cm1: Base introduction and some elements of linear algebra

This commit is contained in:
Samuel Ortion 2023-09-22 17:32:56 +02:00
parent 14928631ec
commit a92a13d354
11 changed files with 389 additions and 18 deletions

6
Makefile Normal file
View File

@ -0,0 +1,6 @@
options=-shell-escape -file-line-error
all: main.pdf
%.pdf: %.tex
lualatex $(options) $<

View File

@ -0,0 +1 @@
\part{}

View File

@ -0,0 +1,112 @@
\section{Generalized Linear Model}
\[
g(\EE(Y)) = X \beta
\]
with $g$ being
\begin{itemize}
\item Logistic regression: $g(v) = \log \left(\frac{v}{1-v}\right)$, for instance for boolean values,
\item Poission regression: $g(v) = \log(v)$, for instance for discrete variables.
\end{itemize}
\subsection{Penalized Regression}
When the number of variables is large, e.g, when the number of explicative variable is above the number of observations, if $p >> n$ ($p$: the number of explicative variable, $n$ is the number of observations), we cannot estimate the parameters.
In order to estimate the parameters, we can use penalties (additional terms).
Lasso regression, Elastic Net, etc.
\subsection{Simple Linear Model}
\begin{align*}
\Y &= \X & \beta & + & \varepsilon.\\
n \times 1 & n \times 2 & 2 \times 1 & + & n \times 1 \\
\begin{pmatrix}
Y_1 \\
Y_2 \\
\vdots \\
Y_n
\end{pmatrix}
&= \begin{pmatrix}
1 & X_1 \\
1 & X_2 \\
\vdots & \vdots \\
1 & X_n
\end{pmatrix}
& \begin{pmatrix}
\beta_0 \\
\beta_1
\end{pmatrix}
& + &
\begin{pmatrix}
\varepsilon_1 \\
\varepsilon_2 \\
\vdots \\
\varepsilon_n
\end{pmatrix}
\end{align*}
\subsection{Assumptions}
\begin{itemize}
\item
\end{itemize}
\subsection{Statistical Analysis Workflow}
\begin{enumerate}[label={\bfseries\color{primary}Step \arabic*.}]
\item Graphical representation;
\item ...
\end{enumerate}
\section{Parameter Estimation}
\subsection{Simple Linear Regression}
\subsection{General Case}
If $\X^\T\X$ is invertible, the OLS estimator is:
\begin{equation}
\hat{\beta} = (\X^\T\X)^{-1} \X^\T \Y
\end{equation}
\subsection{Ordinary Least Square Algorithm}
We want to minimize the distance between $\X\beta$ and $\Y$:
\[
\min \norm{\Y - \X\beta}^2
\]
(See \autoref{ch:elements-of-linear-algebra}).
\begin{align*}
\Rightarrow& \X \beta = proj^{(1, \X)} \Y\\
\Rightarrow& \forall v \in w,\, vy = v proj^w(y)\\
\Rightarrow& \forall i: \\
& \X_i \Y = \X_i X\hat{\beta} \qquad \text{where $\hat{\beta}$ is the estimator of $\beta$} \\
\Rightarrow& \X^\T \Y = \X^\T \X \hat{\beta} \\
\Rightarrow& {\color{red}(\X^T \X)^{-1}} \X^\T \Y = {\color{red}(\X^T \X)^{-1}} (\X^T\X) \hat{\beta} \\
\Rightarrow& \hat{\beta} = (X^\T\X)^{-1} \X^\T \Y
\end{align*}
This formula comes from the orthogonal projection of $\Y$ on the subspace define by the explicative variables $\X$
$\X \hat{\beta}$ is the closest point to $\Y$ in the subspace generated by $\X$.
If $H$ is the projection matrix of the subspace generated by $\X$, $X\Y$ is the projection on $\Y$ on this subspace, that corresponds to $\X\hat{\beta}$.
\section{Coefficient of Determination: $R^2$}
\begin{definition}[$R^2$]
\[
0 \leq R^2 = \frac{\norm{\X\hat{\beta} - \bar{\Y}\One}^2}{\norm{\Y - \bar{\Y}\One}^2} = 1 - \frac{\norm{\Y - \X\hat{\beta}}^2}{\norm{\Y - \bar{\Y}\One}^2} \leq 1
\] proportion of variation of $\Y$ explicated by the model.
\end{definition}

