Deep Learning & Machine Learning Algorithms

Welcome to my repository for Deep Learning and Machine Learning algorithms! 🚀

This repository serves as a personal playground and portfolio where I upload code to practice, experiment with, and master various algorithms and frameworks. It is currently under active development, and new models and implementations will be added regularly.

📂 Repository Contents

Here is an overview of what has been implemented so far:

🧠 Generative Adversarial Networks (GANs)

• Conditional GANs (CGANs): Implementation of Conditional Generative Adversarial Networks used for generating data (e.g., MNIST images) conditioned on class labels.
  • Location: GANs/CGANs.ipynb

📈 Classification & Regression

Try it here: https://slab10000.github.io/deep-learning-algorithms/classification-and-regression

Applied machine learning projects covering both classification and regression tasks.

• Classification - Petfinder Dataset:

  • A project focusing on classifying pet adoption profiles.
  • Includes a Multi-Layer Perceptron (MLP) model saved in ONNX format (mlp_net_model.onnx).
  • Location: classification-and-regression/classification/

• Regression - Songs Dataset:

  • A regression analysis project, likely predicting song popularity or features.
  • Utilizes XGBoost (Extreme Gradient Boosting) and includes the serialized model (xgboost_songs_model.onnx).
  • Location: classification-and-regression/regression/

🎬 Neural Input Optimization (NIO) - Movie Optimization

• Project Blockbuster: NIO Optimisation for Movies:
  • A project that uses Neural Input Optimization (NIO) to reverse-engineer the optimal movie blueprint for maximizing both commercial success (Gross Revenue) and critical acclaim (IMDB Score).
  • Dataset: IMDB 5000 Movie Dataset from Kaggle (~5000 movies, 28 features)
  • Model Architecture: Residual Neural Network (ResNet) implemented in PyTorch with residual connections and dropout regularization
  • Optimization Goal: Find optimal movie characteristics (Budget, Cast, Genre) that maximize Return on Investment (ROI = Gross/Budget) while maintaining:
    • IMDB Score between 9.0 and 10.0
    • Budget between $20M and $200M
  • Key Results: The NIO algorithm identified that a mid-range budget (~$110M) with high star power and specific genre combinations yields optimal ROI (9.09x) while maintaining critical acclaim
  • Location: NIO/

🖼️ Convolutional Neural Networks (CNNs) - Shape Classification

• Geometric Shape Classifier:
  • A CNN-based image classification project that identifies geometric shapes by counting their number of sides (Triangles: 3, Squares: 4, Pentagons: 5, Hexagons: 6).
  • Dataset: 10,000 images of geometric shapes (128×128 pixels, resized to 64×64 for training) with corresponding labels in CSV format
  • Model Architecture: "ShapeClassifier" CNN with:
    • 3 convolutional layers (16 → 32 → 64 channels) with ReLU activation and MaxPooling
    • Fully connected layers (128 hidden units) with dropout (0.5) for regularization
    • 4-class output layer for shape classification
  • Training: 20 epochs with batch size 64, CrossEntropyLoss, Adam optimizer, GPU-accelerated
  • Performance: Achieved excellent results with:
    • Precision: 0.971
    • Recall: 0.971
    • F1 Score: 0.971
  • Custom Dataset: Implemented ShapesDataset class for loading images and mapping side counts to class indices
  • Location: CNN/

🔢 Tensor Operations

Foundational notebooks for understanding data manipulation and tensor math.

• NumPy & PyTorch: Introductory notebooks covering the basics of NumPy arrays and PyTorch tensor operations, essential for any Deep Learning workflow.
  • Location: tensor-operations/

🛠️ Tech Stack & Tools with

• Languages: Python
• Deep Learning Frameworks: PyTorch
• Machine Learning Libraries: XGBoost, Scikit-Learn (implied)
• Data Manipulation: NumPy, Pandas
• Model Exchange: ONNX (Open Neural Network Exchange)
• Environment: Jupyter Notebooks

🚧 Status

This project is in a Work-In-Progress state. I am constantly learning and adding new implementations, including but not limited to:

• Computer Vision models (ViTs, more advanced CNNs)
• NLP architectures (Transformers, RNNs)
• Reinforcement Learning algorithms
• More advanced GAN architectures

Feel free to explore the code!