https://pypi.org/project/blist/

blist: an asymptotically faster list-like type for Python — blist 1.3.6 documentation http://stutzbachenterprises.com/blist/

The blist is a drop-in replacement for the Python list that provides better performance when modifying large lists. The blist package also provides sortedlistsortedsetweaksortedlistweaksortedsetsorteddict, and btuple types.

Full documentation is at the link below:

http://stutzbachenterprises.com/blist-doc/

Python’s built-in list is a dynamically-sized array; to insert or remove an item from the beginning or middle of the list, it has to move most of the list in memory, i.e., O(n) operations. The blist uses a flexible, hybrid array/tree structure and only needs to move a small portion of items in memory, specifically using O(log n) operations.

For small lists, the blist and the built-in list have virtually identical performance.

To use the blist, you simply change code like this:

>>> items = [5, 6, 2]
>>> more_items = function_that_returns_a_list()

to:

>>> from blist import blist
>>> items = blist([5, 6, 2])
>>> more_items = blist(function_that_returns_a_list())

Here are some of the use cases where the blist asymptotically outperforms the built-in list:

Use Case blist list
Insertion into or removal from a list O(log n) O(n)
Taking slices of lists O(log n) O(n)
Making shallow copies of lists O(1) O(n)
Changing slices of lists O(log n + log k) O(n+k)
Multiplying a list to make a sparse list O(log k) O(kn)
Maintain a sorted lists with bisect.insort O(log**2 n) O(n)

So you can see the performance of the blist in more detail, several performance graphs available at the following link: http://stutzbachenterprises.com/blist/

Example usage:

>>> from blist import *
>>> x = blist([0]) # x is a blist with one element
>>> x *= 2**29 # x is a blist with > 500 million elements
>>> x.append(5) # append to x
>>> y = x[4:-234234] # Take a 500 million element slice from x
>>> del x[3:1024] # Delete a few thousand elements from x

Other data structures

The blist package provides other data structures based on the blist:

  • sortedlist
  • sortedset
  • weaksortedlist
  • weaksortedset
  • sorteddict
  • btuple

These additional data structures are only available in Python 2.6 or higher, as they make use of Abstract Base Classes.

The sortedlist is a list that’s always sorted. It’s iterable and indexable like a Python list, but to modify a sortedlist the same methods you would use on a Python set (add, discard, or remove).

>>> from blist import sortedlist
>>> my_list = sortedlist([3,7,2,1])
>>> my_list
sortedlist([1, 2, 3, 7])
>>> my_list.add(5)
>>> my_list[3]
5
>>>

The sortedlist constructor takes an optional “key” argument, which may be used to change the sort order just like the sorted() function.

>>> from blist import sortedlist
>>> my_list = sortedlist([3,7,2,1], key=lambda i: -i)
sortedlist([7, 3, 2, 1]
>>>

The sortedset is a set that’s always sorted. It’s iterable and indexable like a Python list, but modified like a set. Essentially, it’s just like a sortedlist except that duplicates are ignored.

>>> from blist import sortedset
>>> my_set = sortedset([3,7,2,2])
sortedset([2, 3, 7]
>>>

The weaksortedlist and weaksortedset are weakref variations of the sortedlist and sortedset.

The sorteddict works just like a regular dict, except the keys are always sorted. The sorteddict should not be confused with Python 2.7’s OrderedDict type, which remembers the insertion order of the keys.

>>> from blist import sorteddict
>>> my_dict = sorteddict({1: 5, 6: 8, -5: 9})
>>> my_dict.keys()
[-5, 1, 6]
>>>

The btuple is a drop-in replacement for the built-in tuple. Compared to the built-in tuple, the btuple offers the following advantages:

  • Constructing a btuple from a blist takes O(1) time.
  • Taking a slice of a btuple takes O(n) time, where n is the size of the original tuple. The size of the slice does not matter.
>>> from blist import blist, btuple
>>> x = blist([0]) # x is a blist with one element
>>> x *= 2**29 # x is a blist with > 500 million elements
>>> y = btuple(x) # y is a btuple with > 500 million elements

Installation instructions

Python 2.5 or higher is required. If building from the source distribution, the Python header files are also required. In either case, just run:

python setup.py install

If you’re running Linux and see a bunch of compilation errors from GCC, you probably do not have the Python header files installed. They’re usually located in a package called something like “python2.6-dev”.

The blist package will be installed in the ‘site-packages’ directory of your Python installation. (Unless directed elsewhere; see the “Installing Python Modules” section of the Python manuals for details on customizing installation locations, etc.).

If you downloaded the source distribution and wish to run the associated test suite, you can also run:

python setup.py test

which will verify the correct installation and functioning of the package. The tests require Python 2.6 or higher.

Feedback

We’re eager to hear about your experiences with the blist. You can email me at daniel@stutzbachenterprises.com. Alternately, bug reports and feature requests may be reported on our bug tracker at: http://github.com/DanielStutzbach/blist/issues

How we test

In addition to the tests include in the source distribution, we perform the following to add extra rigor to our testing process:

  1. We use a “fuzzer”: a program that randomly generates list operations, performs them using both the blist and the built-in list, and compares the results.
  2. We use a modified Python interpreter where we have replaced the array-based built-in list with the blist. Then, we run all of the regular Python unit tests.

