Python数据分析：实用向

文件处理

导包

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline

添加镜像

https://mirrors.tuna.tsinghua.edu.cn/
https://developer.aliyun.com/mirror/
http://mirrors.163.com/ubuntu/
https://mirrors.ustc.edu.cn/
http://mirrors.zju.edu.cn/
http://mirrors.sohu.com/
http://ftp.sjtu.edu.cn/
http://mirror.bjtu.edu.cn/
http://mirror.bjtu.edu.cn/

语法

其中http和https是可选的

! pip install xxx -i https://mirrors.tuna.tsinghua.edu.cn/

导入文件

excel

data=pd.read_excel(r"C:\Users\ranxi\Desktop\附录1 目标客户体验数据.xlsx", sheet_name='data')
data.head()

csv

data=pd.read_csv()

EDA报告

#生成报告
import pandas_profiling
data.profile_report()
#输出报告文件
pfr = pandas_profiling.ProfileReport(data)
pfr.to_file('report.html')

dataframe导出excel文件

data.to_excel('data.xlsx')

数据处理

数据筛选

分类均值展示

cvr_summary = data.groupby("cvr_group_high")
cvr_summary.mean().reset_index()

标签编码

print("client","--" ,data.client.unique()) 
from sklearn.preprocessing import LabelEncoder
data.client = LabelEncoder().fit_transform(data.client)
print("client","--" ,data.client.unique()) 

交叉比例表

pd.crosstab(data['invited_is'],data["cvr_group_high"],normalize=0)

计算分布比例

def percent_value_counts(df, feature):
    """This function takes in a dataframe and a column and finds the percentage of the value_counts"""
    percent = pd.DataFrame(round(df.loc[:,feature].value_counts(dropna=False, normalize=True)*100,2))
    ## creating a df with th
    total = pd.DataFrame(df.loc[:,feature].value_counts(dropna=False))
    ## concating percent and total dataframe

    total.columns = ["Total"]
    percent.columns = ['Percent']
    return pd.concat([total, percent], axis = 1)

percent_value_counts(data, "B7")

多列apply函数

with_N['B7'] = with_N.apply(lambda x: child_estimator(x['B6'], x['B5']), axis=1)

卡方检验

#分组间确实是有显著性差异，频数比较的结论才有可信度，故需进行”卡方检验“
from scipy.stats import chi2_contingency   #统计分析 卡方检验
#自定义卡方检验函数
def KF(x):
    df1=pd.crosstab(data2['购买意愿'],data2[x])
    li1=list(df1.iloc[0,:])
    li2=list(df1.iloc[1,:])
    kf_data=np.array([li1,li2])
    kf=chi2_contingency(kf_data)
    if kf[1]<0.05:
        print('购买意愿 by {} 的卡方临界值是{:.2f}，小于0.05，表明{}组间有显著性差异,可进行【交叉分析】'.format(x,kf[1],x),'\n')
    else:
        print('购买意愿 by {} 的卡方临界值是{:.2f}，大于0.05，表明{}组间无显著性差异,不可进行交叉分析'.format(x,kf[1],x),'\n')
#对 kf_var进行卡方检验
print('kf_var的卡方检验结果如下:','\n')
print(list(map(KF, kf_var)))

条件筛选

specific=data[(data['a1']>100)|(data['a2']>100)|(data['a3']>100)|(data['a4']>100)|(data['a5']>100)|(data['a6']>100)|(data['a7']>100)|(data['a8']>100)]
specific

specific=data[(data['']>x)|&()]
data[data.Cabin=='N']

map函数分组

def hour_group_fun(hour):
    x = ''
    if 0<=hour<8:
        x=1
    elif 8<=hour<16: 
        x=2
    else:
        x=3
    return x
        
## Applying function to the column.
police['hour_group'] =police['hour'].map(hour_group_fun)

apply多列赋值

with_N['B7'] = with_N.apply(lambda x: child_estimator(x['B6'], x['B5']), axis=1)

这是一个分布比例函数

def percent_value_counts(df, feature):
    """This function takes in a dataframe and a column and finds the percentage of the value_counts"""
    percent = pd.DataFrame(round(df.loc[:,feature].value_counts(dropna=False, normalize=True)*100,2))
    ## creating a df with th
    total = pd.DataFrame(df.loc[:,feature].value_counts(dropna=False))
    ## concating percent and total dataframe

    total.columns = ["Total"]
    percent.columns = ['Percent']
    return pd.concat([total, percent], axis = 1)

