This project recreates the PCA algorithm from scratch using numpy for matrix operations. It then compares it to the PCA implementation from scikit-learn. The dataset used is the Iris flower dataset containing sepal length, width and petal length, width for 3 different types of irises' (Setosa, Versicolour, and Virginica) stored in a 150x4. Each type of irises has 50 instances in the dataset. The labels are removed for PCA since the goal is dimensionality reduction.
Recreating the PCA algorithm gives the benefit of
- understanding what happens under the hood of the algorithm
- understanding if the dimension reductionality for a dataset is best achieved using PCA
- understanding the mathematical concepts leveraged by the algorithm
- Allows for model customization
For a detailed explanation on the theory used for this computation, check out the accompanying article on medium
You can find the code for this project here.
File overview:
Build_PCA_(Principal_Component_Analysis)_from_Scratch.ipynb- the full code from this project
To follow this project, please install the following locally:
- Python 3.8+
- Python packages
- pandas
- numpy
- scikit-learn
The data used for this implementation is the salary data originally on Kaggle.
You can download the file we'll use in this project here:
- IRIS.csv - the Iris dataset used in this project.