Preface

既然认清了自己不是寄系人的事实, 我发现我已经飘了, 这下是复习不完了. 不过你要说这是啥整活操作… 好像也是欸.

(注: 因为时间来不及, 所以一些没有写代码. )

代码

;;; Data Struct
(defpackage :data-struct
  (:use :cl))

(in-package :data-struct)

总的代码应该会放在 这个 地方.

一些帮助代码

虽然叫做帮助代码, 但是可能没啥帮助作用, 折叠了

杂项

生成一个比较函数:

(defun eq-to (value)
  "Generate a compare func to test if equal to VALUE."
  (lambda (another) (eq value another)))

根据规则提取列表中的元素:

(defun pick-elem-of (lst &optional (rule 'random))
  "Pick elem of LST by RULE."
  (if (null lst)
      NIL
      (cond ((eq rule 'first)  (first lst))
            ((eq rule 'last)   (car (last lst)))
            ((eq rule 'middle) (pick-elem-of 2))
            ((eq rule 'random) (nth (random (length lst)) lst))
            ((integerp rule)   (nth (floor (/ (length lst) rule)) lst))
            (T (pick-elem-of lst 'random)))))

过滤函数:

(defun filter (func lst &key (map NIL))
  "Filter a LST by FUNC."
  (loop for elem in lst
        if (funcall func elem)
          collect (if map (funcall map elem) elem) into success
        else
          collect (if map (funcall map elem) elem) into fail
        finally (return (values success fail))))

列表

将一个列表中的元素使用某个元素连接在一起:

(defun join (lst &optional (spliter "~%"))
  "Join elements in LST, splitted by SPLITER."
  (cond ((null lst) "")
        ((= (length lst) 1) (format nil "~A" (first lst)))
        (T (format nil "~A~A~A"
              (first lst)
              (format nil spliter)
              (join (cdr lst) spliter)))))

矩阵计算

将矩阵进行一个转置操作:

(defun transpose (matrix)
  (apply #'mapcar #'list matrix))

或者提取出矩阵中的某个元素:

(defmacro at (nested &rest indexs)
  "Nested nth for NESTED via INDEXS. 
For example: (at matrix row col) -> (nth col (nth row matrix))"
  (if (null indexs)
      nested
      (cons 'at (cons `(nth ,(car indexs) ,nested) (cdr indexs)))))

树的表示

一棵树, 需要有节点和节点的叶子.

假设一颗树通过一个链表来进行表现:

(defvar *tree-example*
  '(1
    (2 (5 6) 7)
    (3 (8 9 10) 11)
    (4 12 13 14)))
数据结构的一些读取

测试一个元素是否为树的叶子:

(defun leafp (elem)
  "Test if an ELEM is a leaf."
  (atom elem))

读取节点的值:

(defun tree-node (elem)
  "Read node info from ELEM."
  (if (leafp elem)
      elem
      (first elem)))

读取节点的叶子:

(defun tree-leaf (elem)
  "Read leaves of ELEM."
  (if (leafp elem)
      NIL
      (rest elem)))

历遍树

倘若想要历遍一颗树, 通常有两种方法. (对于二叉树, 实际上还会多一种先序遍历. )

(defun pre-order (tree func &optional (parent NIL))
  "Pre-order iterate the TREE via FUNC."
  (let ((node (tree-node tree))
        (leaf (tree-leaf tree)))
    (cons
     (funcall func node parent)
     (loop for elem in leaf
           if (leafp elem)
             collect (funcall func elem node)
           else
             collect (pre-order elem func node)))))

或者是后序遍历:

(defun post-order (tree func &optional (parent NIL))
  "Post-order iterate the TREE via FUNC."
  (let* ((node (tree-node tree))
         (leaf (tree-leaf tree))
         (leaves (loop for elem in leaf
                       if (leafp elem)
                         collect (funcall func elem node)
                       else
                         collect (post-order elem func node))))
    (cons (funcall func node parent) leaves)))
绘制代码
(defun draw-tree (tree &key
                         (labelfn NIL)
                         (iter #'pre-order)
                         (caller NIL)
                         (embedding NIL)
                         (headers '("node [shape=\"circle\"];")))
  "Turn Tree to Graphviz code."
  (labels ((node-name (node)
             (if (atom node) node (join node "_"))))
    (when embedding (format t "digraph {~%"))
    (format t "~A" (join headers))
    (funcall iter tree
             (lambda (node parent)
               (format t "node_~A [label=\"~A\"];~%"
                       (node-name node)
                       (if labelfn (funcall labelfn node) (node-name node)))
               (when parent
                 (format t "node_~A -> node_~A;~%"
                         (node-name parent) (node-name node)))
               (when caller
                 (funcall caller node parent))))
    (when embedding (format t "}"))))