特征工程

时间数据处理

police['date'] = pd.to_datetime(police['接警日期'],errors='coerce')

police['year'] =police['date'].dt.year.fillna(0).astype("int")   #转化提取年 
police['month'] = police['date'].dt.month.fillna(0).astype("int")  #转化提取月
police['day'] = police['date'].dt.day.fillna(0).astype("int")      #转化提取天

police['dates'] = police['month'].map(str) + '-' + police['day'].map(str) #转化获取月-日

police['time'] = pd.to_datetime(police['接警时间点'],errors='coerce').dt.time

police['hour'] = pd.to_datetime(police['接警时间点'],errors='coerce').dt.hour.fillna(0).astype("int")    #转化提取小时

SMOTE过抽样

from imblearn.over_sampling import SMOTE
model_smote=SMOTE()
X,y=model_smote.fit_resample(X,y)
X=pd.DataFrame(X,columns=t.columns)
#分拆数据集：训练集 和 测试集
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.3,random_state=0)
print('过抽样数据特征：', X.shape,
      '训练数据特征：',X_train.shape,
      '测试数据特征：',X_test.shape)

print('过抽样后数据标签：', y.shape,
      '   训练数据标签：',y_train.shape,
      '   测试数据标签：',y_test.shape)

输出缺失值

print ("Train age missing value: " + str((train.Age.isnull().sum()/len(train))*100)+str("%"))

影响分析

xgb输出特征重要性

model_xgb= XGBClassifier()
model_xgb.fit(X,y)
from xgboost import plot_importance
plot_importance(model_xgb,height=0.5,color='green',title='')
# plt.savefig('imp.png')
plt.show()

计算相关系数并画图

plt.style.use('classic')
plt.rcParams['font.sans-serif'] = ['SimHei']  # 黑体
plt.rcParams['axes.unicode_minus'] = False    # 解决无法显示符号的问题
plt.rc("figure", facecolor="white")           #去除灰色边框
plt.figure(figsize=(15,6),dpi=300)
df_onehot.corr()['购买意愿'].sort_values(ascending=False).plot(kind='bar',color='dodgerblue')
plt.savefig('buyvary1.png', dpi=300)
plt.show()

data.corr(method='pearson')
data.corr(method='spearman')
data.corr(method='kendall')

Pandas处理

常用操作

为dataframe添加1列

data['age']=list

合并表格再排序

data = pd.concat([with_N, without_N], axis=0)

data.sort_values(by = '目标客户编号', inplace=True)

dataframe排序

useful=useful.sort_values(by = ['购买难度'], ascending = [True])

选取指定行（以列的值筛选）

first1=data3[(data3['品牌编号']==1)]

获取列名

kf=list(data2.columns[1:7])
for x in [9,11,12,20,21,24,25,26]:
    kf.append(data2.columns[x])
print(kf)

修改列名

#1、修改列名a，b为A、B。
df.columns = ['A','B']
#2、只修改列名a为A
df.rename(columns={'a':'A'})

删除一列

data3=data3.drop(1,axis=0)

列表转dataframe(嵌套列表）

from pandas.core.frame import DataFrame
data7=DataFrame(week)
data7

类型转换

Dataframe到Series

Series = Dataframe['column']

Series到list

list = Series.to_list()

list 转 array

array = np.array(list)

array 转 torch.Tensor

tensor = torch.from_numpy(array)

torch.Tensor 转 array

array = tensor.numpy()
# gpu情况下需要如下的操作
array = tensor.cpu().numpy()

torch.Tensor 转 list

# 先转numpy，后转list
list = tensor.numpy().tolist()

array 转 list

list = array.tolist()

list 转 torch.Tensor

tensor=torch.Tensor(list)

array或者list转Series

series = pd.Series({'a': array})
series2 = pd.Series({'a': list})

list转dataframe

data4=DataFrame(li)

array转dataframe

df = pd.DataFrame(data=data[0:,0:],columns='pregnants','Plasma_glucose_concentration','blood_pressure','Triceps_skin_fold_thickness','serum_insulin','BMI','Diabetes_pedigree_function','Age','Target'] )

python需要注意的地方

变量

列表的复制：直接采用a=b的方式会指向同一个内存地址

全局变量：函数内部的变量，外部是无法访问的，在函数内部定义global 后函数运行过才可访问

循环

• continue： 跳出本次循环
• break： 跳出本层循环

运算

矩阵numpy乘法：

• 点乘： np.dot(xy)
• 数乘： np.mat(x,int)

