MachineLearning_Notebooks

Here are all my Machine Learning/LLMs/Deep Learning Notebooks.

Project List:

1. Applied Machine Learning Techniques
2. NanoGPT
3. Company Brochure Generator
4. Meeting Minutes Generator
5. Python Code to C++ Code Converter

Applied Machine Learning Techniques:

• Implemented and fine-tuned machine learning algorithms: Random Forests, XGBoost (XGB), LightGBM (LGBM), Logistic Regression, Linear Regression, Decision Trees, and Ensemble Techniques.
• Worked with diverse datasets (thousands to millions of records) for classification, regression, and feature engineering.
• Achieved up to 95% predictive accuracy in various projects.
• Conducted model evaluation, cross-validation, and hyperparameter tuning for robust and optimized performance.

NanoGPT :

• Designed and implemented a small-scale GPT-inspired language model trained on Shakespeare’s complete works.
• Generated text mimicking Shakespeare’s language, tone, poetic style, and literary devices.
• Preprocessed and tokenized Shakespeare’s texts to create a high-quality training dataset.
• Utilized transformer-based architecture with attention mechanisms to capture complex syntax and metaphors.
• Fine-tuned the model using state-of-the-art techniques for coherent, stylistically accurate text generation.
• Demonstrated expertise in NLP, deep learning, and creative text generation by adapting modern techniques to classical literature.

Company Brochure Generator:

• Designed and implemented a web-based summarization tool using a custom-built web scraper to extract and summarize website content.
• Developed a scraper to handle diverse HTML structures and dynamically loaded elements for accurate data extraction.
• Applied advanced text preprocessing (e.g., cleaning HTML, removing redundancy) to prepare data for summarization.
• Utilized NLP techniques to identify key sentences and generate concise, context-preserving summaries.
• Enabled users to reduce lengthy content into digestible highlights while retaining key insights.
• Demonstrated expertise in web scraping, data processing, NLP, and building end-to-end solutions to address information overload.

Meeting Minutes Generator :

• Designed and developed a Meeting Minutes Generator to automate meeting summaries from audio recordings.
• Integrated OpenAI’s Whisper for accurate transcription of meeting discussions.
• Utilized Google’s Gemma for summarization to extract key insights (summary, discussion points, takeaways, action items).
• Automated the generation of structured meeting minutes in markdown format, including attendees, location, date, and assigned action items.
• Enhanced productivity by reducing manual effort and improving accessibility of meeting information.
• Built the tool to be scalable and adaptable for various meeting types and industries.

Python Code to C++ Code Converter :

• Developed a Python-to-C++ Code Generator using ChatGPT-4.0-mini and Claude Sonnet3.5 APIs for code generation and optimization.
• Built an interactive Gradio-based UI for users to input Python code, customize parameters, and view real-time C++ output.
• Enhanced usability and efficiency, enabling seamless Python-to-C++ translation with minimal developer effort.
• Integrated modern programming standards and best practices to ensure high-quality, clean C++ code generation.
• Delivered a practical tool for improving performance or transitioning Python prototypes to production-grade C++.

Predictive Modeling for Kaggle Competitions:

• Designed and implemented machine learning models for Kaggle competitions (House Prices Prediction, Titanic Disaster), achieving over 80% accuracy in both.
• Utilized feature engineering, exploratory data analysis (EDA), and data preprocessing to extract insights and enhance model performance.
• Applied ensemble methods like Random Forests and Gradient Boosting to address prediction challenges.
• Conducted hyperparameter tuning and model evaluation to optimize performance and ensure robustness.
• Demonstrated problem-solving skills by applying data-driven approaches to real-world predictive tasks.

PyTorch: Linear Regression, CNN, and ANN

1. Introduction

• Overview of PyTorch and its libraries
• Implementing Linear Regression, CNN, and ANN efficiently

2. Setup

• Install and import necessary libraries
• Enable GPU acceleration if available

3. Linear Regression

• Generate synthetic data using torch.randn
• Define model with torch.nn.Linear
• Train using MSE loss and SGD optimizer
• Plot regression results

4. Convolutional Neural Network (CNN)

• Load MNIST/CIFAR-10 with torchvision.datasets
• Normalize and batch data using DataLoader
• Define CNN with convolution, pooling, and fully connected layers
• Train using Cross-Entropy loss and Adam optimizer
• Evaluate accuracy on test data

5. Artificial Neural Network (ANN)

• Prepare tabular data with torch.utils.data.TensorDataset
• Define MLP using torch.nn.Sequential
• Train using BCE/Cross-Entropy loss and Adam optimizer
• Evaluate performance

6. Summary

• Key insights from Linear Regression, CNN, and ANN
• Future exploration in deep learning