机器学习-数学基础2：字母表，二叉树

本贴是对闵老师的博客的理解

一.字母表

1.1字母表

常见的字母表包括:
Σ = { 0 , 1 } \Sigma= \{0, 1\} Σ={0,1}
Σ = { a , … , z } \Sigma = \{\mathrm{a}, \dots, \mathrm{z}\} Σ={a,…,z}

我们需要的字母表：
Σ = { l , r } \Sigma = \{\mathrm{l},\mathrm{r}\} Σ={l,r}

1.2正闭包

The positive closure of alphabet is given by ${\Sigma }^{+}={\Sigma }^1\cup {\Sigma }^2\cup ...$

Σ 1 = { l , r } \Sigma^1=\{\mathrm{l},\mathrm{r}\} Σ1={l,r}
Σ 2 = { l l , l r , r r , r l } \Sigma^2=\{\mathrm{ll},\mathrm{lr},\mathrm{rr},\mathrm{rl}\} Σ2={ll,lr,rr,rl}
$\dots$
Σ + = { l , r , l l , l r , r l , r r , l l l , …   } \Sigma^+ = \{\mathrm{l}, \mathrm{r}, \mathrm{ll}, \mathrm{lr}, \mathrm{rl}, \mathrm{rr}, \mathrm{lll}, \dots \} Σ+={l,r,ll,lr,rl,rr,lll,…}

1.3克林闭包

空串 $\epsilon$ (varepsilon)
The Cling closure of alphabet Σ ∗ = Σ 0 ∪ Σ + = { ε } ∪ Σ + \Sigma^* = \Sigma^0 \cup \Sigma^+ = \{\varepsilon\} \cup \Sigma^+ Σ∗=Σ0∪Σ+={ε}∪Σ+
字母表克林闭包的元素, 就称为字符串

1.4跳转函数

为了描述在某个状态接受字符串的跳转, 可定义跳转函数为:
Let $\mathbf{S}$ denote the set of states. The positive state transition function is given by $f:\mathbf{S}\times {\Sigma }^{+}\to \mathbf{S}$ , where $\forall s\in \mathbf{S}$ and $a_1{a}_2\dots a_k\in {\Sigma }^{+}$,
$f(s,a_1{a}_2\dots a_k)=f(f(s,a_1),a_2{a}_3\dots a_k)$

二.二叉树

2.1初始版本

Let Σ = { l , r } \Sigma = \{\mathrm{l}, \mathrm{r}\} Σ={l,r}be the alphbet and $\varphi$ be a null node. A binary tree is a triple $T=(\mathbf{V},r,c)$, where V = { v 1 , … , v n } \bm{V} = \{v_1, \dots, v_n\} V={v1​,…,vn​} is the set of nodes, $r\in \mathbf{V}$ is the root, and c : V ∪ { ϕ } × Σ + → V ∪ { ϕ } c: \bm{V} \cup \{\phi\} \times \Sigma^+ \to \bm{V} \cup \{\phi\} c:V∪{ϕ}×Σ+→V∪{ϕ} satisfying
a) $c(\varphi ,\mathrm{l})=c(\varphi ,\mathrm{r})=\varphi$;
b) ∀ v ∈ V ∖ { r } \forall v \in \bm{V} \setminus \{r\} ∀v∈V∖{r}, $\exists 1$ $s\in {\Sigma }^{+}$ st. $(\mathrm{r},\mathrm{s})=v$;
c) $\forall v\in \mathbf{V}$, $a\in \Sigma$ , $c(\mathrm{v},\mathrm{a})\ne r$

这是一个三元组。
b）代表从根节点到任意节点的路径仅存在唯一的路径
c) 代表根节点没有父节点

注：由b) 和 c)我们推出了一个性质 $\mathbf{V}$中节点没有环

2.2打磨版本

2.2.1定义

Let Σ = { l , r } \Sigma = \{\mathrm{l}, \mathrm{r}\} Σ={l,r}be the alphbet and $\varphi$ be a null node. A binary tree is a triple $T=(\mathbf{V},r,c)$, where V = { v 1 , … , v n } \bm{V} = \{v_1, \dots, v_n\} V={v1​,…,vn​} is the set of nodes, $r\in \mathbf{V}$ is the root, and c : V ∪ { ϕ } × Σ ∗ → V ∪ { ϕ } c: \bm{V} \cup \{\phi\} \times \Sigma^* \to \bm{V} \cup \{\phi\} c:V∪{ϕ}×Σ∗→V∪{ϕ} satisfying
$\forall v\in \mathbf{V}$, $\exists 1$ $s\in {\Sigma }^{\ast }$ st. $(\mathrm{r},\mathrm{s})=v$;

a)和c)都是冗余的，但是只保留b),会出现一种特殊情况

所以我们引入了空串， Σ ∗ = Σ 0 ∪ Σ + = { ε } ∪ Σ + \Sigma^* = \Sigma^0 \cup \Sigma^+ = \{\varepsilon\} \cup \Sigma^+ Σ∗=Σ0∪Σ+={ε}∪Σ+
$c(r,\epsilon )=r$. 即从 $r$ 读入空串到自己

2.2.2性质

①： 二叉树的任何节点 (空节点除外) 不会有到自己的环.