View File

@ -0,0 +1,212 @@
\chapter{Elements of Linear Algebra}
\label{ch:elements-of-linear-algebra}
\begin{remark}[vector]
Let $u$ a vector, we will use interchangeably the following notations: $u$ and $\vec{u}$
\end{remark}
Let $u = \begin{pmatrix}
u_1 \\
\vdots \\
u_n
\end{pmatrix}$ and $v = \begin{pmatrix}
v_1 \\
\vdots \\
v_n
\end{pmatrix}$
\begin{align*}
\langle u, v\rangle & = \left(u_1, \ldots, u_v\right) \begin{pmatrix}
v_1 \\
\vdots \\
v_n
\end{pmatrix} \\
& = u_1 v_1 + u_2 v_2 + \ldots + u_n v_n
\end{align*}
\begin{definition}[Norm]
Length of the vector.
\[
\norm{u} = \sqrt{\scalar{u, v}}
\]
$\norm{u, v} > 0$
\end{definition}
\begin{definition}[Distance]
\[
dist(u, v) = \norm{u-v}
\]
\end{definition}
\begin{definition}[Orthogonality]
\[
u \perp v \Leftrightarrow \scalar{u, v} = 0
\]
\end{definition}
\begin{remark}
\[
(dist(u, v))^2 = \norm{u - v}^2,
\] and
\[
\scalar{v-u, v-u}
\]
\end{remark}
Scalar product properties:
\begin{itemize}
\item $\scalar{u, v} = \scalar{v, u}$
\item $\scalar{(u+v), w} = \scalar{u, w} + \scalar{v, w}$
\item $\scalar{u, v}$
\item $\scalar{\vec{u}, \vec{v}} = \norm{\vec{u}} \times \norm{\vec{v}} \times \cos(\widehat{\vec{u}, \vec{v}})$
\end{itemize}
\begin{align*}
\scalar{v-u, v-u} & = \scalar{v, v} + \scalar{u, u} - 2 \scalar{u, v} \\
& = \norm{v}^2 + \norm{u}^2 \\
& = -2 \scalar{u, v}
\end{align*}
\begin{align*}
\norm{u - v}^2 & = \norm{u}^2 + \norm{v}^2 - 2 \scalar{u,v} \\
\norm{u + v}^2 & = \norm{u}^2 + \norm{v}^2 + 2 \scalar{u,v}
\end{align*}
If $u \perp v$, then $\scalar{u, v} = 0$
\begin{proof}[Indeed]
$\norm{u-v}^2 = \norm{u+v}^2$,
\begin{align*}
\Leftrightarrow & -2 \scalar{u, v} = 2 \scalar{u, v} \\
\Leftrightarrow & 4 \scalar{u, v} = 0 \\
\Leftrightarrow & \scalar{u, v} = 0
\end{align*}
\end{proof}
\begin{theorem}{Pythagorean theorem}
If $u \perp v$, then $\norm{u+v}^2 = \norm{u}^2 + \norm{v}^2$ .
\end{theorem}
\begin{definition}[Orthogonal Projection]
\end{definition}
Let $y = \begin{pmatrix}
y_1 \\
. \\
y_n
\end{pmatrix} \in \RR[n]$ and $w$ a subspace of $\RR[n]$
$\mathcal{Y}$ can be written as the orthogonal projection of $y$ on $w$:
\[
\mathcal{Y} = proj^w(y) + z,
\]
where
\[
\begin{cases}
z \in w^\perp \\
proj^w(y) \in w
\end{cases}
\]
There is only one vector $\mathcal{Y}$ that ?
The scalar product between $z$ and (?) is zero.
\begin{property}
$proj^w(y)$ is the closest vector to $y$ that belongs to $w$.
\end{property}
\begin{definition}[Matrix]
A matrix is an application, that is, a function that transform a thing into another, it is a linear function.
\end{definition}
\begin{example}[Matrix application]
Let $A$ be a matrix:
\[
A = \begin{pmatrix}
a & b \\
c & d
\end{pmatrix}
\] and
\[
x = \begin{pmatrix}
x_1 \\
x_2
\end{pmatrix}
\]
Then,
\begin{align*}
Ax & = \begin{pmatrix}
a & b \\
c & d
\end{pmatrix}
\begin{pmatrix}
x_1 \\
x_2
\end{pmatrix} \\
& = \begin{pmatrix}
a x_1 + b_x2 \\
c x_1 + d x_2
\end{pmatrix}
\end{align*}
Similarly,
\begin{align*}
\begin{pmatrix}
a & b & c & d \\
e & f & g & h \\
i & j & k & l
\end{pmatrix}
\begin{pmatrix}
x_1 \\
x_2 \\
x_3 \\
x_4
\end{pmatrix}
& = \begin{pmatrix}
a x_1 + b x_2 + c x_3 \ldots
\end{pmatrix}
\end{align*}
\end{example}
The number of columns has to be the same as the dimension of the vector to which the matrix is applied.
\begin{definition}[Tranpose of a Matrix]
Let $A = \begin{pmatrix}
a & b \\
c & d
\end{pmatrix}$, then $A^\T = \begin{pmatrix}
a & c \\
b & d
\end{pmatrix}$
\end{definition}
\begin{example}
\begin{align*}
Y & = X \beta + \varepsilon \\
\begin{pmatrix}
y_1 \\
y_2 \\
y_3 \\
y_4
\end{pmatrix}
& = \begin{pmatrix}
1 & x_{11} & x_{12} \\
1 & x_{21} & x_{22} \\
1 & x_{31} & x_{32} \\
1 & x_{41} & x_{42}
\end{pmatrix}
\begin{pmatrix}
\beta_0 \\
\beta_1 \\
\beta_2
\end{pmatrix} +
\begin{pmatrix}
\varepsilon_1 \\
\varepsilon_2 \\
\varepsilon_3 \\
\varepsilon_4
\end{pmatrix}
\end{align*}
\end{example}