/_img/pieces/data-struct/tree-example.svg

一些绘图的代码
(let ((previous NIL)
      (counter 1))
  (draw-tree
   *tree-example*
   :iter #'pre-order
   :caller (lambda (node -)
             (when previous
               (format t "node_~A -> node_~A [label=\"~A\",color=grey, constraint=false, style=dashed];~%"
                       previous node counter)
               (incf counter))
               (setq previous node))))

/_img/pieces/data-struct/pre-order-iter-tree.svg

(let ((previous NIL)
      (counter 1))
  (draw-tree
   *tree-example*
   :iter #'post-order
   :caller (lambda (node -)
             (when previous
               (format t "node_~A -> node_~A [label=\"~A\",color=grey, constraint=false, style=dashed];~%"
                       previous node counter)
               (incf counter))
             (setq previous node))))

/_img/pieces/data-struct/post-order-iter-tree.svg

(m, M) 树

为了让树的节点更加的统一, 加入修正 (约束) 使得节点数量有限, 比如最大叶子数不超过 M 个, 最小叶子数不少于 m 个的树, 并且修改节点为一个链表, 表现叶子和节点链表关系, 称为 m-M-树.

比如一个节点为 (2 4), 那么其叶子为 1, 3, 5, 即数据表示如下:

(defvar *m-M-tree-example*
  '((10 20 27)
    ((5))
    ((17))
    ((23 25)
     ((21)) ((24)) ((26)))
    ((28))))

/_img/pieces/data-struct/m-M-tree-example.svg

查找

实际上和 \(k\) 分法很像, 因为节点 \(n\) 有数据 \((a_1^n, a_2^n, …, a_l^n), m \leq l \leq M\), 于是对于要查找的值 \(v\), 若不在 \(a_j^n\) 中, 则应落在 \(a_i^n, a_{i+1}^n\) 之间, 于是去查找该节点的第 \(i\) 个节点:

(defun search-m-M-tree (value tree)
  "Search VALUE in TREE (m-M-tree)."
  (let ((node (first tree))
        (idx  0))
    (if (null node)
        NIL                             ; Not Found
        (if (find value node)
            value
            (progn
              (loop while (and (nth idx node)
                               (< (nth idx node) value))
                    do (incf idx))
              (search-m-M-tree value (nth (1+ idx) tree)))))))

插入和删除

插入和删除需要遇到的一个问题是节点的调整:

  1. 若操作后节点 \(n\) 的数据数量 \(l\) 小于 \(m\), 则需要从其他节点处借用
  2. 若操作后节点 \(n\) 的数据数量 \(l\) 大于 \(M\), 则需要分散到其他节点处

B 树和 B+ 树

在正经 m-M 树上做更多的约束, 和一个链表对应, 但 B+ 树里面放的不是数据, 而是链表的索引.

红黑树

可以参考 Wikipedia. 在平衡树的基础上加上颜色, 满足黑色节点高度的平衡.

胜者树和败者树

胜者树和败者树与堆很像.

图的表示

  • 可以用邻接矩阵来表示图:
    (defvar *matrix-graph-example*
      '((NIL A   B   C   D)
        (A   NIL T   T   NIL)
        (B   NIL NIL T   NIL)
        (C   NIL NIL NIL T)
        (D   NIL T   NIL NIL)))
    
    一些访问和读取
    • 绘制矩阵对应的图:
      (defun draw-matrix-graph (graph-matrix &key
                                               (embedding NIL)
                                               (headers '("layout=fdp;"
                                                          "node [shape=\"circle\"];")))
        "Trun GRAPH-MATRIX into digraph."
        (when embedding (format t "digraph {~%"))
        (format t "~A~%" (join headers))
        (let ((nodes (rest (first graph-matrix))))
          (loop for node in nodes do
            (format t "node_~A [label=\"~A\"];~%" node node))
          (loop for from-node in nodes
                for row from 1 do
                  (loop for to-node in nodes
                        for col from 1 do
                        (when (at graph-matrix row col)
                          (format t "node_~A -> node_~A;~%"
                                   from-node to-node)))))
        (when embedding (format t "}")))
      

    /_img/pieces/data-struct/matrix-graph-draw.svg

  • 可以用邻接链表来表示图:
    (defvar *arc-graph-example*
      '(:nodes (A B C D)
        :arcs ((A B NIL)
               (B C NIL)
               (C A NIL)
               (C D NIL))))
    
