linear_regression_multi_var.py

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

sale=pd.read_csv('sale.csv')
X=np.array(sale.iloc[:,1:4])
Y=np.array(sale.iloc[:,-1])
a=X.max()
b=X.min()
X=(X-b)/(a-b)
ay=Y.max()
by=Y.min()
Y=(Y-by)/(ay-by)

n=len(Y)
def y(w1,w2,w3,x):
    return w1*x[0]+w2*x[1]+w3*x[2]
def MSE(w1,w2,w3):
    ans=0
    for i in range (n):
        ans+=(Y[i]-(w1*X[i][0]+w2*X[i][1]+w3*X[i][2]))**2
    return ans/2/n

def fw1(w1,w2,w3):
    ans=0
    for i in range (n):
        ans+=X[i][0]*(Y[i]-y(w1,w2,w3,X[i]))
    return 2*ans/(-n)

def fw2(w1,w2,w3):
    ans=0
    for i in range (n):
        ans+=X[i][1]*(Y[i]-y(w1,w2,w3,X[i]))
    return 2*ans/(-n)
def fw3(w1,w2,w3):
    ans=0
    for i in range (n):
        ans+=X[i][2]*(Y[i]-y(w1,w2,w3,X[i]))
    return 2*ans/(-n)

LR=0.001
epochs=3005
w1=10
w2=10
w3=10
mse=[]
for i in range(epochs):
    #print(w1)
    #print(w2)
    #print(w3)
    #print(MSE(w1,w2,w3))
    #print('===============')
    t1,t2,t3=w1,w2,w3
    w1-=LR*fw1(t1,t2,t3)
    w2-=LR*fw2(t1,t2,t3)
    w3-=LR*fw3(t1,t2,t3)
    mse.append(MSE(w1,w2,w3))

plt.plot(range(epochs),mse,c='green')
plt.title('Error rate')
plt.ylabel('MSE')
plt.xlabel('Iterations')

plt.show()