Logistic Regression with a Neural Network Mindset 🧠💻

This project demonstrates a neural network mindset in approaching logistic regression, applying it to a binary classification problem (cat vs. non-cat images). It builds a logistic regression classifier using a deep learning framework, with Python and essential libraries. The project also introduces neural network concepts such as forward propagation, cost function, and gradient descent.

Credit: This project is inspired by the course Deep Learning Specialization by DeepLearning.AI. Special thanks to the creators for providing excellent resources on deep learning! 🎓💡

🚀 Project Overview

Problem Set

The objective is to classify images as either containing a cat 🐱 or not containing a cat 🙅‍♂️. The dataset consists of:

• Training set: 209 images
• Testing set: 50 images
• Each image size: 64x64x3 (RGB)

Key Steps:

1. Data Preprocessing:

   • Normalize the dataset by scaling pixel values to [0, 1] range.
   • Reshape the image data into vectors for further processing.

2. Logistic Regression Model:

   • Implement the logistic regression model using a neural network mindset.
   • Build helper functions for sigmoid activation, initializing parameters, and propagation.
   • Use gradient descent for optimization.

3. Evaluation:

   • Use forward and backward propagation to compute the cost and gradients.
   • Optimize the model parameters (weights and bias) using gradient descent.
   • Assess the performance on the test set.

Libraries Used:

• numpy for numerical computations
• matplotlib for data visualization
• PIL and scipy for image processing
• h5py for handling datasets

🛠️ How to Use the Code

Prerequisites:

• Python 3.x
• Jupyter Notebook (optional for inline plotting)

Setup Instructions:

1. Clone the repository:

   git clone https://github.com/MohammedSaqibMS/Logistic-Regression-as-a-Neural-Network.git

2. Install the necessary Python libraries:

   pip install numpy matplotlib Pillow scipy h5py

3. Run the notebook or the script:

   jupyter notebook logistic_regression_neural_network.ipynb

📊 Dataset Overview

• Classes: Binary (cat/non-cat)
• Train set: 209 images
• Test set: 50 images

Example of dataset exploration:

plt.imshow(train_set_x_orig[10])  # Display an image from the dataset
print(f"y = {train_set_y[0, 10]}, it's a '{classes[np.squeeze(train_set_y[:, 10])].decode('utf-8')}' picture.")

💡 Key Concepts

1. Sigmoid Activation Function:

   def sigmoid(z):
       return 1 / (1 + np.exp(-z))

2. Cost Function: The negative log-likelihood cost function used for logistic regression.

3. Gradient Descent Optimization: Optimize weights w and bias b by updating them iteratively based on the computed gradients:

   w = w - learning_rate * dw
   b = b - learning_rate * db

🎯 Results

The model achieves reasonable accuracy in predicting whether the images contain cats. The cost reduces significantly during training, thanks to gradient descent optimization.

Example of Cost Output:

Cost after iteration 0: 0.693
Cost after iteration 100: 0.218

🤝 Acknowledgments

This project is based on the coursework from the Deep Learning Specialization by DeepLearning.AI. Many thanks to Andrew Ng and the entire team for their contributions to AI education! 🙏