    一些辅助代码
    (defun draw-arc-graph (arc-graph &key
                                       (embedding NIL)
                                       (headers '("layout=fdp;"
                                                  "node [shape=\"circle\"];")))
      "Trun ARC-GRAPH into digraph."
      (when embedding (format t "digraph{~%"))
      (format t "~A~%" (join headers))
      (mapcar (lambda (node) (format t "node_~A [label=\"~A\"];~%" node node))
              (getf arc-graph :nodes))
      (mapcar (lambda (arcs) (format t "node_~A -> node_~A [label=\"~A\"];~%"
                                     (first arcs) (second arcs)
                                     (if (nth 2 arcs) (nth 2 arcs) "")))
              (getf arc-graph :arcs))
      (when embedding (format t "}")))
    

    /_img/pieces/data-struct/arc-graph-draw.svg

一些辅助代码

读取所有的节点:

(defun nodes-of-graph (graph &key (type 'arc-graph))
  "Find all nodes of GRAPH, in TYPE."
  (cond ((eq type 'arc-graph)    (getf graph :nodes))
        ((eq type 'matrix-graph) (rest (first graph)))
        (T NIL)))

读取当前节点的邻接边:

(defun next-arcs-of (graph node &key (type 'arc-graph))
  "Find all next arcs of NODE in GRAPH, in TYPE."
  (cond ((eq type 'arc-graph)    (let ((arcs (getf graph :arcs)))
                                   (remove-if-not (eq-to node) arcs :key #'first)))
        ((eq type 'matrix-graph) (let ((header (rest (first graph)))
                                       (filter (rest (assoc node (rest graph)))))
                                   (loop for next in header
                                         for weight in filter
                                         if weight
                                           collect (list node next weight))))
        (T NIL)))

图的历遍

主要有两种: 深度优先和广度优先. (主要以顶点边的表示为主)

  • 深度优先 \(O(n + e)\), 其中 \(n\) 为节点个数, \(e\) 为边个数 (近似为 \(n^2\))
    (defun depth-first-iter (func graph &key
                                          (arc-func NIL)
                                          (start NIL)
                                          (type 'arc-graph))
      "Depth first iteration, apply FUNC to GRAPH nodes."
      (let* ((nodes (copy-list (nodes-of-graph graph :type type)))
             (begin (if start start (first nodes)))
             (collection '()))
        (labels ((depth-first (node)
                   (when (find node nodes)
                     ;; Eval node
                     (push (funcall func node) collection)
                     (setq nodes (remove node nodes))
                     ;; Search next immediately
                     (loop for arc in (next-arcs-of graph node :type type) do
                       (progn
                         ;; (arc-func arc back-arc?)
                         (if arc-func (funcall arc-func arc (find (second arc) nodes)))
                         (depth-first (second arc)))))))
          ;; If remain nodes, keeping searching.
          (loop while (not (null nodes))
                do (let ((node (if (find begin nodes) begin (first nodes))))
                     (depth-first node)))
          collection)))
    
  • 广度优先
    (defun breadth-first-iter (func graph &key
                                            (arc-func NIL)
                                            (start NIL)
                                            (type 'arc-graph))
      "Breadth first iteration, apply FUNC to GRAPH nodes."
      (let* ((nodes (copy-list (nodes-of-graph graph :type type)))
             (begin (if start start (first nodes)))
             (collection '()))
        (labels ((breadth-first (&rest to-search)
                   (let ((nexts '()))
                     ;; Next to search
                     (loop for node in to-search
                           if (find node nodes) do
                             (progn
                               ;; Eval nodes
                               (push (funcall func node) collection)
                               (setq nodes (remove node nodes))
                               ;; Add next to search nodes
                               (loop for arc in (next-arcs-of graph node :type type)
                                     do (progn
                                          (when arc-func
                                            (funcall arc-func
                                                     arc (find (second arc) nodes)))
                                          (when (find (second arc) nodes)
                                            (push (second arc) nexts))) )))
                     ;; If there's nodes next to search
                     (if nexts (apply #'breadth-first nexts)))))
          (loop while (not (null nodes))
                do (let ((node (if (find begin nodes) begin (first nodes))))
                     (breadth-first node)))
          collection)))
    
  • 一些应用:
    • 历遍的边收集起来就是一个生成树 (并不是最小的)
      (defun iter-generate-tree (graph &key
                                         (iter #'depth-first-iter)
                                         (type 'arc-graph))
        "Generate a tree of GRAPH by DEPTH-FIRST-ITER or BREADTH-FIRST-ITER."
        (let ((nodes (nodes-of-graph graph :type type))
              (back-arcs '())
              (arcs '()))
          (funcall iter #'identity graph
                   :type type
                   :arc-func (lambda (arc back?)
                               (if back?
                                   (push arc arcs)
                                   (push arc back-arcs))))
          (values (list :nodes nodes :arcs arcs)
                  back-arcs)))
      
