every hidden layer, the output layer can use other activation
To use this package, you need to have numpy installed. If you don't have numpy installed, you can install it using pip:
pip install numpy
Clone repo
git clone https://github.com/IamKrill1n/Deep-neural-networks-from-scratch.git
Check out the example.ipynb file for further instructions
Regression example
from my_dnn import model, optimizers, loss, metrics
# layer_dims = [number_of_feature_X, hidden_layer1, hidden_layer2, ..., hidden_layerL-1, output_layer]
my_model = model.SimpleMlp(layer_dims=[X_train.shape[1], 32, 32, 16, 1], output_activation='relu')
my_model.compile(optimizer = optimizers.RMSprop(), loss=loss.MSE(), metrics=metrics.RMSE())
# X_train must be a numpy array of shape (number_of_examples, number_of_feature)
# y_train must be a numpy array of shape (number_of_examples, )
my_model.fit(X_train, y_train, validation_data = (X_val, y_val), epochs=50, batch_size=32, verbose=0)
# X_test must be a numpy array of shape (number_of_examples, number_of_feature)
y_pred = my_model.predict(X_test)Classification example
from my_dnn import model, optimizers, loss, metrics
my_model = model.SimpleMlp(layer_dims = [X_train.shape[1], 32, 32, 16, number_of_class], output_activation = 'softmax')
my_model.compile(optimizer = optimizers.Adam(), loss = loss.CategoricalCrossEntropy(), metrics = metrics.SparseCategoricalAccuracy())
my_model.fit(X_train, y_train, validation_data = (X_val, y_val), epochs = 50, batch_size = 32, verbose = 0)
y_pred = my_model.predict(X_test)Inspired by Coursera Deep learning Specialization https://www.coursera.org/specializations/deep-learning