ML-Labs

This repository is a course-spanning machine learning workspace. It collects foundational lab notebooks, class tasks, fast experiments, supporting slides and reports, and a few small application-style projects built with Flask and Streamlit.

The work moves from early data handling and classical machine learning into clustering, neural networks, transformers, retrieval-augmented generation, and YOLO-based computer vision. It is best understood as a learning archive plus a set of prototype projects rather than a single packaged application.

Repository at a glance

• Classical ML notebooks covering data loading, preprocessing, feature engineering, regression, classification, and evaluation
• Unsupervised learning work on clustering and related exploratory analysis
• Neural network exercises using TensorFlow and custom NumPy implementations
• NLP experiments with Hugging Face pipelines and GPT-2
• Small web apps built with Flask and Streamlit
• RAG and local-document question answering experiments
• A multi-task YOLO11 computer vision demo
• Course slides, PDFs, reports, exported notebooks, and zipped lab submissions

Main contents

1. Core lab notebooks

Top-level notebooks contain the main machine learning lab progression:

• 1167-Lab1-BSCS_A_7.ipynb Data loading practice with CSV, Excel, JSON, dictionaries, and URL-based sources using pandas.
• 1167-Lab2-BSCS-A_7.ipynb Early regression concepts including logistic and polynomial regression, plus train/test split workflow.
• 1167-Lab3-BSCS_A_7.ipynb Data preprocessing and feature selection.
• 1167-Lab3-BSCS_A_7_ClassTask.ipynb Logistic regression from scratch, logistic regression with scikit-learn, and polynomial regression practice.
• 1167-Lab4-BSCS-A_7_Activity.ipynb KNN, SVM, voting classifiers, and credit card fraud detection with SMOTE on imbalanced data.
• 1167-Lab4-BSCS_A.ipynb Short notebook tied to the imbalance topic from Lab 4.
• 1167-Lab5-BSCS_A_7_Activity.ipynb Unsupervised learning and clustering using the Mall Customers dataset.
• 1167-Lab5-BSCS_A_7_ClassTask.ipynb Unsupervised learning workflow on the mushrooms dataset with exploration and visualization.
• 1167-Lab6-BSCS_A_7_ClassTask.ipynb Unsupervised learning work on the wine dataset with scaling and analysis setup.
• 1167-Lab8-9-BSCS_A_ClassTask.ipynb ANN work on Iris using TensorFlow/Keras, including hidden-layer activation comparisons.

2. Fast labs and experiments

• fast_lab_1.ipynb Synthetic moons dataset work with augmentation, visualization, and model-focused experimentation.
• fast_lab_2.ipynb A more hands-on deep learning exercise with a custom DeepNeuralNetwork implementation built in NumPy, including forward pass, backpropagation, and training helpers.
• test.ipynb Small scratch notebook.

3. NLP and transformers

The Lab10/ folder contains transformer-focused experiments:

• Lab10/1167-Lab10-BSCS_A_ClassTask.ipynb Text classification, summarization, translation, and image-to-text oriented tasks.
• Lab10/simple_emotion_pipeline.ipynb Emotion/sentiment classification with Hugging Face pipelines.
• Lab10/Transformers Hugging face.ipynb Summarization and translation pipeline experiments.
• Lab10/Text_Generation_with__GPT2.ipynb GPT-2 text generation workflow.

4. Reports and supporting material

• Ensemble Report/ Ensemble learning notebooks, an exported HTML notebook, and PDF reports.
• ML_Lab1.pptx, ML_Lab2_3.pptx, ML_Lab4.pptx, ML_Lab_05.pptx, ML_Lab7.pptx Course slide decks used alongside the labs.
• Lab 08-09 - Instructions.pdf Instructions for the later neural-network-related labs.
• Lab11/*.pdf PDF assets related to the final lab and offline PDF QA work.

5. App-style projects

These are the most application-oriented parts of the repository.

Flask/

A basic Flask + SQLAlchemy to-do app with Jinja templates and SQLite persistence.

Features visible in the code:

• Create, edit, list, and delete tasks
• SQLite-backed storage via SQLAlchemy
• Template inheritance with Jinja
• Simple Tailwind-oriented template structure

Key files:

• Flask/app.py
• Flask/templates/
• instance/database.db

Lab11/part1_flask/

A Flask interface for training and evaluating simple ML models against either built-in datasets or uploaded CSV files.

Features visible in the code:

• Built-in datasets: Iris, Digits, Wine
• Models: Logistic Regression, KNN, SVM
• Standard train/test workflow with scaling
• Metrics: accuracy, precision, recall, F1-score
• Auto-generated bar chart of evaluation results

Key files:

• Lab11/part1_flask/app.py
• Lab11/part1_flask/templates/index.html
• Lab11/part1_flask/static/metrics.png

Lab11/part2_RAG_StreamLit/

A Streamlit RAG prototype for asking questions over an uploaded PDF document.

