前言
最近工作工作中遇到一個(gè)需求,是要根據(jù)CDN日志過濾一些數(shù)據(jù),例如流量、狀態(tài)碼統(tǒng)計(jì),TOP IP、URL、UA、Referer等。以前都是用 bash shell 實(shí)現(xiàn)的,但是當(dāng)日志量較大,日志文件數(shù)G、行數(shù)達(dá)數(shù)千萬億級時(shí),通過 shell 處理有些力不從心,處理時(shí)間過長。于是研究了下Python pandas這個(gè)數(shù)據(jù)處理庫的使用。一千萬行日志,處理完成在40s左右。
代碼
#!/usr/bin/python # -*- coding: utf-8 -*- # sudo pip install pandas author = 'Loya Chen' import sys import pandas as pd from collections import OrderedDict """ Description: This script is used to analyse qiniu cdn log. ================================================================================ 日志格式 IP - ResponseTime [time +0800] "Method URL HTTP/1.1" code size "referer" "UA" ================================================================================ 日志示例 [0] [1][2] [3] [4] [5] 101.226.66.179 - 68 [16/Nov/2016:04:36:40 +0800] "GET http://www.qn.com/1.jpg -" [6] [7] [8] [9] 200 502 "-" "Mozilla/5.0 (compatible; MSIE 9.0; Windows NT 6.1; Trident/5.0)" ================================================================================ """ if len(sys.argv) != 2: print('Usage:', sys.argv[0], 'file_of_log') exit() else: log_file = sys.argv[1] # 需統(tǒng)計(jì)字段對應(yīng)的日志位置 ip = 0 url = 5 status_code = 6 size = 7 referer = 8 ua = 9 # 將日志讀入DataFrame reader = pd.read_table(log_file, sep=' ', names=[i for i in range(10)], iterator=True) loop = True chunkSize = 10000000 chunks = [] while loop: try: chunk = reader.get_chunk(chunkSize) chunks.append(chunk) except StopIteration: #Iteration is stopped. loop = False df = pd.concat(chunks, ignore_index=True) byte_sum = df[size].sum() #流量統(tǒng)計(jì) top_status_code = pd.DataFrame(df[6].value_counts()) #狀態(tài)碼統(tǒng)計(jì) top_ip = df[ip].value_counts().head(10) #TOP IP top_referer = df[referer].value_counts().head(10) #TOP Referer top_ua = df[ua].value_counts().head(10) #TOP User-Agent top_status_code['persent'] = pd.DataFrame(top_status_code/top_status_code.sum()*100) top_url = df[url].value_counts().head(10) #TOP URL top_url_byte = df[[url,size]].groupby(url).sum().apply(lambda x:x.astype(float)/1024/1024) .round(decimals = 3).sort_values(by=[size], ascending=False)[size].head(10) #請求流量最大的URL top_ip_byte = df[[ip,size]].groupby(ip).sum().apply(lambda x:x.astype(float)/1024/1024) .round(decimals = 3).sort_values(by=[size], ascending=False)[size].head(10) #請求流量最多的IP # 將結(jié)果有序存入字典 result = OrderedDict([("流量總計(jì)[單位:GB]:" , byte_sum/1024/1024/1024), ("狀態(tài)碼統(tǒng)計(jì)[次數(shù)|百分比]:" , top_status_code), ("IP TOP 10:" , top_ip), ("Referer TOP 10:" , top_referer), ("UA TOP 10:" , top_ua), ("URL TOP 10:" , top_url), ("請求流量最大的URL TOP 10[單位:MB]:" , top_url_byte), ("請求流量最大的IP TOP 10[單位:MB]:" , top_ip_byte) ]) #
pandas 學(xué)習(xí)筆記
Pandas 中有兩種基本的數(shù)據(jù)結(jié)構(gòu),Series 和 Dataframe。 Series 是一種類似于一維數(shù)組的對象,由一組數(shù)據(jù)和索引組成。 Dataframe 是一個(gè)表格型的數(shù)據(jù)結(jié)構(gòu),既有行索引也有列索引。
from pandas import Series, DataFrame import pandas as pd
Series
In [1]: obj = Series([4, 7, -5, 3]) In [2]: obj Out[2]: 0 4 1 7 2 -5 3 3