    • 历遍的时候遇到的返回边就是一个环
      (defun count-ring-of (graph &key
                                    (iter #'depth-first-iter)
                                    (type 'arc-graph))
        "Count ring number of a GRAPH by DEPTH-FIRST-ITER or BREADTH-FIRST-ITER."
        (let ((count 0))
          (funcall iter #'identity graph
                   :type type
                   :arc-func (lambda (- back?)
                               (when (not back?) (incf count))))
          count))
      
    • 可以通过历遍来判断是否是连通图 (若存在中断, 则为非连通图)

图的生成树

实际上关心的是最小生成树 (Prim 算法). (和 Dijkstra 算法比较类似)

  • 初始起点出发, 扔到 Blue Cloud 中
  • 对于从 Blue Cloud 中节点能够出发到达的新的节点, 选择最短的边能到的节点并加入到 Blue Cloud 中.
  • 循环直到没有剩下的节点

图的最短路径

Dijkstra 和 Floyd 算法

Dijkstra

核心是一个 Blue Cloud 的扩张和松弛边:

  • 初始将所有的节点的距离设为 \(∞\), 而起点设为 \(0\)
  • 将起点添加到 Blue Cloud 中, 并更新邻近的节点的距离为当前节点距离加上边权重, 并且更新邻近的顶点的距离
  • 将最小的 (非 Blue Cloud 内的) 顶点添加到 Blue Cloud 中并重复操作.

Floyd

节点 \(i\) 到 \(j\) 之间的最短距离用 \(d_{ij}\) 表示, 则取一个中间节点 \(k\), 那么 \(d_{ij} = d_{ik} + d_{kj}\ \mathrm{iif}\ d_{ij} < d_{ik} + d_{kj}\).

(defun floyd-compare (a b)
  "Compare A and B in Floyd Algorithm."
  (if (and a b) (< a b)                 ; a < b if
      (if a NIL T)))                    ; a = NIL(infinity) => NIL

(defun floyd-shotest-path (graph-matrix &key
                                          (compare #'floyd-compare))
  "Floyd shortest path."
  (let ((matrix (copy-list graph-matrix))
        (max (1- (length graph-matrix))))
    (labels ((add (a b)
               (if (and a b) (+ a b) NIL)))
      (loop for mid from 1 to max do
        (loop for start from 1 to max do
          (loop for end from 1 to max
                if (funcall compare
                            (at matrix start end)
                            (add (at matrix start mid)
                                 (at matrix mid start)))
                  do (setf (at matrix start end)
                           (add (at matrix start mid)

                                (at matrix mid start)))))))
    matrix))

(注: compare 还有点问题, 没时间改了. )

拓扑序

算法应该不只一种, 提供一种倒着数的方法: 选择没有出边的节点, 删除节点和相邻的边并且对该节点标号, 然后循环直到节点为空.

排序

  • \(O(n^2)\): 冒泡, 选择, 插入 (希尔)
  • \(O(n log n)\): 归并, 堆, 快排
  • \(O(n+k)\): 桶 (基数)

内部排序

插入排序

每次都选择未排序的部分中最符合的部分, 然后插入到排序好了的部分的前面:

(defun insert-sort (lst &key
                          (key #'identity)
                          (compare #'<))
  "Insert sort."
  (labels ((iter (sorted unsort)
             (if (null unsort)
                 sorted
                 (multiple-value-bind (pick rest)
                     (select-most unsort :key key :compare compare)
                   (iter (append sorted (list pick)) rest)))))
    (iter '() lst)))
其中的 select-most 部分
(defun select-most (lst &key
                          (key #'identity)
                          (compare #'<))
  "Select the most element in LST."
  (let ((most NIL)
        (most-elem NIL)
        (rest '()))
    (loop for elem in lst do
      (let ((value (funcall key elem)))
        (if (not most)
            (setq most value
                  most-elem elem)
            (if (funcall compare value most)
                (progn (push most-elem rest)
                       (setq most value
                             most-elem elem))
                (push elem rest)))))
    (values most-elem (reverse rest))))

快速排序

将列表分割成大小两块, 分别进行排序.

