NumPy Mastery

A structured, hands-on repository documenting my path to NumPy proficiency, from first principles to a production-style image processing pipeline built with zero external math libraries.

---

What's in here

numpy-mastery/
├── notebooks/                        # 6 annotated learning notebooks
│   ├── 01_array_creation.ipynb
│   ├── 02_indexing_slicing.ipynb
│   ├── 03_operations_broadcasting.ipynb
│   ├── 04_linear_algebra.ipynb
│   ├── 05_statistics_random.ipynb
│   └── 06_performance_tricks.ipynb   # coming soon
│
├── exercises/                        # 50+ graded problems (Easy → Hard)
│   ├── ex01_creation.ipynb
│   ├── ex02_indexing.ipynb
│   ├── ex03_operations.ipynb
│   ├── ex04_linear_algebra.ipynb
│   ├── ex05_statistics.ipynb
│   └──06_performance_tricks.ipynb
│
└── project/                          # Capstone: image processing pipeline
    ├── filters.py                    # Convolution: Sobel, Gaussian, Laplacian
    ├── transforms.py                 # Rotate, resize, flip, crop
    ├── pipeline.py                   # Composable | operator chain
    ├── demo.ipynb                    # Visual before/after walkthrough
    ├── benchmarks.ipynb              # NumPy vs naive Python speed tests
    └── tests/                        # pytest suite (35 tests)

---

---

Notebooks overview

#	Notebook	Key concepts
01	Array Creation	arange, linspace, zeros/ones/full, dtypes, memory layout
02	Indexing & Slicing	Basic, 2-D, fancy indexing, boolean masking, np.where
03	Operations & Broadcasting	Arithmetic, aggregation with axis, all 3 broadcasting rules
04	Linear Algebra	@ operator, linalg.solve, SVD, eigendecomposition
05	Statistics & Random	Descriptive stats, the modern default_rng API, Monte Carlo
06	Performance	Strides, views vs copies, vectorisation patterns, np.save

---

Capstone project — Image Processing Pipeline

An image processing library written in pure NumPy — no OpenCV, no PIL, no scipy.

Features implemented

• Filters: Gaussian blur, Sobel edge detection (x/y/magnitude), Laplacian, sharpen, emboss
• Transforms: bilinear resize, arbitrary rotation, horizontal/vertical flip, centre crop
• Pipeline: composable | operator to chain operations, optional timing
• Benchmarks: head-to-head timing vs plain Python loops on the same operations

from project.pipeline import Pipeline
from project.filters import gaussian_blur, sobel_magnitude
from project.transforms import rotate, resize

result = (
    Pipeline(image)
    | gaussian_blur(sigma=1.5)
    | rotate(angle=15)
    | sobel_magnitude()
    | resize(target=(256, 256))
).run()

Sample outputs

Original	Gaussian blur	Sobel edges	Rotated
(see demo.ipynb)

---

Quick start

git clone https://github.com/<your-username>/numpy-mastery
cd numpy-mastery
pip install -r requirements.txt
jupyter notebook notebooks/01_array_creation.ipynb

Requirements: numpy, matplotlib, pytest, jupyter

---

Skills demonstrated

• Vectorised computation, no Python loops where NumPy can do the job
• Broadcasting across N-dimensional arrays
• Memory-efficient operations via views and stride tricks
• Linear algebra from first principles (regression, SVD, eigenvalues)
• Statistically correct random number generation with the modern Generator API
• Clean, tested, documented Python code

---

Running the tests

cd project
pytest tests/ -v

---

Built as part of a structured AI/ML engineering curriculum.