Description
In this project, you’ll be implementing a facial analysis program using Principal Component Analysis (PCA), using the skills you learned from the linear algebra + PCA lecture. You’ll also continue to build your Python skills.
We’ll walk you through the process step-by-step (at a high level).
Packages Needed for this Project
In this project, you’ll need the following packages:
>>> from scipy.linalg import eigh
>>> import numpy as np
>>> import matplotlib.pyplot as plt
Dataset
You will be using part of Yale face dataset (Links to an external site.) (processed). You can download the dataset here: YaleB_32x32.npy
The dataset contains 2414 sample images, each of size 32×32. We will use n to refer to number of images, n=2414 and d to refer to number of features for each sample image, d=1024 (32×32). We will test your code only using the provided data set. Note, we’ll use LaTeX: x_ix i to refer to the ith sample image which would be a d-dimensional feature vector.
Program Specification
Implement these six Python functions to do PCA on our provided dataset, in a file called pca.py:
load_and_center_dataset(filename) — load the dataset from a provided .npy file, re-center it around the origin and return it as a NumPy array of floats
get_covariance(dataset) — calculate and return the covariance matrix of the dataset as a NumPy matrix (d x d array)
get_eig(S, m) — perform eigen decomposition on the covariance matrix S and return a diagonal matrix (NumPy array) with the largest m eigenvalues on the diagonal, and a matrix (NumPy array) with the corresponding eigenvectors as columns
get_eig_perc(S, perc) — similar to get_eig, but instead of returning the first m, return all eigenvalues and corresponding eigenvectors in similar format as get_eig that explain more than perc % of variance
project_image(image, U) — project each image into your m-dimensional space and return the new representation as a d x 1 NumPy array
display_image(orig, proj) — use matplotlib to display a visual representation of the original image and the projected image side-by-side
Load and Center the Dataset
First, if you haven’t, download our sample dataset to the machine you’re working on: YaleB_32x32.npy.
Once you have it, you’ll want to use the numpy function load() to load the file into Python. (You may need to install NumPy first.)
>>> x = np.load(filename)
This should give you an n x d dataset (n: the number of images in the dataset and d: the dimensions of each image)
Each row represents an image feature vector. For this particular dataset, we have n = 2414 (no. of images) and d=32 x 32=1024 (32 by 32 pixels).
Your next step is to center this dataset around the origin. Recall the purpose of this step from lecture—it is a technical condition that makes it easier to perform PCA—but does not lose any information.
To center the dataset is simply to subtract the mean LaTeX: \mu_xμ x from each data point LaTeX: x_ix i (image in our case), i.e. LaTeX: x_i^{cent}\:=\:x_i-\mu_xx i c e n t = x i − μ x where mean LaTeX: \mu_x=\frac{\:1}{n}\sum^n_{i=1}x_iμ x = 1 n ∑ i = 1 n x i (n is the no. of images)
You can take advantage of the fact that x (as defined above) is a NumPy array and as such, has this convenient behavior:
>>> x = np.array([[1,2,5],[3,4,7]])
>>> np.mean(x, axis=0)
=> array([2., 3., 6.])
>>> x – np.mean(x, axis=0)
=> array([[-1., -1., -1.],
[ 1., 1., 1.])
After you’ve implemented this function, it should work like this:
>>> x = load_and_center_dataset(‘YaleB_32x32.npy’)
>>> len(x)
=> 2414
>>> len(x[0])
=> 1024
>>> np.average(x)
=> -8.315174931741023e-17
(Its center isn’t exactly zero, but taking into account precision errors over 2414 arrays of 1024 floats, it’s what we call Close Enough.)
Note, From now on, we will use LaTeX: x_ix i to refer to LaTeX: x_i^{cent}x i c e n t.
Find Covariance Matrix
Recall, from lecture, that one of the interpretations of PCA is that it is the eigendecomposition of the sample covariance matrix. We will rely on this interpretation in this assignment, with all of the information you need below.
The covariance matrix is defined as
LaTeX: S\:=\frac{\:1\:}{n-1}\sum^n_{i=1}\:x_ix_i^TS = 1 n − 1 ∑ i = 1 n x i x i T
Note that LaTeX: x_ix i is one of the n images in the dataset (centered) and it is considered as a column vector (size d x 1) in this formula.
If you consider it as a row vector (size 1 x d), then the formula would like below.
LaTeX: S\:=\frac{\:1\:}{n-1}\sum^n_{i=1}\:x^T_ix_iS = 1 n − 1 ∑ i = 1 n x i T x i
To calculate this, you’ll need a couple of tools from NumPy again:
>>> x = np.array([[1,2,5],[3,4,7]])
>>> np.transpose(x)
=> array([[1, 3],
[2, 4],
[5, 7]])
>>> np.dot(x, np.transpose(x))
=> array([[30, 46],
[46, 74]])
>>> np.dot(np.transpose(x), x)
=> array([[10, 14, 26],
[14, 20, 38],
[26, 38, 74]])
The result of this function for our sample dataset should be a d x d (1024 x 1024) matrix.
