daniel7an / nn_from_scratch

Neural Network From Scratch in Python (Digit classification).

Neural Network From Scratch

This project involves implementing a neural network from scratch and training it on the MNIST dataset for handwritten digit classification. The code is developed based on the principles outlined in the book "Neural Network from Scratch".

This neural network implementation consists of the following components:

• Layer_Dense: Implements a dense layer with random initialization of weights and zero biases.
• Activation_ReLU: Implements the ReLU activation function.
• Activation_Softmax: Implements the softmax activation function for multiclass classification.
• Loss: Defines the base class for loss functions.
• Loss_CategoricalCrossentropy: Implements the categorical cross-entropy loss function.
• Activation_Softmax_Loss_CategoricalCrossentropy: Combines the softmax activation and categorical cross-entropy loss.
• Optimizer_SGD, Optimizer_AdaGrad, Optimizer_RMSprop, Optimizer_Adam: Implement various optimization algorithms for updating weights and biases during training.