Homework Pandas Statsmodels

Part 1

用Dataframe结构的groupby函数可以将整表依照某index分组为多个，这样对之进行mean()，var()就方便很多，但不方便作corrwith()操作，因为需要self为Dataframe，other为Serial。所以又建立一个列表，专门求相关系数和最小二乘。

import numpy as np
import scipy as sp
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import statsmodels.api as sm
import statsmodels.formula.api as smf

anascombe = pd.read_csv('anscombe.csv')
grouped=anascombe.groupby('dataset')
dgroup = [0]*5
for i in range(0,4):
    dgroup[i]=anascombe.ix[i*11:i*11+10,['x','y']]

print(" Mean:")
print(grouped.mean()) #平均
print("\n Variance:")
print(grouped.var()) #方差
print('\n Correlation coefficient:')
for i in range(0,4):
    print(i,'\t',dgroup[i][['x']].corrwith(dgroup[i]['y']),'\n') #相关系数

for i in range(0,4):   
    X = sm.add_constant(dgroup[i]['x'])
    Y = dgroup[i]['y']
    lin_model = sm.OLS(Y,X)
    result = lin_model.fit()
    print(result.summary())
    #最小二乘

结果表明，从统计数据上观测，这4组样本的特征几乎完全一致。

Part 2

若用lmplot绘制，则各组数据貌似相近；但若通过FacetGrid绘制plt.scatter子图，比对后会发现，四组数据本质差别极大：图一为正常的散点图，图二为抛物线，图三为带一异常值的直线，图四同理。通过这次作业我更加熟悉了python图表的绘制。

import numpy as np
import scipy as sp
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import statsmodels.api as sm
import statsmodels.formula.api as smf
sns.set()

anascombe = pd.read_csv('anscombe.csv')
g=sns.FacetGrid(anascombe, col="dataset",size=4) #以dataset为分割作子图
g.map(plt.scatter,"x","y") #用plt.scatter函数绘制
plt.show()

#g = sns.lmplot(x ='x',y ='y',data = anascombe,hue ='dataset')
#plt.show()