随机数

import random
print( random.randint(1,10) )        # 产生 1 到 10 的一个整数型随机数  
print( random.random() )             # 产生 0 到 1 之间的随机浮点数
print( random.uniform(1.1,5.4) )     # 产生  1.1 到 5.4 之间的随机浮点数，区间可以不是整数
print( random.choice('tomorrow') )   # 从序列中随机选取一个元素
print( random.randrange(1,100,2) )   # 生成从1到100的间隔为2的随机整数
a=[1,3,5,6,7]                # 将序列a中的元素顺序打乱
random.shuffle(a)
print(a)

import random
import string
# 随机整数：
print random.randint(1,50)
# 随机选取0到100间的偶数：
print random.randrange(0, 101, 2)
# 随机浮点数：
print random.random()
print random.uniform(1, 10)
# 随机字符：
print random.choice('abcdefghijklmnopqrstuvwxyz!@#$%^&*()')
# 多个字符中生成指定数量的随机字符：
print random.sample('zyxwvutsrqponmlkjihgfedcba',5)
# 从a-zA-Z0-9生成指定数量的随机字符：
ran_str = ''.join(random.sample(string.ascii_letters + string.digits, 8))
print ran_str
# 多个字符中选取指定数量的字符组成新字符串：
print ''.join(random.sample(['z','y','x','w','v','u','t','s','r','q','p','o','n','m','l','k','j','i','h','g','f','e','d','c','b','a'], 5))
# 随机选取字符串：
print random.choice(['剪刀', '石头', '布'])
# 打乱排序
items = [1, 2, 3, 4, 5, 6, 7, 8, 9, 0]
print random.shuffle(items)

画图

画图准备

解决中文符号显示问题

plt.rcParams['font.sans-serif'] = ['SimHei']  # 黑体
plt.rcParams['axes.unicode_minus'] = False    # 解决无法显示符号的问题

sns.set(font='SimHei', font_scale=0.8)        # 解决Seaborn中文显示问题

设置背景样式

plt.style.use('classic')
plt.rc("figure", facecolor="white")           #去除灰色边框

绘图

这是一个画箱线图代码

import matplotlib.pyplot as plt
sns.set_style('darkgrid')
fig, ax = plt.subplots(figsize=(16,12),ncols=2)
ax1 = sns.boxplot(x="Embarked", y="Fare", hue="Pclass", data=train, ax = ax[0]);
ax2 = sns.boxplot(x="Embarked", y="Fare", hue="Pclass", data=test, ax = ax[1]);
ax1.set_title("Training Set", fontsize = 18)
ax2.set_title('Test Set',  fontsize = 18)
fig.show()

画缺口饼图

churn_value=data['cvr_group_high'].value_counts()
labels=data['cvr_group_high'].value_counts().index
plt.figure(figsize=(7,7))
plt.pie(churn_value,labels=['一般客户', '高价值客户'],colors=["#75bbfd","#00ffff"], explode=(0.05,0),autopct='%1.1f%%', shadow=False)
plt.rcParams['font.sans-serif']=['SimHei']
plt.rcParams['axes.unicode_minus'] = False
plt.title("高价值客户占比23.4%")
#plt.savefig('pie.png', dpi=300) 

画相关性系数图

mask = np.zeros_like(data.corr(), dtype=np.bool)
#mask[np.triu_indices_from(mask)] = True

plt.subplots(figsize = (15,12))
sns.heatmap(data.corr(), 
            annot=True,
#             mask = mask,
            cmap = 'RdBu', ## in order to reverse the bar replace "RdBu" with "RdBu_r"
            linewidths=.9, 
            linecolor='gray',
            fmt='.2g',
            center = 0,
            square=True)
plt.title("Correlations Among Features", y = 1.03,fontsize = 20, pad = 40) #相关性矩阵
plt.savefig('cor.png', dpi=300) 
plt.show()  

画核密度估计

fig = plt.figure(figsize=(15,8),)
## I have included to different ways to code a plot behigh, choose the one that suites you. 
ax=sns.kdeplot(data.client[data.cvr_group_high == 0] , 
               color='gray',
               shade=True,
               label='high')
ax=sns.kdeplot(data.loc[(data['cvr_group_high'] == 1),'client'] , 
               color='g',
               shade=True, 
               label='high', 
              )
plt.title('client - high vs high', fontsize = 25, pad = 40)
plt.ylabel("Frequency of cvr", fontsize = 15, labelpad = 20)
plt.xlabel("Client", fontsize = 15,labelpad =20)
## Converting xticks into words for better understanding
labels = ['H5', 'android', 'ios','pc','wap']
plt.xticks(sorted(data.client.unique()), labels)
plt.legend()