Series的字符串表現(xiàn)形式為:索引在左邊,值在右邊。沒有指定索引時(shí),會(huì)自動(dòng)創(chuàng)建一個(gè)0到N-1(N為數(shù)據(jù)的長度)的整數(shù)型索引??梢酝ㄟ^Series的values和index屬性獲取其數(shù)組表示形式和索引對象:
In [3]: obj.values Out[3]: array([ 4, 7, -5, 3]) In [4]: obj.index Out[4]: RangeIndex(start=0, stop=4, step=1)
通常創(chuàng)建Series時(shí)會(huì)指定索引:
In [5]: obj2 = Series([4, 7, -5, 3], index=['d', 'b', 'a', 'c']) In [6]: obj2 Out[6]: d 4 b 7 a -5 c 3
通過索引獲取Series中的單個(gè)或一組值:
In [7]: obj2['a'] Out[7]: -5 In [8]: obj2[['c','d']] Out[8]: c 3 d 4
排序
In [9]: obj2.sort_index() Out[9]: a -5 b 7 c 3 d 4 In [10]: obj2.sort_values() Out[10]: a -5 c 3 d 4 b 7
篩選運(yùn)算
In [11]: obj2[obj2 > 0] Out[11]: d 4 b 7 c 3 In [12]: obj2 * 2 Out[12]: d 8 b 14 a -10 c 6
成員
In [13]: 'b' in obj2 Out[13]: True In [14]: 'e' in obj2 Out[14]: False
通過字典創(chuàng)建Series
In [15]: sdata = {'Shanghai':35000, 'Beijing':40000, 'Nanjing':26000, 'Hangzhou':30000} In [16]: obj3 = Series(sdata) In [17]: obj3 Out[17]: Beijing 40000 Hangzhou 30000 Nanjing 26000 Shanghai 35000
如果只傳入一個(gè)字典,則結(jié)果Series中的索引就是原字典的鍵(有序排列)
In [18]: states = ['Beijing', 'Hangzhou', 'Shanghai', 'Suzhou'] In [19]: obj4 = Series(sdata, index=states) In [20]: obj4 Out[20]: Beijing 40000.0 Hangzhou 30000.0 Shanghai 35000.0 Suzhou NaN
當(dāng)指定index時(shí),sdata中跟states索引相匹配的3個(gè)值會(huì)被找出并放到響應(yīng)的位置上,但由于‘Suzhou'所對應(yīng)的sdata值找不到,所以其結(jié)果為NaN(not a number),pandas中用于表示缺失或NA值
pandas的isnull和notnull函數(shù)可以用于檢測缺失數(shù)據(jù):
In [21]: pd.isnull(obj4) Out[21]: Beijing False Hangzhou False Shanghai False Suzhou True In [22]: pd.notnull(obj4) Out[22]: Beijing True Hangzhou True Shanghai True Suzhou False
Series也有類似的實(shí)例方法
In [23]: obj4.isnull() Out[23]: Beijing False Hangzhou False Shanghai False Suzhou True
Series的一個(gè)重要功能是,在數(shù)據(jù)運(yùn)算中,自動(dòng)對齊不同索引的數(shù)據(jù)
In [24]: obj3 Out[24]: Beijing 40000 Hangzhou 30000 Nanjing 26000 Shanghai 35000 In [25]: obj4 Out[25]: Beijing 40000.0 Hangzhou 30000.0 Shanghai 35000.0 Suzhou NaN In [26]: obj3 + obj4 Out[26]: Beijing 80000.0 Hangzhou 60000.0 Nanjing NaN Shanghai 70000.0 Suzhou NaN
Series的索引可以通過復(fù)制的方式就地修改
In [27]: obj.index = ['Bob', 'Steve', 'Jeff', 'Ryan'] In [28]: obj Out[28]: Bob 4 Steve 7 Jeff -5 Ryan 3
DataFrame
pandas讀取文件
In [29]: df = pd.read_table('pandas_test.txt',sep=' ', names=['name', 'age']) In [30]: df Out[30]: name age 0 Bob 26 1 Loya 22 2 Denny 20 3 Mars 25
DataFrame列選取
df[name]
In [31]: df['name'] Out[31]: 0 Bob 1 Loya 2 Denny 3 Mars Name: name, dtype: object
DataFrame行選取
df.iloc[0,:] #第一個(gè)參數(shù)是第幾行,第二個(gè)參數(shù)是列。這里指第0行全部列 df.iloc[:,0] #全部行,第0列
In [32]: df.iloc[0,:] Out[32]: name Bob age 26 Name: 0, dtype: object In [33]: df.iloc[:,0] Out[33]: 0 Bob 1 Loya 2 Denny 3 Mars Name: name, dtype: object
獲取一個(gè)元素,可以通過iloc,更快的方式是iat
In [34]: df.iloc[1,1] Out[34]: 22 In [35]: df.iat[1,1] Out[35]: 22
DataFrame塊選取
In [36]: df.loc[1:2,['name','age']] Out[36]: name age 1 Loya 22 2 Denny 20
根據(jù)條件過濾行
在方括號中加入判斷條件來過濾行,條件必需返回 True 或者 False
In [37]: df[(df.index >= 1) & (df.index <= 3)] Out[37]: name age city 1 Loya 22 Shanghai 2 Denny 20 Hangzhou 3 Mars 25 Nanjing In [38]: df[df['age'] > 22] Out[38]: name age city 0 Bob 26 Beijing 3 Mars 25 Nanjing
