SVD + LDA Image Classification Project (Starter)

This repository is a starter kit for the final programming project.

You will implement:

• power_method for the dominant eigenvalue/eigenvector of a symmetric matrix.
• svd_compress for rank-k image approximation using SVD.
• svd_features to build feature vectors from image singular values.
• lda_train to train a two-class Linear Discriminant Analysis (LDA) classifier.
• lda_predict to apply the trained LDA classifier.

The autograder will import project.py and call these functions with specific signatures. Do not change the function names or signatures.

Files

• project.py — main implementation file with function stubs.
• project_data_example.npz — tiny synthetic dataset for local testing.
• requirements.txt — minimal Python dependencies.
• .gitignore — standard Python ignores.

Quick start

1. Create a virtual environment (optional but recommended).

2. Install dependencies:

   pip install -r requirements.txt

3. Open project.py and implement each function where indicated.

4. Try running the self-test:

   python project.py

   This will:

   • Load project_data_example.npz if present.
   • Extract SVD features.
   • Train an LDA classifier.
   • Print the test accuracy on the tiny synthetic dataset.

In the real assignment, the Gradescope autograder will provide a different project_data.npz with more realistic images and labels.