algorithm - 最快的字符串过滤算法
问题描述
我有 5,000,000 个以这种方式格式化的无序字符串(Name.Name.Day-Month-Year 24hrTime):
"John.Howard.12-11-2020 13:14"
"Diane.Barry.29-07-2020 20:50"
"Joseph.Ferns.08-05-2020 08:02"
"Joseph.Ferns.02-03-2020 05:09"
"Josephine.Fernie.01-01-2020 07:20"
"Alex.Alexander.06-06-2020 10:10"
"Howard.Jennings.07-07-2020 13:17"
"Hannah.Johnson.08-08-2020 00:49"
...
找到时间 t 在某个 n 和 m 之间的所有字符串的最快方法是什么?(即删除所有时间 < n || m < time 的字符串的最快方法)
这种过滤将在不同的范围内进行多次。时间范围必须始终在同一天,并且开始时间始终早于结束时间。
在 java 中,这是我目前的方法,给出了一些时间字符串 M 和 N 以及 500 万个字符串列表:
ArrayList<String> finalSolution = new ArrayList<>();
String[] startingMtimeArr = m.split(":");
String[] startingNtimeArr = n.split(":");
Integer startingMhour = Integer.parseInt(startingMtimeArr[0]);
Integer startingMminute = Integer.parseInt(startingMtimeArr[1]);
Integer endingNhour = Integer.parseInt(startingNtimeArr[0]);
Integer endingNminute = Integer.parseInt(startingNtimeArr[1]);
for combinedString in ArraySizeOf5Million{
String[] arr = combinedString.split(".");
String[] subArr = arr[2].split(" ");
String[] timeArr = subArr[1].split(":");
String hour = timeArr[0];
String minute = timeArr[1];
If hour >= startingMhour
&& minute >= startingMminute
&& hour <= endingNhour
&& minute <= endingNminute {
finalSolution.add(hour)
}
}
Java 是我的母语,但任何其他语言也可以。更好/更快的逻辑是我所追求的
解决方案
正如@Paddy3118 已经指出的那样,二进制搜索可能是要走的路。
- (如果您的数据在磁盘上):加载输入数据并按日期/时间排序。
- i0 是结果集的开始索引,i1 是结果集的结束索引(都是从二进制搜索获得的):枚举结果条目。
我使用的代码(在 Lisp 中)显示在此答案的末尾。它没有丝毫优化(我想通过一些优化工作可以使加载和初始排序更快)。
这就是我的交互式会话的样子(包括时间信息,我的 foo.txt 输入文件包含 500 万个条目)。
rlwrap sbcl --dynamic-space-size 2048
这是 SBCL 2.1.1.debian,ANSI Common Lisp 的实现。有关 SBCL 的更多信息,请访问http://www.sbcl.org/。SBCL 是免费软件,按原样提供,绝对不提供任何担保。它主要在公共领域;有些部分是在 BSD 风格的许可证下提供的。有关详细信息,请参阅发行版中的 CREDITS 和 COPYING 文件。
(ql:quickload :cl-ppcre)
加载“cl-ppcre”:
加载 1 个 ASDF 系统:
cl-ppcre
;加载“cl-ppcre”
..
(:CL-PPCRE)
(load "fivemillion.lisp")
T
(time (defparameter data (load-input-for-queries "foo.txt")))
"sorting..."
评估花费:
实时
32.091 秒总运行时间 32.090620 秒(31.386722 用户,0.703898 系统)
[运行时间包括 2.641 秒 GC 时间和 29.450 秒非 GC 时间。]
100.00% CPU
15 lambdas 转换
115,308,171,684 个处理器周期
6,088,198,752 字节 consed
DATA
(time (defparameter output (query-interval data '(2018 1 1) '(2018 1 2))))
评估花费:
0.000 秒的实时
0.000111 秒总运行时间(0.000109 个用户,0.000002 个系统)
100.00% CPU
395,172 个处理器周期
65,536 个字节 consed
OUTPUT
(时间(defparameteroutput (query-interval data '(2018 1 1) '(2018 1 2 8))))
评估时间:
0.000 秒的实时时间
0.000113 秒的总运行时间(0.000110 用户,0.000003 系统)
100.00% CPU
399,420 处理器周期
65,536字节 consed
OUTPUT
(time (defparameter output (query-interval data '(2018 1 1) '(2019 1 1))))
评估花费:
0.020 秒的实时
0.022469 秒的总运行时间(0.022469 用户,0.000000 系统)
110.00 % CPU
80,800,092 个处理器周期
15,958,016 字节 consed
OUTPUT
因此,虽然加载和排序时间(一次完成)没什么好写的(但可以优化),但(query-interval ...)
调用速度非常快。查询的结果集越大,函数返回的列表越长(conses 越多,运行时间越长)。我本可以更聪明,只返回结果集的开始和结束索引,并将条目的收集留给调用者。
这里是源代码,其中还包括生成我使用的测试数据集的代码:
(defun random-uppercase-character ()
(code-char (+ (char-code #\A) (random 26))))
(defun random-lowercase-character ()
(code-char (+ (char-code #\a) (random 26))))
(defun random-name-part (nchars)
(with-output-to-string (stream)
(write-char (random-uppercase-character) stream)
(loop repeat (- nchars 1) do
(write-char (random-lowercase-character) stream))))
(defun random-day-of-month ()
"Assumes every month has 31 days, because it does not matter
for this exercise."