增加列
In [39]: df['city'] = ['Beijing', 'Shanghai', 'Hangzhou', 'Nanjing'] In [40]: df Out[40]: name age city 0 Bob 26 Beijing 1 Loya 22 Shanghai 2 Denny 20 Hangzhou 3 Mars 25 Nanjing
排序
按指定列排序
In [41]: df.sort_values(by='age') Out[41]: name age city 2 Denny 20 Hangzhou 1 Loya 22 Shanghai 3 Mars 25 Nanjing 0 Bob 26 Beijing
# 引入numpy 構(gòu)建 DataFrame import numpy as np
In [42]: df = pd.DataFrame(np.arange(8).reshape((2, 4)), index=['three', 'one'], columns=['d', 'a', 'b', 'c']) In [43]: df Out[43]: d a b c three 0 1 2 3 one 4 5 6 7
# 以索引排序 In [44]: df.sort_index() Out[44]: d a b c one 4 5 6 7 three 0 1 2 3 In [45]: df.sort_index(axis=1) Out[45]: a b c d three 1 2 3 0 one 5 6 7 4 # 降序 In [46]: df.sort_index(axis=1, ascending=False) Out[46]: d c b a three 0 3 2 1 one 4 7 6 5
查看
# 查看表頭5行 df.head(5) # 查看表末5行 df.tail(5) # 查看列的名字 In [47]: df.columns Out[47]: Index(['name', 'age', 'city'], dtype='object') # 查看表格當(dāng)前的值 In [48]: df.values Out[48]: array([['Bob', 26, 'Beijing'], ['Loya', 22, 'Shanghai'], ['Denny', 20, 'Hangzhou'], ['Mars', 25, 'Nanjing']], dtype=object)
轉(zhuǎn)置
df.T Out[49]: 0 1 2 3 name Bob Loya Denny Mars age 26 22 20 25 city Beijing Shanghai Hangzhou Nanjing
使用isin
In [50]: df2 = df.copy() In [51]: df2[df2['city'].isin(['Shanghai','Nanjing'])] Out[52]: name age city 1 Loya 22 Shanghai 3 Mars 25 Nanjing
運(yùn)算操作:
In [53]: df = pd.DataFrame([[1.4, np.nan], [7.1, -4.5], [np.nan, np.nan], [0.75, -1.3]], ...: index=['a', 'b', 'c', 'd'], columns=['one', 'two']) In [54]: df Out[54]: one two a 1.40 NaN b 7.10 -4.5 c NaN NaN d 0.75 -1.3
#按列求和 In [55]: df.sum() Out[55]: one 9.25 two -5.80 # 按行求和 In [56]: df.sum(axis=1) Out[56]: a 1.40 b 2.60 c NaN d -0.55
group
group 指的如下幾步:
Splitting the data into groups based on some criteria
Applying a function to each group independently
Combining the results into a data structure
See the Grouping section
In [57]: df = pd.DataFrame({'A' : ['foo', 'bar', 'foo', 'bar', ....: 'foo', 'bar', 'foo', 'foo'], ....: 'B' : ['one', 'one', 'two', 'three', ....: 'two', 'two', 'one', 'three'], ....: 'C' : np.random.randn(8), ....: 'D' : np.random.randn(8)}) ....: In [58]: df Out[58]: A B C D 0 foo one -1.202872 -0.055224 1 bar one -1.814470 2.395985 2 foo two 1.018601 1.552825 3 bar three -0.595447 0.166599 4 foo two 1.395433 0.047609 5 bar two -0.392670 -0.136473 6 foo one 0.007207 -0.561757 7 foo three 1.928123 -1.623033
group一下,然后應(yīng)用sum函數(shù)
In [59]: df.groupby('A').sum() Out[59]: C D A bar -2.802588 2.42611 foo 3.146492 -0.63958 In [60]: df.groupby(['A','B']).sum() Out[60]: C D A B bar one -1.814470 2.395985 three -0.595447 0.166599 two -0.392670 -0.136473 foo one -1.195665 -0.616981 three 1.928123 -1.623033 two 2.414034 1.600434
【相關(guān)推薦】
1. 特別推薦:“php程序員工具箱”V0.1版本下載
2. Python免費(fèi)視頻教程
3. Python基礎(chǔ)入門教程
聲明:本網(wǎng)頁內(nèi)容旨在傳播知識,若有侵權(quán)等問題請及時(shí)與本網(wǎng)聯(lián)系,我們將在第一時(shí)間刪除處理。TEL:177 7030 7066 E-MAIL:11247931@qq.com