Unrolling a 20x20 grid of pixels into a 400 dimensional vector.

During an exercise of a machine learning course, the task was to use logistic regression and neural networks to recognize handwirtten digits. For this, we are given a dataset containing 5000 training examples of handwritten digits where each example is a 20 x 20 pixel grayscale image of the digit. (I am not looking the answer on to how to do it)

When loading the data, it loads a 5000x400 matrix X where every row is a training example.

I would like to know how they have "unrolled" (as they call it) the 20x20 grid of pixels into a 400-dimensional vector and created this 5000x400 matrix X.

Thank you for yout help!

Answer 1

[Jan]

If the data are a collection of 500 matrices of the size 20x20:

data = rand(5000, 20, 20)

You can create the 2D matrix by:

data2D = reshape(data, 5000, 400)