Eleko / FNN-Builder

A simple feedforward neural networks builder

FNN Builder

A simple NumPy feedforward neural networks builder

Usage

Build

model = FNN(lossFunction = Loss.CE, inputs = 4)
model.normalize(x = True, y = False)
model.addLayer(activation = Activation.RELU, neurons = 4)
model.addLayer(activation = Activation.SIGMOID, neurons = 3)
model.addLayer(activation = Activation.SOFTMAX, neurons = 3)

Train

model.train(x_train, y_train, epochs = 10000, rate = .01, batch_size = 5)

Predict

model.predict(x_test)

Test

model.test(y_test)

y_train, y_test : pandas.core.series.Series
x_train, x_test : pandas.core.frame.DataFrame

---

Usage | Features | Requirements | Notebook

Features

Activation functions

• Sigmoïd Activation.SIGMOID
• ReLU Activation.RELU
• Softmax Activation.SOFTMAX

Loss functions

• Categorical Cross Entropy Loss.CE
• Mean Squared Error Loss.MSE

Normalization

• Rescaling (min-max normalization)

Gradient-based Optimization

• Stochastic Gradient Descent : batch_size = 1
• Mini-batch : batch_size = n
• Batch : batch_size = len(x_train.index)

Requirements

• numpy
• pandas
• matplotlib