Python 探索性數據分析(Exploratory Data Analysis,EDA)

探索性數據分析，主要針對原始數據進行初次了解。了解數據的分布情況、了解分析方向、排除該單個變量的異常值 等。此腳本讀取的是 SQL Server ，只需給定表名或視圖名稱，如果有數據，將輸出每個字段符合要求的每張數據分布圖。

# -*- coding: UTF-8 -*-
# python 3.5.0
# 探索性數據分析(Exploratory Data Analysis,EDA)
__author__ = 'HZC'

import math
import sqlalchemy
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

class EDA:
    def __init__(self,d):
        self.engine = sqlalchemy.create_engine("mssql+pymssql://%s:%s@%s/%s" %(d['user'],d['pwd'],d['ins'],d['db']))
        
    def get_df_from_table(self,table_name):
        df = pd.read_sql_table(table_name, self.engine)
        return df
    
    def get_df_from_query(self,sql):
        df = pd.read_sql_query(sql, self.engine)
        return df
    
    #讀取表各字段數據類型
    def get_table_type(self,table_name):
        sql = """select c.name as colname,t.name as typename
        from sys.sysobjects o inner join syscolumns c on o.id=c.id and o.name<>'dtproperties' 
        inner join sys.systypes t on c.xusertype=t.xusertype 
        where o.name='%s'""" % table_name
        df = self.get_df_from_query(sql)
        return df
    
    #繪圖
    def eda_plot(self,table_name):
        list_char = ['char','nchar','varchar','nvarchar','text','ntext','sysname']
        list_num = ['tinyint','smallint','int','real','money','float','decimal','numeric','smallmoney','bigint']
        df_type = self.get_table_type(table_name)
        df_date = self.get_df_from_table(table_name)
        date_count = df_date.shape[0]
        k = 0
        for row in df_type.itertuples():
            k = k + 1
            #字符類型，繪柱狀圖
            if row.typename in list_char:
                col = df_date.groupby([row.colname]).agg({row.colname:['count']})
                row_count = col.shape[0]
                #col_count = col.shape[1]
                col = col.sort_index()
                val = col.values.tolist()
                #只繪不重復數占總數比小於 5% 的
                if math.floor(row_count*100/date_count) <5:
                    df_ = pd.DataFrame(col.index.values.tolist(), columns=[row.colname])
                    df_['count'] = list(i[0] for i in val)
                    x_axle = range(len(df_[row.colname]))
                    y_axle = df_['count'].tolist()
                    x_label = df_[row.colname].tolist()
                    fig, (ax1, ax2) = plt.subplots(2)
                    ax1.bar(x_axle,y_axle)
                    ax1.set_xticks(x_axle)
                    ax1.set_xticklabels(x_label)
                    ax1.set_title('表[%s]  %s  分布' % (table_name,row.colname))
                    ax2.pie(y_axle,labels=x_label, autopct='%1.2f%%')
                    
            #數值類型，其他分布圖    
            elif row.typename in list_num:
                df__ = pd.DataFrame(df_date[row.colname])
                df__ = df__[(df__[row.colname].notnull())].sort_values(row.colname, ascending=True).reset_index(drop=True)
                k = k + 1
                plt.figure(k)
                plt.subplot(1,3,1)
                plt.hist(df__[row.colname])
                plt.subplot(1,3,2)
                plt.boxplot(df__[row.colname])
                plt.gca().set_title('表[%s]  %s  分布' % (table_name,row.colname))
                plt.subplot(1,3,3)
                plt.violinplot(df__[row.colname])
                plt.tight_layout()
            else:
                pass
        plt.show()

if __name__ == "__main__":
    #conn = {'user':'kk','pwd':'kk','ins':'HYH0109-189\CAT2014','db':'CSMS3'} 
    conn = {'user':'用戶名','pwd':'密碼','ins':'實例','db':'數據庫'} 
    eda = EDA(conn)
    eda.eda_plot("表或視圖名")

顯示圖分為字符型（離散型）和數值型（連續型），示例結果如下：