Feature EngineeringPreprocessingComputer VisionNLP

Feature Extraction & Preprocessing

Turning text and images into numbers a model can learn from.

Role: Learning project
When: September 2026
Tools: Python, pandas, NumPy, scikit-learn, Jupyter Notebook

Description

A working set of notebooks on the step before modelling: getting raw text and images into a numerical form that a model can use at all.

This is the part of the pipeline I find most interesting, because it is where domain knowledge enters. The algorithm cannot recover information that the features threw away.

What it covers

Bag-of-words representations for text.
One-hot encoding for categorical variables.
Standardisation and scaling.
Pixel intensities as image features, and their limits.
Points of interest in images.
SIFT — scale-invariant feature transform.
Extracting text from images.

What the notebooks produce

SIFT keypoints on her own avatar illustration — each circle is a feature the detector found, sized by its scale and turned by its orientation.

The same portrait in greyscale with detected points of interest marked in red. — The same image in greyscale, with the detected points of interest marked.