View File

View File

@ -0,0 +1,25 @@
\chapter{Introduction}
\begin{definition}[Long Term Nonprocessor (LTNP)]
Patient who will remain a long time in good health condition, even with a large viral load (cf. HIV).
\end{definition}
\begin{example}[Genotype: Qualitative or Quantitative?]
\[
\text{SNP}:
\begin{cases}
\text{AA} \\
\text{AB} \\
\text{BB}
\end{cases}
\rightarrow
\begin{pmatrix}
0 \\
1 \\
2
\end{pmatrix},
\]
thus we might consider genotype either as a qualitative variable or quantitative variable.
\end{example}
When the variable are quantitative, we use regression, whereas for qualitative variables, we use an analysis of variance.

Binary file not shown.

View File

@ -0,0 +1,12 @@
\documentclass[tikz]{standalone}
\usepackage{tikz}
\usepackage{tkz-euclide}
\begin{document}
\begin{tikzpicture}
\tkzInit[xmax=5,ymax=5,xmin=-5,ymin=-5]
\tkzGrid
\tkzAxeXY
\draw[thick, latex-latex] (-1,4) -- (4,-6) node[anchor=south west] {$a$};
\end{tikzpicture}
\end{document}

BIN
main.pdf Normal file

Binary file not shown.

View File

@ -17,23 +17,23 @@
\usepackage{mus}
\titlehead{GENIOMHE}
\title{Multivariate Statistics}
\title{Multivariate\newline{}Statistics}
\author{Samuel Ortion}
\teacher{Cyril Dalmasso}
\cursus{GENIOMHE}
\university{Université Paris-Saclay, Université d'Évry val d'Essonne}
\semester{M1 - S1}
\date{}
\date{Fall 2023}
\definecolor{myblue}{HTML}{5654fa}
\colorlet{primary}{myblue}
\input{definitions}
\input{preamble}
\hypersetup{
pdftitle={
Course - Multivariate Statistics
},
pdfauthor={
Samuel Ortion
},
pdftitle={Course - Multivariate Statistics},
pdfauthor={Samuel Ortion},
pdfsubject={},
pdfkeywords={},
pdfcreator={LaTeX}

View File

@ -0,0 +1,3 @@
\usepackage{pgffor}
\usetikzlibrary{math}
\usepackage{standalone}