Coursework and ML Notebooks
- Practice and coursework
- Coursework Archive
- 2023
- Open source
Python, Jupyter, scikit-learn, TensorFlow and 3 more
These five repositories hold coursework and practice from my B.Tech in AIML at The NorthCap University, 2021 to 2025. Between them are lab manuals for 16 courses, six end-semester projects, three Java console programs and 156 Jupyter notebooks.
Not all of it is mine alone, and this page says which parts are not. At least three of the six end-semester projects were group work. Most of the machine learning repository follows public tutorials, and the deep learning notebooks were collected from elsewhere.
# six classifiers, one predict call reusedcd "machine-learning/Loan Approval Prediction"grep -o '[a-z_]*\.predict(X_test)' *.ipynblog_reg.predict(X_test)knn.predict(X_test)knn.predict(X_test)knn.predict(X_test)knn.predict(X_test)rf_classifier.predict(X_test)
2 commands6 lines of output
- Chapter
- Practice and coursework
- Category
- Coursework Archive
- Year
- Availability
- Open source
- Last commit
- Python
- Jupyter
- scikit-learn
- TensorFlow
- Keras
- Java
- Arduino
Why it is public
The lab repository’s README gives three reasons for publishing it. They are a personal reference, material to share with other students, and a record of each experiment. The courses asked for work in fixed formats, so the repository holds notebooks, Java programs, SQL with screenshots, Word workbooks and PDFs as they were. I did not convert any of it into a common format. The machine learning and Kaggle repositories began in October 2023, beside the lab work. The deep learning repository followed in December 2024.
Five repositories
Each course has a folder named after its code, such as NCU-CSL312 - DL - Lab Manual. Inside it sits one Experiment N folder per experiment. The courses include Analysis and Design of Algorithms, Operating System, Deep Learning and Reinforcement Learning. Others cover expert systems, cyber security and NLP. Six End Semester Project folders sit beside them. I counted notebooks with git ls-files '*.ipynb' in each repository.
| Repository | What it holds | Notebooks |
|---|---|---|
ncu-lab-manual-and-end-semester-projects | 16 lab manuals, 6 end-semester projects | 112 |
| machine-learning | 5 projects at the root, 26 under #/ | 32 |
| deep-learning-projects | 10 collected projects | 10 |
| kaggle-competitions | Titanic and Digit Recognizer | 2 |
| java-projects | Blood bank, calendar, timetable | 0 |
End-semester projects
Loan approval prediction was a three-person project for Introduction to AI and ML. It compares six scikit-learn classifiers on a 4,269-row Kaggle dataset, and Random Forest reached 97.9% accuracy on 854 held-out rows. Bitcoin price prediction trains a Keras LSTM with 64 units on 15-day windows of BTC-USD closing prices. A Gradio notebook serves it, and the CSL229 lab manual documents it with an SRS and UML diagrams. The AI for Robotics project, built by four of us, is an Arduino Uno car with Bluetooth and an ultrasonic sensor. Within 12 cm of an obstacle, code.c backs off, measures left and right with a servo, and turns toward the clearer side. Two projects are Java console programs: a blood bank that keeps records in text files, and a calendar. The sixth is a two-person Python blackjack game whose dealer hits on 16 or less.
Notebooks and programs
The MNIST notebook trains five scikit-learn classifiers on 70,000 digits, and the SVM scores best at 97.6% on 14,000 held-out images. The Iris notebook fits logistic regression and scores 96.7% on 30 test rows. The Kaggle Digit Recognizer notebook reuses that pipeline and writes 28,000 predictions in the ImageId,Label submission format. The deep learning notebooks use VGG16 for pneumonia X-rays and landmarks, VGG19 for face masks, and a U-Net autoencoder for colourisation. In java-projects, the calendar stores 100 years of dates in arrays, each date holding linked lists of birthdays, reminders and to-dos. A timetable scheduler books 12 daily slots greedily.
The lesson kept
Reading the notebooks again, the lesson is evaluation discipline. In the loan notebook, one copied knn.predict(X_test) call means three of the six reported scores are really KNN's. The Bitcoin LSTM reports a test error ten times its training error, with no previous-day baseline to judge it against. Today I compare every model with a naive baseline and check which model produced each number before I report it.
What this is not
It is not research, and none of it is maintained. Most projects under machine-learning/#/ follow public video tutorials, and the deep learning notebooks were collected from courses and exercises. The last coursework commit is from November 2024. Each repository carries an MIT licence added in May 2026, which covers only my own files.