More than 5 years have passed since last update.

python プログラムいろいろ

Python

Last updated at 2019-11-07Posted at 2018-06-20

書いたものをいろいろ載せてみました。

重回帰：



import pandas as pd
import numpy as np
import matplotlib.pyplot as plt 
from matplotlib import pyplot
from mpl_toolkits.mplot3d.axes3d import Axes3D
from scipy import genfromtxt


df_sample=\
pd.DataFrame([
    [167,167.5,168.4,172,155.3,151.4,163,174,168,160.4,164.7,171,162.6,164.8,163.3,167.6,169.2,168,167.4,172],
    [84,87,86,85,82,87,92,94,88,84.9,78,90,88,87,82,84,86,83,85.2,82],
    [61,55.5,57,57,50,50,66.5,65,60.5,49.5,49.5,61,59.5,58.4,53.5,54,60,58.8,54,56]
]).T

df_sample.columns=["y","x1","x2"]
df_sample.index=[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]

Expect_and_sd_data=\
pd.DataFrame([
    [np.mean(df_sample["y"]),np.std(df_sample["y"])],
    [np.mean(df_sample["x1"]),np.std(df_sample["x1"])],
    [np.mean(df_sample["x2"]),np.std(df_sample["x2"])]
]).T

Expect_and_sd_data.columns=["y","x1","x2"]
Expect_and_sd_data.index=["mean","std"]

data=np.array([df_sample["y"],df_sample["x1"],df_sample["x2"]])
cor=np.zeros([3,3])

for j in range(0,3,1):
    for i in range(0,3,1):
        cor[i][j]=sum((data[i]-np.mean(data[i]))*(data[j]-np.mean(data[j])))
        
print(cor)

COR=cor

cor=np.zeros([3,3])
cor[0:2,0:2]=COR[1:3,1:3]
cor[2][2]=COR[0][0]
for j in range(0,2,1):
    cor[j][2]=COR[0][j+1]
    cor[j+1][2]=COR[j+1][0]
    
A=cor

inv_A=np.linalg.inv(cor[0:2,0:2])

b1=sum(A[2,0:2]*inv_A[0])
b2=sum(A[2,0:2]*inv_A[1])
b0=np.mean(df_sample["y"])-b1*np.mean(df_sample["x1"])-b2*np.mean(df_sample["x2"])
print("linear_equation is",b1,"*x1+",b2,"*x2+",b0)

ayy=A[2][2]

EV=sum((df_sample["y"]-(b0+b1*df_sample["x1"]+b2*df_sample["x2"]))**2)
RV=sum

多項式回帰：


import numpy as np

import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd

df_sample=\
pd.DataFrame([
    [167,167.5,168.4,172,155.3,151.4,163,174,168,160.4,164.7,171,162.6,164.8,163.3,167.6,169.2,168,167.4,172],
    #[84,87,86,85,82,87,92,94,88,84.9,78,90,88,87,82,84,86,83,85.2,82],
    [61,55.5,57,57,50,50,66.5,65,60.5,49.5,49.5,61,59.5,58.4,53.5,54,60,58.8,54,56]
]).T

df_sample.columns=["y","x1"]
df_sample.index=[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]

print(df_sample)

value=np.array([df_sample["x1"]])
ans=np.ones(len(df_sample.index))
vec=ans

DIM=2

dim=min(DIM,len(df_sample.index))
for j in range(0,dim):
    ans=ans*value
    vec=np.vstack((vec,ans))
    
X=vec
Y=np.array([df_sample["y"]])
#print(Y)

X=np.matrix(X)
Y=np.transpose(Y)

a=((np.linalg.inv(X.dot(np.transpose(X)))).dot(X)).dot(Y)

print(a)
print(X)
y=np.transpose(X).dot(a)
error=sum(np.power(abs(y-Y),2))/len(y)
print("平均二乗誤差は",error,"で次元は",DIM,"です。")
R2=1-(sum(np.power(y-Y,2))/sum(np.power(Y-np.mean(Y),2)))
R=(np.sign(R2))*(np.power(R2,1/2))
print("coefficent of determination R is",R)

リッジ回帰：


import pandas as pd
import numpy as np
import matplotlib.pyplot as plt 
from matplotlib import pyplot

df_sample=\
pd.DataFrame([
    [167,167.5,168.4,172,155.3,151.4,163,174,168,160.4,164.7,171,162.6,164.8,163.3,167.6,169.2,168,167.4,172],
    [84,87,86,85,82,87,92,94,88,84.9,78,90,88,87,82,84,86,83,85.2,82],
    [61,55.5,57,57,50,50,66.5,65,60.5,49.5,49.5,61,59.5,58.4,53.5,54,60,58.8,54,56]
]).T

df_sample.columns=["y","x1","x2"]
df_sample.index=[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]

Expect_and_sd_data=\
pd.DataFrame([
    [np.mean(df_sample["y"]),np.std(df_sample["y"])],
    [np.mean(df_sample["x1"]),np.std(df_sample["x1"])],
    [np.mean(df_sample["x2"]),np.std(df_sample["x2"])]
]).T

Expect_and_sd_data.columns=["y","x1","x2"]
Expect_and_sd_data.index=["mean","std"]


Z=np.array([df_sample["x1"]-np.mean(df_sample["x1"]),df_sample["x2"]-np.mean(df_sample["x2"])])
Y=np.array([df_sample["y"]])
            
gamma=5
            
unit=np.identity(2)
            