(defun quick-sort (lst &key
                         (pivot 'random)
                         (key #'identity)
                         (compare #'<))
  "Quick sort."
  (if (or (<= (length lst) 1)
          (eq (first lst) (car (last lst))))
      lst
      (let ((pivot-value (pick-elem-of lst pivot)))
        (multiple-value-bind (less greater)
            (filter (lambda (elem)
                      (funcall compare (funcall key elem) pivot-value))
                    lst)
          (append
           (quick-sort less :pivot pivot :key key :compare compare)
           (quick-sort greater :pivot pivot :key key :compare compare))))))

桶排序和基数排序

桶排序和基数排序基本差不多 (我觉得倒是很像是 Hash 表), 可以一起理解:

  • 令映射函数 \(f\) 使得 \(a_i \mapsto f(a_i)_{\mathrm{inorder}}\), 将 \(f(a_i)\) 作为索引进行重排
  • 对于整数, 显然用 \(\mathrm{id}: a_i \mapsto a_i\) 即可满足需求, 但是因为利用率不高, 可以将其变成 \(p\) 进制的第 \(i\) 位, 然后按位排序
  • 对于非整数, 也可以差不多这么干
(defun bucket-sort (lst &key
                          (key #'identity)
                          (compare #'<))
  "Bucket Sort."
  (let ((bucket '()))
    (labels ((add-to-bucket (elem)
               (let ((value (funcall key elem)))
                 (if (assoc value bucket)
                     (push elem (cdr (assoc value bucket)))
                     (push (cons value (list elem)) bucket)))))
      (loop for elem in lst do (add-to-bucket elem))
      (apply #'append (mapcar #'rest (sort bucket compare :key #'first))))))

(注: 实际上应该将结果映射到一个有序紧凑的链表上的, 但是懒得写, 所以就这样吧.)

堆排序

堆是一种完全二叉树 记概念什么的真是麻烦.

关于堆的一些代码

假设在一个数组 (idx 从 0 开始) 上的堆结构, 那么应该有相对关系:

(defun heap-parent-idx (idx)
  "Get parent idex relavent to IDX."
  (floor (/ (1- idx) 2)))

(defun heap-left-child-idx (idx)
  "Get left child relavent to IDX."
  (+ (* 2 idx) 1))

(defun heap-right-child-idx (idx)
  "Get right child relavent to IDX."
  (+ (* 2 idx) 2))

于是可以有建堆过程:

(defmacro swap (a b)
  "Swap A and B."
  `(let ((temp ,a))
     (setf ,a ,b
           ,b temp)))

(defun make-heap-from (lst &key (compare #'<))
  "Make heap from LST."
  (let ((heap (copy-list lst))
        (change-p T))
    (loop while change-p do
      (progn
        (setq change-p NIL)
        (loop for idx from 1 upto (1- (length lst)) do
          (when (funcall compare (nth (heap-parent-idx idx) heap) (nth idx heap))
            (setq change-p T)
            (swap (nth idx heap) (nth (heap-parent-idx idx) heap))))))
    heap))

以及输出一个好看的堆的形式:

(defun draw-heap (heap &key
                         (headers '("node [shape=\"circle\"];")))
  (format t "~A~%" (join headers))
  (loop for idx from 0 to (1- (length heap)) do
      (let ((node (nth idx heap))
            (lchild (nth (heap-left-child-idx idx) heap))
            (rchild (nth (heap-right-child-idx idx) heap)))
        (format t "node_~A [label=\"~A\"];~%" node node)
        (when lchild
          (format t "node_~A -> node_~A;~%" node lchild))
        (when rchild
          (format t "node_~A -> node_~A;~%" node rchild)))))
(defvar *heap-example*
  (make-heap-from '(6 5 3 1 8 7 2 4)))
(draw-heap *heap-example*)
digraph {
  $data
}

对于一个输入:

  • 建堆 make-heap-from
  • 删除并输出第一个元素并将最后一个元素放到第一位, 对剩下的元素进行重新建堆, 循环直到剩下的为空.
(defun heap-sort (lst &key (compare #'<))
  (let ((heap (make-heap-from lst :compare compare)))
    (if (<= (length heap) 1)
        heap
        (let* ((last (car (last heap)))
               (new-heap (cons last (subseq heap 1 (1- (length heap))))))
          (cons (first heap) (heap-sort new-heap))))))

归并排序

将链表分裂成 \(n\) 个子链表, 分别排序后合并. 分别排序的方法可以是任意的排序方法. 对于多路归并的方式, 可以使用胜者树和败者树.

外部排序

实际上就像是归并排序. 原因是内存无法一次读取过多数据, 只能一点一点地读取 (相当于每次只能够读取一部分的子链表).

\(k\) 路归并

实际上就是核心部分了. 可以参考 Wikipedia. 对 \(k\) 路归并来说, 败者树是一个深度 \(d = ⌈ log k ⌉ + 1\) 的堆. 堆的底部连接待归并的链表.

End

就这样吧…