Features visible in the code:

• PDF ingestion with PyPDF2
• Text chunking with LangChain text splitters
• Embeddings with sentence-transformers/all-MiniLM-L6-v2
• Vector search using FAISS
• Local answer generation through GPT4All

Important note:

• The code currently points to a local GGUF model path outside this repository, and .gitignore also excludes q4_0-orca-mini-3b.gguf. That means the app is not fully self-contained out of the box without supplying a local GPT4All model.

Key file:

• Lab11/part2_RAG_StreamLit/RAG.py

Lab11/Lab11_HomeTask/yolo/

A more ambitious Flask-based computer vision demo wrapping multiple YOLO11 tasks behind separate routes.

Tasks implemented:

• Object detection
• Instance segmentation
• Image classification
• Pose estimation
• Oriented bounding box detection

Features visible in the code:

• Individual Flask endpoints per task
• Separate service functions for each inference path
• A single HTML front end for uploading images and viewing structured results
• Pretrained YOLO11 weights stored in the repo under models/

Key files:

• Lab11/Lab11_HomeTask/yolo/app.py
• Lab11/Lab11_HomeTask/yolo/routes/yolo_routes.py
• Lab11/Lab11_HomeTask/yolo/services/yolo_service.py
• Lab11/Lab11_HomeTask/yolo/models/yolo_model.py
• Lab11/Lab11_HomeTask/yolo/templates/main.html

6. Miscellaneous notes

• Playwright/PlaywrightNotes.ipynb Browser automation notes separate from the ML material.
• Lab11/part1_flask.zip, Lab11/part2_RAG_StreamLit.zip, Lab11/Lab11_HomeTask/1167-lab11-hometask.zip Archived project bundles/submissions.

Tech stack

Based on pyproject.toml, the repository uses or references:

• Python 3.12
• Jupyter Notebook / IPython kernel
• NumPy, pandas, matplotlib, seaborn
• scikit-learn, imbalanced-learn, Optuna, Keras Tuner
• TensorFlow
• PyTorch, torchvision, torchaudio
• Hugging Face transformers, sentencepiece, sacremoses
• Diffusers
• Flask, Flask-SQLAlchemy, Streamlit
• PyPDF2, LangChain, sentence-transformers, FAISS, GPT4All
• Ultralytics YOLO
• uv for environment management
• ruff for linting

Getting started

Environment setup

This repo is configured as a Python project with uv.

uv sync

If you prefer working directly in notebooks:

uv run jupyter notebook

Running the app demos

Flask to-do app:

uv run python Flask/app.py

Flask ML evaluation app:

uv run python Lab11/part1_flask/app.py

Streamlit RAG app:

uv run streamlit run Lab11/part2_RAG_StreamLit/RAG.py

YOLO multi-task demo:

uv run python Lab11/Lab11_HomeTask/yolo/app.py

Notes and limitations

• Some notebooks expect local datasets under Datasets/, and that folder is ignored in .gitignore.
• The RAG prototype depends on a locally available GPT4All model file that is not committed.
• A number of files in the repository are course artifacts, exports, or archives rather than polished reusable modules.
• There is a committed SQLite database under instance/database.db for the Flask to-do example.
• The repository currently reads best as a documented learning portfolio rather than a single install-and-run product.

Recommended way to read the repo

If you want to review the work in sequence:

1. Start with the top-level numbered lab notebooks for the course progression.
2. Move to fast_lab_1.ipynb and fast_lab_2.ipynb for more implementation-heavy experiments.
3. Review Lab10/ for transformer and NLP work.
4. Review Ensemble Report/ for the report-oriented material.
5. Finish with the application-style work in Flask/ and Lab11/.

Best standalone portfolio candidates

If this repository is later split into smaller, presentation-ready projects, the strongest candidates are:

1. Lab11/Lab11_HomeTask/yolo/ Best standalone demo in the repo. It has a clear product shape, multiple computer vision tasks, API-style routes, and a presentable UI.
2. Lab11/part2_RAG_StreamLit/ Strong portfolio topic because local-document RAG is easy to explain and relevant, though it needs cleanup around model paths and setup.
3. Lab11/part1_flask/ Good smaller project showing the bridge between machine learning and web deployment, especially if expanded with better dataset handling and result persistence.

Summary

This repository captures a broad machine learning learning journey: from data loading and preprocessing, through classical ML and unsupervised learning, into neural networks, NLP, RAG, and computer vision deployment demos. It is a useful archive of both coursework and practical experimentation, with the Lab 11 application projects standing out as the best material for standalone portfolio use.