Skip to content

README.md

Latest commit

 

History

History
45 lines (34 loc) · 1.36 KB

README.md

File metadata and controls

45 lines (34 loc) · 1.36 KB

🧠 Neural Network from Scratch - Fashion MNIST Classifier

This repository contains a full implementation of a deep neural network from scratch using NumPy, trained on the Fashion-MNIST dataset. The goal is to classify clothing images into one of 10 categories without using high-level deep learning libraries like TensorFlow or PyTorch.

📁 Dataset

  • The dataset is based on Fashion-MNIST, which is a drop-in replacement for MNIST but contains fashion product images.
  • Each image is 28x28 grayscale pixels.
  • There are 10 classes:
    • 0: T-shirt/top
    • 1: Trouser
    • 2: Pullover
    • 3: Dress
    • 4: Coat
    • 5: Sandal
    • 6: Shirt
    • 7: Sneaker
    • 8: Bag
    • 9: Ankle boot

🧩 Features Included

✅ Data Preprocessing

  • Normalization: Pixel values are divided by 255.0 to scale them between 0 and 1.
  • Standardization: Each feature is zero-centered and scaled to unit variance.
  • One-hot encoding: Labels are converted from scalar integers to binary class matrices using OneHotEncoder.

✅ Activation Functions

  • ReLU (Rectified Linear Unit) and its derivative
  • Sigmoid and its derivative
  • Softmax for multi-class classification in the output layer

✅ Network Architecture

The architecture is flexible and defined by:

layers = [784, 512, 256, 128, 10]
activation = ['ReLU', 'ReLU', 'ReLU', 'softmax']