模型训练

导入模块

#加载模块
from sklearn.preprocessing import StandardScaler
import warnings
warnings.filterwarnings("ignore") #过滤掉警告的意思
from pyforest import *
import pandas as pd
import numpy as np

from sklearn.ensemble import RandomForestClassifier     #随机森林
from sklearn.svm import SVC,LinearSVC                   #支持向量机
from sklearn.linear_model import LogisticRegression     #逻辑回归
from sklearn.neighbors import KNeighborsClassifier      #KNN算法
from sklearn.cluster import KMeans                     #K-Means 聚类算法
from sklearn.naive_bayes import GaussianNB              #朴素贝叶斯
from sklearn.tree import DecisionTreeClassifier         #决策树

import xgboost as xgb
from xgboost import XGBClassifier                      
from catboost import CatBoostClassifier                
from sklearn.ensemble import AdaBoostClassifier
from sklearn.ensemble import GradientBoostingClassifier 

from sklearn.metrics import classification_report,precision_score,recall_score,f1_score,accuracy_score #分类报告
from sklearn.metrics import confusion_matrix           #混淆矩阵
from sklearn.metrics import silhouette_score           #轮廓系数（评价k-mean聚类效果）
from sklearn.model_selection import GridSearchCV       #交叉验证
from sklearn.metrics import make_scorer
from sklearn.ensemble import VotingClassifier          #投票

def plot_predictions(test,predicted):
    #整体平移
    x=np.arange(0,len(test))+1
#     x[0]=1
#     my_x_ticks = np.arange(1, 14, 1)
#     plt.xticks(my_x_ticks)
    plt.plot(x,test,label='Real')
    plt.plot(x,predicted,color='darkOrange',linestyle='--',label='Predicted')
#     plt.xlabel('month')
    plt.ylabel('count')
    plt.legend()
import math
def mse_loss(y_true, y_pred):
    return np.sum(np.power(y_true - y_pred, 2)) / y_true.shape[0] / 2
def return_rmse(test,predicted):
    rmse = math.sqrt(mse_loss(test, predicted))
    return rmse
#     print("The mean squared error is {}.".format(rmse))

Classifiers=[
            ["Random Forest",RandomForestClassifier()],
             ["Support Vector Machine",SVC()],
             ["LogisticRegression",LogisticRegression()],
             ["KNN",KNeighborsClassifier(n_neighbors=5)],
             ["Naive Bayes",GaussianNB()],
             ["Decision Tree",DecisionTreeClassifier()],
             ["AdaBoostClassifier",AdaBoostClassifier()],
             ["GradientBoostingClassifier", GradientBoostingClassifier()],
             ["XGB", XGBClassifier()],
]

设置训练集

X=train.drop(['目标客户编号','品牌类型','购买意愿'], axis = 1)
# X=train.drop(['目标客户编号','品牌类型'], axis = 1)
t=X
headers = X.columns
X= X.astype(float)
y = train["购买意愿"]

训练模型

import warnings
warnings.filterwarnings('ignore')
Classify_result=[]
names=[]
prediction=[]
for name,classifier in Classifiers:
    classifier=classifier
    classifier.fit(X_train,y_train)
    y_pred=classifier.predict(X_test)
    recall=recall_score(y_test,y_pred,average='macro')
    precision=precision_score(y_test,y_pred,average='macro')
    f1score = f1_score(y_test, y_pred,average='macro')
    mse = return_rmse(y_test,y_pred)
    class_eva=pd.DataFrame([recall,precision,f1score,mse])
    Classify_result.append(class_eva)
    name=pd.Series(name)
    names.append(name)
    y_pred=pd.Series(y_pred)
    prediction.append(y_pred)
    plot_predictions(y_test,y_pred)
#     # plt.savefig('seven1.png', dpi=300)
    plt.show()

模型评估

names=pd.DataFrame(names)
names=names[0].tolist()
result=pd.concat(Classify_result,axis=1)
result.columns=names
result.index=["recall","precision","f1score","mse"]
result

小工具

tqdm显示进度条

from tqdm import tqdm

for I in tqdm():

记录时间

Import time
time_begin = time.time()
#code,你的程序
time_end = time.time()
time = time_end - time_begin
print('time:', time)

jupyter操作

• Shift+上下键 # 按住Shift进行上下键操作可复选多个cell
• Shift-M # 合并所选cell或合并当前cell和下方的cell
• Ctrl + Shift + - # 从光标所在的位置拆分cell

原创作者：孤飞-博客园
原文链接：