This repository contains a series of laboratory works completed for the "Data Science" course. The projects demonstrate a progression from exploratory data analysis and classical machine learning to deep learning with PyTorch and model deployment.
A predictive modeling project based on the popular Kaggle competition. The goal is to predict which passengers were transported to an alternate dimension during a collision with a spacetime anomaly.
- Core Tasks: Data cleaning, handling missing values, and feature engineering.
- Dimensionality Reduction: Utilization of PCA (Principal Component Analysis) to identify feature importance and reduce noise.
- Modeling: Implementation of classification algorithms (Decision Trees/Random Forests) with performance evaluation.
Implementation of a multi-class classification model for the classic Iris dataset using PyTorch.
- Architecture: A fully connected neural network built using
torch.nn. - Workflow: Data scaling with
StandardScaler, customDatasetandDataLoaderimplementation, and training loops with backpropagation. - Evaluation: Visualization of model performance through Confusion Matrices and loss/accuracy curves.
A deep learning project focused on Computer Vision, involving transfer learning and model deployment.
- Fine-tuning: Utilizing a pre-trained model and fine-tuning it to classify 7 different species of hares.
- Optimization: Implementation of data augmentation, learning rate scheduling, and logging of training metrics to CSV.
- Deployment: An interactive web interface developed with Gradio, allowing users to classify images via file upload or direct URL.
- End-to-End Pipelines: From raw data preprocessing to deploying a functional UI.
- Deep Learning: Extensive use of PyTorch for both tabular data and image classification.
- Visual Analytics: Detailed plotting of PCA components, training history, and model metrics using
MatplotlibandSeaborn. - Interactive Inference: Real-time predictions through a web-based UI in
Gradio.
- Frameworks:
PyTorch,Scikit-learn - Data Handling:
Pandas,NumPy - Visualization:
Matplotlib,Seaborn - Deployment:
Gradio - Environment: Jupyter Notebook / Python 3.x