Get m Largest Eigenvalues/Eigenvectors
Again, recall from lecture that eigenvalues and eigenvectors are useful objects that characterize matrices. Better yet, PCA can be performed by doing an eigendecomposition and taking the eigenvectors corresponding to the largest eigenvalues. This replaces the recursive deflation step we discussed in class.
You’ll never have to do eigen decomposition by hand again (Links to an external site.)! Use scipy.linalg.eigh() to help, particularly the optional eigvals argument.
We want the largest m eigenvalues of S.
Return the eigenvalues as a diagonal matrix, in descending order, and the corresponding eigenvectors as columns in a matrix.
To return more than one thing from a function in Python:
def multi_return():
return “a string”, 5
mystring, myint = multi_return()
Make sure to return the diagonal matrix of eigenvalues FIRST, then the eigenvectors in corresponding columns. You may have to rearrange the output of eigh() to get the eigenvalues in decreasing order and make sure to keep the eigenvectors in the corresponding columns after that rearrangement.
>>> Lambda, U = get_eig(S, 2)
>>> print(Lambda)
[[1369142.41612494 0. ]
[ 0. 1341168.50476773]]
>>> print(U)
[[-0.01304065 -0.0432441 ]
[-0.01177219 -0.04342345]
[-0.00905278 -0.04095089]
…
[ 0.00148631 0.03622013]
[ 0.00205216 0.0348093 ]
[ 0.00305951 0.03330786]]
Get all Eigenvalues/Eigenvectors that Explain More than Certain % of Variance
We want all the eigenvalues that explain more than a certain percentage of variance.
Let LaTeX: \lambda_iλ i be an eigenvalue of the covariance matrix S. Then the percentage of variance explained is calculated as:
LaTeX: \frac{\lambda_i}{\sum^n_{i=1}\lambda_i}λ i ∑ i = 1 n λ i
Return the eigenvalues as a diagonal matrix, in descending order, and the corresponding eigenvectors as columns in a matrix.
Make sure to return the diagonal matrix of eigenvalues FIRST, then the eigenvectors in corresponding columns. You may have to rearrange the output of eigh() to get the eigenvalues in decreasing order and make sure to keep the eigenvectors in the corresponding columns after that rearrangement.
>>> Lambda, U = get_eig_perc(S, 0.07)
>>> print(Lambda)
[[1369142.41612494 0. ]
[ 0. 1341168.50476773]]
>>> print(U)
[[-0.01304065 -0.0432441 ]
[-0.01177219 -0.04342345]
[-0.00905278 -0.04095089]
…
[ 0.00148631 0.03622013]
[ 0.00205216 0.0348093 ]
[ 0.00305951 0.03330786]]
Project the Images
Given one of the images from your dataset and the results of your get_eig() function, create and return the projection of that image.
For any image LaTeX: x_ix i, we project it into the m dimensional space as
LaTeX: x^{pro}_i\:=\:\sum^m_{j=1}\alpha_{ij}u_jx i p r o = ∑ j = 1 m α i j u j where LaTeX: \alpha_{ij}\:=\:u^T_jx_iα i j = u j T x i , LaTeX: \alpha_i\in\mathbb{R}^mα i ∈ R m and LaTeX: x^{pro}_i\in\mathbb{R}^dx i p r o ∈ R d.
LaTeX: u_ju j is an eigenvector column (size d x 1) of U from the previous function.
Find the alphas for your image, then use them together with the eigenvectors to create your projection.
>>> projection = project_image(x[0], U)
>>> print(projection)
[6.84122225 4.83901287 1.41736694 … 8.75796534 7.45916035 5.4548656 ]
Visualize
We’ll be using matplotlib’s imshow (Links to an external site.). First, make sure you have the matplotlib library (Links to an external site.), for creating figures.
Follow these steps to visualize your images:
Reshape the images to be 32×32 (you should have calculated them as 1d vectors of 1024 numbers).
Create a figure with one row of two subplots (Links to an external site.).
The first subplot (on the left) should be titled (Links to an external site.) “Original”, and the second (on the right) should be titled “Projection”.
Use imshow() with optional argument aspect=’equal’ to render the original image in the first subplot and the projection in the second subplot.
Use the return value of imshow() to create a colorbar (Links to an external site.) for each image.
Render (Links to an external site.) your plots!
>>> x = load_and_center_dataset(‘YaleB_32x32.npy’)
>>> S = get_covariance(x)
>>> Lambda, U = get_eig(S, 2)
>>> projection = project_image(x[0], U)
>>> display_image(x[0], projection)