(+ 1 (random 31)))
(defun random-month-of-year ()
(+ 1 (random 12)))
(defun random-year ()
"Some year between 2017 and 2022"
(+ 2017 (random 5)))
(defun random-hour-of-day ()
(random 24))
(defun random-minute-of-hour ()
(random 60))
(defun random-entry (stream)
(format stream "\"~a.~a.~d-~d-~d ~d:~d\"~%"
(random-name-part 10)
(random-name-part 10)
(random-day-of-month)
(random-month-of-year)
(random-year)
(random-hour-of-day)
(random-minute-of-hour)))
(defun generate-input (entry-count file-name)
(with-open-file (stream
file-name
:direction :output
:if-exists :supersede)
(loop repeat entry-count do
(random-entry stream))))
(defparameter *line-scanner*
(ppcre:create-scanner
"\"(\\w+).(\\w+).(\\d+)-(\\d+)-(\\d+)\\s(\\d+):(\\d+)\""))
;; 0 1 2 3 4 5 6
;; fname lname day month year hour minute
(defun decompose-line (line)
(let ((parts (nth-value
1
(ppcre:scan-to-strings
*line-scanner*
line))))
(make-array 7 :initial-contents
(list (aref parts 0)
(aref parts 1)
(parse-integer (aref parts 2))
(parse-integer (aref parts 3))
(parse-integer (aref parts 4))
(parse-integer (aref parts 5))
(parse-integer (aref parts 6))))))
(defconstant +fname-index+ 0)
(defconstant +lname-index+ 1)
(defconstant +day-index+ 2)
(defconstant +month-index+ 3)
(defconstant +year-index+ 4)
(defconstant +hour-index+ 5)
(defconstant +minute-index+ 6)
(defvar *compare-<-criteria*
(make-array 5 :initial-contents
(list +year-index+
+month-index+
+day-index+
+hour-index+
+minute-index+)))
(defun compare-< (dl1 dl2)
(labels ((comp (i)
(if (= i 5)
nil
(let ((index (aref *compare-<-criteria* i)))
(let ((v1 (aref dl1 index))
(v2 (aref dl2 index)))
(cond
((< v1 v2) t)
((= v1 v2) (comp (+ i 1)))
(t nil)))))))
(comp 0)))
(defun time-stamp-to-index (hours minutes)
(+ minutes (* 60 hours)))
(defun load-input-for-queries (file-name)
(let* ((decomposed-line-list
(with-open-file (stream file-name :direction :input)
(loop for line = (read-line stream nil nil)
while line
collect (decompose-line line))))
(number-of-lines (length decomposed-line-list))
(decomposed-line-array (make-array number-of-lines
:initial-contents
decomposed-line-list)))
(print "sorting...") (terpri)
(sort decomposed-line-array #'compare-<)))
(defun unify-date-list (date)
(let ((date-length (length date)))
(loop
for i below 5
collecting (if (> date-length i) (nth i date) 0))))
(defun decomposed-line-date<date-list (decomposed-line date-list)
(labels ((comp (i)
(if (= i 5)
nil
(let ((index (aref *compare-<-criteria* i)))
(let ((v1 (aref decomposed-line index))
(v2 (nth i date-list)))
(cond
((< v1 v2) t)
((= v1 v2) (comp (+ i 1)))
(t nil)))))))
(comp 0)))
(defun index-before (data key predicate
&key (left 0) (right (length data)))
(if (and (< left right) (> (- right left) 1))
(if (funcall predicate (aref data left) key)
(let ((mid (+ left (floor (- right left) 2))))
(if (funcall predicate (aref data mid) key)
(index-before data key predicate
:left mid
:right right)
(index-before data key predicate
:left left
:right mid)))
left)
right))
(defun query-interval (data start-date end-date)
"start-date and end-date are given as lists of the form:
'(year month day hour minute) or shorter versions e.g.
'(year month day hour), omitting trailing values which will be
appropriately defaulted."
(let ((d0 (unify-date-list start-date))
(d1 (unify-date-list end-date)))
(let* ((start-index (index-before
data
d0
#'decomposed-line-date<date-list))
(end-index (index-before
data
d1
#'decomposed-line-date<date-list
:left (cond
((< start-index 0) 0)
((>= start-index (length data))
(length data))
(t start-index)))))
(loop for i from start-index below end-index
collecting (aref data i)))))
推荐阅读
- c# - 如何画一个弧形箭头?
- c# - VSTO Excel 文档级别自定义无法使用“msbuild”命令构建
- python - 当我尝试将数据插入到我的 posgresql 数据库中时,什么也没有发生
- javascript - 为什么当我在复选框上触发“更改”事件时,事件对象不会更新为新状态
- java - 有没有比在 Java 中使用 imageIcon 数组更性能友好的方式来遍历动画的图像?
- c++ - 如何将 Visual Studio 与 CMake 一起使用并保留文件结构
- python - 如何在 m 长列表和 int y 的一行中获取两个输入
- php - 成对函数不同的计算
- r - R根据参考列的最小值或最大值创建包含另一列值的列
- r - 当grid.arrange()时如何对齐我的ggplots?