Data Science Portfolio

This repository contains a collection of Jupyter notebooks showcasing various projects on data analytics, machine learning, and neural networks - in particular, projects that were created during the Deep Learning School (MIPT), fall 2024.

Projects

1. Game of Thrones Survival Prediction

View Notebook

• Developed a classification model to predict character survival in Game of Thrones
• Performed extensive exploratory data analysis with visualization
• Created and analyzed new features through feature engineering
• Implemented and compared multiple models:
  • Random Forest
  • Linear Regression
  • AdaBoost Classifier
  • KNeighborsClassifier
• Conducted hyperparameter optimization to improve model performance
• Tools used: Python, scikit-learn, pandas, seaborn

2. Kaggle Competition Project

View Notebook

• Participated in Kaggle competition focusing on prediction tasks
• Performed comprehensive data preprocessing:
  • Handled missing values
  • Data type optimization
  • Feature engineering
• Implemented gradient boosting models using CatBoost and XGBoost
• Applied data filtering techniques to improve model accuracy
• Tools used: Python, CatBoost, XGBoost, pandas, numpy

3. MNIST Classification with CNN and LeNet

View Notebook

• Implemented various neural network architectures from scratch
• Explored different activation functions and their impacts
• Built a fully-connected neural network
• Developed convolution operations for image processing
• Implemented LeNet architecture
• Compared performance across different approaches
• Tools used: Python, PyTorch, NumPy

4. Transfer Learning and CNN Optimization

View Notebook

• Applied transfer learning techniques
• Implemented data augmentation strategies
• Developed and optimized convolutional neural networks
• Experimented with layer freezing techniques
• Addressed class imbalance problems
• Tools used: Python, PyTorch, scikit-learn

5. Exploring Autoencoder Architectures

View Notebook

• Implemented and compared different types of autoencoders:
  • Basic autoencoder for facial image reconstruction
  • Variational Autoencoder (VAE) with MNIST dataset
  • Conditional VAE (CVAE) with MNIST dataset
• Analyzed and visualized latent space dimensions
• Performed sampling from trained models
• Compared VAE and CVAE performance and characteristics
• Tools used: Python, PyTorch, torchvision

Skills Demonstrated

• Machine Learning
  • Classification algorithms
  • Feature engineering
  • Model optimization
  • Gradient boosting
• Deep Learning
  • Neural network architecture design
  • Convolutional Neural Networks
  • Transfer learning
  • Data augmentation
  • Autoencoders
    • Basic autoencoder architecture
    • Variational Autoencoders (VAE)
    • Conditional VAE
    • Latent space analysis
• Data Analysis
  • Exploratory data analysis
  • Data visualization
  • Feature correlation analysis
  • Data preprocessing

Contact

• Email: vs.v.atrix@gmail.com
• LinkedIn: https://www.linkedin.com/in/v-atrix/

Setup Instructions

To run these notebooks:

1. Clone this repository
2. Install required packages: pip install -r requirements.txt
3. Open the notebooks in Jupyter or Google Colab