O(n) O(log n) blist: an asymptotically faster list-like type for Python的更多相关文章

  1. 自动统计安卓log中Anr,Crash,Singnal出现数量的Python脚本 (转载)

    自动统计安卓log中Anr,Crash,Singnal出现数量的Python脚本   转自:https://www.cnblogs.com/ailiailan/p/8304989.html 作为测试, ...

  2. 自动统计安卓log中Anr,Crash,Singnal出现数量的Python脚本

    作为测试,在测试工作中一定会经常抓log,有时log收集时间很长,导致log很大,可能达到几G,想找到能打开如此大的log文件的工具都会变得困难:即使log不大时,我们可以直接把log发给开发同学去分 ...

  3. 脚本自动统计安卓log中Anr、Crash等出现的数量(Python)

    作为测试,在测试工作中一定会经常抓log,有时log收集时间很长,导致log很大,可能达到几G,想找到能打开如此大的log文件的工具都会变得困难:即使log不大时,我们可以直接把log发给开发同学去分 ...

  4. Mininet在创建拓扑的过程中为什么不打印信息了——了解Mininet的log系统

    前言 写这篇博客是为了给我的愚蠢和浪费的6个小时买单! 过程原因分析 我用Mininet创建过不少拓扑了,这次创建的拓扑非常简单,如下图,创建拓扑的代码见github.在以前的拓扑创建过程中,我都是用 ...

  5. openstack 中 log模块分析

    1 . 所在模块,一般在openstack/common/log.py,其实最主要的还是调用了python中的logging模块: 入口函数在 def setup(product_name, vers ...

  6. Log4Net .NET log处理

    1.NuGet 安装Log4Net. 2.新建一个Common的project,并且添加一个LogWriter的类: public class LogWriter { //Error log publ ...

  7. python 每日一练: 读取log文件中的数据,并画图表

    之前在excel里面分析log数据,简直日了*了. 现在用python在处理日志数据. 主要涉及 matplotlib,open和循环的使用. 日志内容大致如下 2016-10-21 21:07:59 ...

  8. ubuntu运行命令tee显示和保存为log

    一般有三种需求: 假如我要执行一个py文件 python class.py 1:将命令输出结果保存到文件log.log python class.py |tee log.log 结果就是:屏幕输出和直 ...

  9. ORACLE LOG的管理

    CREATE OR REPLACE PACKAGE PLOG IS /** * package name : PLOG *<br/> *<br/> *See : <a h ...

随机推荐

  1. 支付宝APP支付(基于Java实现支付宝APP支付)

    贴一下支付核心代码,以供后续参考: 业务层 import com.alipay.api.AlipayApiException; import com.alipay.api.AlipayClient; ...

  2. GoogLeNet网络的Pytorch实现

    1.文章原文地址 Going deeper with convolutions 2.文章摘要 我们提出了一种代号为Inception的深度卷积神经网络,它在ILSVRC2014的分类和检测任务上都取得 ...

  3. 新闻系统——SQLHelper助手优化

    在前面我们机房重构的时候已经用过了SQLHelper,但当时就是只会应用,知道利用SQLHelper能够帮助我们连接数据库,但对于怎样书写还是不太清楚,今天就揭开这层神秘的面纱,来真正体验如何来编写S ...

  4. Robot Framework--完整的接口测试用例

    *** Settings *** Library Collections Library json Library requests Library RequestsLibrary Library H ...

  5. Spring 整合 Quartz框架(定时任务)

    Maven 无法下载 Quartz 依赖,去官网下载 http://www.quartz-scheduler.org/downloads/ Quartz 官方手册:https://www.w3csch ...

  6. Mac OS终端利器iTerm2(完美替代bash)

    iTerm2 iTerm一个为Mac OS X编写的,功能齐全的终端仿真程序:目标是在为用户提供OS X下最佳的命令行体验,iTerm 2有很多能够提升效率的实用功能.比如说窗口分割.热键窗口.智能搜 ...

  7. 通用分页jsp页面显示

    注:本章内容都是在上一篇文章 通用分页后台显示:https://www.cnblogs.com/ly-0919/p/11058942.html  的基础上进行改进,所以有许多的类都在上一篇, 带来不便 ...

  8. RabbitMQ的5种模式

    队列截图,去rabbitMq.com去找学习文档 =========================================================================== ...

  9. umeditor+粘贴word图片

    图片的复制无非有两种方法,一种是图片直接上传到服务器,另外一种转换成二进制流的base64码 目前限chrome浏览器使用,但是项目要求需要支持所有的浏览器,包括Windows和macOS系统.没有办 ...

  10. luogu 3702 [SDOI2017]序列计数 矩阵乘法+容斥

    现在看来这道题真的不难啊~ 正着求不好求,那就反着求:答案=总-全不是质数 这里有一个细节要特判:1不是质数,所以在算全不是质数的时候要特判1 code: #include <bits/stdc ...