Property 1. $\forall v\in \mathbf{V}$, $!\exists s\in {\Sigma }^{+}$,st.$c(v,s)=v$.
Proof. Suppose that $\exists v_i\in \mathbf{V}$ and $s^{\prime }\in {\Sigma }^{+}$,st. $c(v_i,s^{\prime })=v_i$
According to Definition 17, $\exists s_1\in {\Sigma }^{\ast }$ st. $c(r,s_1)=v_i$.
Consequentyly $c(r,s_{1}s\prime )=c(c(r,s_1),s\prime )=c(vi,s\prime )=vi$ , and $s$ takes at least two values $(s1$ and $s1s\prime )$, making it not unique.
This contradition shows that the assumption does not hold.
The proof is finished.

②：空节点的左右孩子都是自己

Property 1. $c(\varphi ,\mathrm{l})=c(\varphi ,\mathrm{r})=\varphi$.
Proof. $s=a_1{a}_2\dots a_{n+1}\in {\Sigma }^{\ast }$ , we consider$c(r,s)$. Let the path corresponding to the calculation of $c(r,s)$ be $v_0^{\prime }{v}_1^{\prime }\dots v_{n+1}^{\prime }$ where $v_0^{\prime }=r$. Since ∣ V ∪ { ϕ } ∣ = n + 1 |\bm{V} \cup \{\phi\}| = n + 1 ∣V∪{ϕ}∣=n+1, according to the Pigeon Cage Principle (鸽笼原理), there must $\exists 0\le i<j\le n+1$ st. $v_i^{\prime }=v_j^{\prime }$ . In other words,$v_i^{\prime }\dots v_j^{\prime }$ is a loop.
Now assume that $\exists i<k<j$ st. $v_k^{\prime }\in \mathbf{V}$. We have $c(r,a_1{a}_2\dots a_k)=v_k^{\prime }$ , and $c(r,a_1{a}_2\dots a_j{a}_{i+1}{a}_{i+2}\dots a_k)=v_k^{\prime }$ , making the path from $r$ to $v_k^{\prime }$ not unique.
Hence the assumption does not hold, and $v_i^{\prime }=v_{i+1}^{\prime }=\cdots =v_j^{\prime }=\varphi$.
In other words, any character takes $\varphi$ to itself.
This completes the proof.
说明:
a) 鸽笼原理: $n+1$只鸽子飞进$n$个鸽笼, 至少有两个鸽子在同一个笼子里. 这是组合数学中重要的定理.
b) 这里用到了有穷状态自动机 (Finite state automata) 的知识. 从任一节点 (状态), 读入一个字符, 到达下一个节点. 这里的 $\varphi$被称为 陷井状态.

1)开始在一个节点，读入n+1个字符，就会经过n+2个节点。有效集合加上{$\varphi$}一共n+1个节点。由鸽笼原理。得知肯定由两个节点相同。两个节点相同由性质1可知 二叉树的任何节点 (空节点除外) 不会有到自己的环，所以这两个相同的点肯定是$\varphi$.
2)接着我们探讨$v_i,v_j$是否存在有效节点。我们假设在i，j中存在有效节点k。由于$v_k^{\prime }$属于有效节点，再由b)可知$c(r,a_1{a}_2\dots a_k)=v_k^{\prime }$ 。 $c(r,a_1{a}_2\dots a_j{a}_{j+1}{a}_{j+2}\dots a_k)=v_k^{\prime }$ = $c(r,a_1{a}_2\dots a_i{a}_{i+1}{a}_{i+2}\dots a_k)=v_k^{\prime }$，这就导致路径不唯一与定义相矛盾。为了防止不冲突，我们只有当读取到$a_i$或者$a_j$时是$\varphi$，后面$a_{i+1}\dots$或者$a_{j+1}\dots$都是$\varphi$节点是才可以。即$v_i^{\prime }=v_{i+1}^{\prime }=\cdots =v_j^{\prime }=\varphi$.证明了空节点自成一个环路