Beta=(np.linalg.inv(Z.dot(Z.T)+gamma*unit).dot(Z)).dot(Y.T)
fmt="""
係数は{0},{1}で
方程式としては、Y={2}*z1+{3}*z2+{4}
となります。

"""
desc=fmt.format(Beta[0],Beta[1],Beta[0],Beta[1],np.mean(df_sample["y"]))
print(desc)
            
y=(np.mean(df_sample["y"])+(Z.T).dot(Beta))
            
print(y)
            
x1=range(1,len(df_sample["y"])+1)
y1=y
x2=range(1,len(df_sample["y"])+1)
y2=Y

pyplot.scatter(x1,y1,c="b",label="predict values")
pyplot.scatter(x2,y2,c="r",label="data values")
pyplot.legend()
pyplot.title("Ridge regression")
pyplot.show()


import pandas as pd
import numpy as np
import matplotlib.pyplot as plt 
from matplotlib import pyplot
from mpl_toolkits.mplot3d.axes3d import Axes3D
from scipy import genfromtxt
from statistics import mean, median,variance,stdev

#ADALINE

df_sample=\
pd.DataFrame([
    [167,167.5,168.4,172,155.3,151.4,163,174,168,160.4,164.7,171,162.6,164.8,163.3,167.6,169.2,168,167.4,172],
    [84,87,86,85,82,87,92,94,88,84.9,78,90,88,87,82,84,86,83,85.2,82],
    [61,55.5,57,57,50,50,66.5,65,60.5,49.5,49.5,61,59.5,58.4,53.5,54,60,58.8,54,56]
]).T

df_sample.columns=["y","x1","x2"]
df_sample.index=[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]

#print(df_sample)

eta=0.0001;ite=100000

X=np.matrix(df_sample.loc[:, ['x1','x2']])

Y=np.array(df_sample["y"])

X=(X-np.array([np.mean(df_sample["x1"]),np.mean(df_sample["x2"])]))/np.array([stdev(df_sample["x1"]),stdev(df_sample["x2"])])

X=np.c_[np.ones(X.shape[0]),X]

w=np.ones(X.shape[1])

for i in range(0,ite): 
    
    for j in range(0,X.shape[1]):
    
      pthi=np.exp(np.array(X.dot(w)))
    
      dw=eta*sum((Y-pthi)*X[:,j])
    
      w[j]=w[j]+dw
 
    pthi=np.exp(np.array(X.dot(w)))
    
    print(w)        

y=pthi    
    
x1=range(1,len(df_sample["y"])+1)
y1=y
x2=range(1,len(df_sample["y"])+1)
y2=Y

print("cor:",np.corrcoef(y1,y2))

pyplot.scatter(x1,y1,c="b",label="predict values")
pyplot.scatter(x2,y2,c="r",label="data values")
pyplot.legend()
pyplot.title("ADALINE")
pyplot.show()


import pandas as pd
import numpy as np
import matplotlib.pyplot as plt 
from matplotlib import pyplot
from mpl_toolkits.mplot3d.axes3d import Axes3D
from scipy import genfromtxt
from statistics import mean, median,variance,stdev
from copy import copy

# Logistic (regularized parameter)

df_sample=\
pd.DataFrame([
    [167,167.5,168.4,172,155.3,151.4,163,174,168,160.4,164.7,171,162.6,164.8,163.3,167.6,169.2,168,167.4,172],
    [84,87,86,85,82,87,92,94,88,84.9,78,90,88,87,82,84,86,83,85.2,82],
    [61,55.5,57,57,50,50,66.5,65,60.5,49.5,49.5,61,59.5,58.4,53.5,54,60,58.8,54,56]
]).T

df_sample.columns=["y","x1","x2"]
df_sample.index=[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]

#print(df_sample)

eta=0.00001;ite=100000;lam=1

X=np.matrix(df_sample.loc[:, ['x1','x2']])/10

Y=np.array(df_sample["y"])/1000

#X=(X-np.array([np.mean(df_sample["x1"]),np.mean(df_sample["x2"])]))/np.array([stdev(df_sample["x1"]),stdev(df_sample["x2"])])

X=np.c_[np.ones(X.shape[0]),X]

w=np.ones(X.shape[1])
 
def J_cost(s,lam):
    
    z=np.array([1/(1+np.exp(-X.dot(s)))])
    
    J=np.sum(-Y*np.array([np.log(z)])-(np.array([np.ones(len(Y))])-Y)*np.log(np.array([np.ones(len(Y))])-z))+lam*np.sum(s**2)
   
    return J


def logit(s):
    
    return np.array([1/(1+np.exp(-X.dot(s)))])

#print(J_cost(w,0.1))

#print(logit(w))



for i in range(0,ite):
      
    for j in range(0,len(w)):
        
        w_sub=copy(w);
        
        w_sub[j]=w_sub[j]+0.01;
        
        dw=np.sum(J_cost(w_sub,lam)-J_cost(w,lam))
        
        w[j]=w[j]-eta*dw/0.01
        
    print(J_cost(w,lam)) 
    
x1=range(1,len(df_sample["y"])+1)
y1=logit(w)
x2=range(1,len(df_sample["y"])+1)
y2=Y

#print("cor:",np.corrcoef(y1,y2))

pyplot.scatter(x1,y1,c="b",label="predict values")
pyplot.scatter(x2,y2,c="r",label="data values")
pyplot.legend()
pyplot.title("Penalized logistic reg")
pyplot.show()

You get articles that match your needs
You can efficiently read back useful information
You can use dark theme

What you can do with signing up