▶ Linear Algebra

1 Scalars

1.1 What are scalars?

  • Value of \(\pi\)
  • Grade for math
  • Height of a person
  • Birth year of Albert Einstein
  • Color of the Canadian flag
  • My favourite season of the year

1.2 Data types

  • Float: \(\mathbb{R}\)
  • Positive floats: \(\mathbb{R}^+\)
  • Fraction: \([0, 1]\)
  • Integer: \(\mathbb{N}\)
  • Categorical: \(\{\mathrm{red}, \mathrm{green}, \mathrm{blue}\}\)

1.3 Different kinds of scalars

  • Constant values
  • Variables:
    • Dependent
    • Independent
    • Unknown

1.4 Measuring closeness of scalars

Consider two scalars \(x\) and \(y\)

  • Similarity measure: \[ \mathrm{sim}(x, y) \in [-1, 1] \] \[ \mathrm{sim}(x, x) = 1 \]

    Here is an example: \[ \mathrm{sim}(x, y) = \frac{\min(x, y)}{\max(x, y)} \]

  • Distance measure: \[ d(x, y) \in [0, \infty) \] \[ d(x, x) = 0 \] \[ \forall x, y, z,\ d(x, y) \leq d(x, z) + d(z, y) \quad \mathrm{triangular\ inequality} \]

    Here is an example: \[ d(x, y) = |x - y| \]

    Here is another example: \[ d(x, y) = (x - y)^2 \]

2 Vectors

2.1 Examples

Some examples of vectors:

  • Grades for math, english and biology: \[ x = \left[\begin{array}{c} g_\mathrm{math} \\ g_\mathrm{english} \\ g_\mathrm{biology} \end{array}\right] \in \mathbb{R}^3 \]

  • Person with height, weight, IQ and income. \[ x = \left[\begin{array}{c} \mathrm{height} \\ \mathrm{weight} \\ \mathrm{IQ} \\ \mathrm{income} \end{array}\right]\in\mathbb{R}^4 \]

2.2 Notation

An entry in a vector is a scalar.

\[ x = [x_1, x_2, x_3, \dots, x_n] \in \mathbb{R}^n \]

\[ x = \left[\begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \vdots \\ x_n \end{array}\right]\in\mathbb{R}^n \]

Each entry \(x_i\in\mathbb{R}\).

2.2.1 Magnitudes of vectors

Let \(x\in\mathbb{R}^n\), its magnitude is given as:

\[ \|x\| = \sqrt{\sum_{i=1}^n x_i^2} \]

We usually express the magnitude squared:

\[ \|x\|^2 = \sum_{i=1}^n x_i^2 \]

Distance between two vectors can be expressed as a magnitude. \[ d(x,y) = \|x - y\| \]

2.2.2 Inner product

Consider two vectors that have the same dimensionality: \(x, y \in \mathbb{R}^n\).

Inner product

\[\left<x,y\right> = \sum_{i=1}^n x_i\cdot y_i\]

Magnitude as an inner product:

\[ \|x\|^2 = \left<x, x\right> \]

2.2.3 Measuring closeness: similarity

Consider two vectors that have the same dimensionality: \(x, y \in \mathbb{R}^n\)

We can measure the similarity of two vectors based on the angle between them.

Note:

\[ \begin{eqnarray} \cos(\theta) &=& \frac{\sum_{i=1}^n x_iy_i}{\sqrt{\left(\sum_{i=1}^n x_i^2\right)\left(\sum_{i=1}^n y_i^2\right)}} \\ &=& \frac{\left<x, y\right>}{\|x\| \|y\|}\\ &\in& [-1, 1] \end{eqnarray} \]

2.2.4 Measuring closeness: distance

\[ \begin{eqnarray} d(x, y) &=& \sqrt{\sum_{i=1}^n (x_i - y_i)^2} \\ &=& \|x - y\| \end{eqnarray} \]

3 Matrices

A matrix is a grid of scalars. A matrix has two axes: rows and columns.

  • \(A_{ij}\in\mathbb{R}\) is an entry where \(i\in[1, m]\) and \(j\in[1, n]\)

  • \(A\) is a matrix of shape \((m, n)\)

  • \(A\in\mathbb{R}^{m\times n}\)

3.1 Reduction

Reduction functions are functions that map vectors or matrices to a scalar.

\[ h: A \mapsto v \]

Some examples are:

  • Summation: \(h(A) = \sum_{i=1}^m\sum_{j=1}^n A_{ij}\)
  • Mean
  • Max / Min

3.2 Reducing along an axis

We can apply any reduction function along an axis. This results in a vector.

\[ h_\mathrm{axis=0} : \mathbb{R}^{m\times n} \to \mathbb{R}^n \]

\[ h_\mathrm{axis=0}(A)[j] = h(A[:,j]) \]

4 Transformations

import numpy as np
import matplotlib.pyplot as pyplot

4.1 A single vector

vector = np.array([1., 2.])

pyplot.scatter([vector[0]], [vector[1]], s=100)
pyplot.grid(True)
pyplot.xlim((-5, 5))
pyplot.ylim((-5, 5));

4.2 A collection of vectors

xs = np.random.normal(0, 1, size=(1000,))
ys = np.random.normal(0, 2, size=(1000,))

vectors = np.vstack([xs, ys])
vectors.shape
(2, 1000)

Let’s plot them.

pyplot.figure(figsize=(6, 6))
pyplot.xlim((-10, 10))
pyplot.ylim((-10, 10))

pyplot.scatter(vectors[0, :], vectors[1, :], s=1);

4.3 Scaling axes by matrices

Here is a scaling matrix:

  • grows x-coordinates by 3
  • shrinks y-coordinates by 1/3
M = np.array([
    [3, 0],
    [0, 1/3]
])

M.shape
(2, 2)

We can apply the scaling to the vectors:

x = M @ vectors
x.shape
(2, 1000)

Let’s see the effect of matrix multiplication when \(M\) is a scaling matrix.

pyplot.figure(figsize=(6, 6))
pyplot.xlim((-10, 10))
pyplot.ylim((-10, 10))

pyplot.scatter(vectors[0, :], vectors[1, :], s=1, color='gray');
pyplot.scatter(x[0, :], x[1, :], s=1, color='red');

4.4 Rotational matrix

The general form of a rotational matrix is:

\[ R = \left[\begin{array}{cc} \cos(\theta) & \sin(\theta) \\ -\sin(\theta) & \cos(\theta) \end{array}\right] \] where \(\theta\) is the clockwise rotation angle.

theta = np.pi / 4        # 45 degrees

R = np.array([
    [np.cos(theta), np.sin(theta)],
    [-np.sin(theta), np.cos(theta)]
])

R.shape
(2, 2)

We can apply the rotational matrix to the set of vectors.

x = R @ vectors
x.shape
(2, 1000)

Let’s see the effect of matrix multiplication when \(R\) is a rotation matrix.

pyplot.figure(figsize=(6, 6))
pyplot.xlim((-10, 10))
pyplot.ylim((-10, 10))

pyplot.scatter(vectors[0, :], vectors[1, :], s=1, color='gray');
pyplot.scatter(x[0, :], x[1, :], s=1, color='red');

4.5 Compositional transformation

Here is the composition that performs rotation and then translation.

x = M @ R @ vectors
x.shape
(2, 1000)
pyplot.figure(figsize=(6, 6))
pyplot.xlim((-10, 10))
pyplot.ylim((-10, 10))

pyplot.scatter(vectors[0, :], vectors[1, :], s=1, color='gray');
pyplot.scatter(x[0, :], x[1, :], s=1, color='red');

Now, consider the composition that performs translation followed by rotation.

x = R @ M @ vectors
x.shape
(2, 1000)
pyplot.figure(figsize=(6, 6))
pyplot.xlim((-10, 10))
pyplot.ylim((-10, 10))

pyplot.scatter(vectors[0, :], vectors[1, :], s=1, color='gray');
pyplot.scatter(x[0, :], x[1, :], s=1, color='red');