মেশিন লারনিং ঃ Implement: Multivariate Regression: Python
শত ভাগ সঠিক নাউ হতে পারে।
Theory reference: https://www.cmpe.boun.edu.tr/~ethem/i2ml/slides/v1-1/i2ml-chap5-v1-1.pdf . This is an approximate solution to start with. Understand the theory and then adjust/fix/improve
import numpy as np
import random
print(‘Iterations: rows: Please enter the number of samples for each variable/dimension’)
n_number_of_samples_rows = int(input())
print(‘Columns: Dimensions: Please enter number of variables’)
m_number_of_variables_cols = int(input())
# initialize the input data variables
print(m_number_of_variables_cols, n_number_of_samples_rows)
X = np.zeros((n_number_of_samples_rows, m_number_of_variables_cols))
#Y_actually_R = np.zeros((n_number_of_samples_rows, m_number_of_variables_cols))
Y_actually_R = np.zeros((n_number_of_samples_rows * 1)) #m_number_of_variables_cols
#generate random data
for n in range (n_number_of_samples_rows):
for m in range(m_number_of_variables_cols):
X[n,m] = random.random()
Y_actually_R[n] = random.random()
print(“X”)
print(X)
print(“Y in row format”)
print(Y_actually_R)
print(“Y in column/vector format”)
Y_actually_R = np.transpose(Y_actually_R) #transpose
print(Y_actually_R)
#convert to matrix
X = np.matrix(X)
Y_actually_R = np.matrix(Y_actually_R)
print(‘————-matrix-print—X and then Y’)
print(X)
print(‘Y matrix’)
print(Y_actually_R)
#THE EQUATION: steps to calculate W matrix: w = (((X.Transpose) * X).invert) * X.Transpose * r
#transpose X
X_transpose = np.transpose(X)
print(‘X_transpose’)
print(X_transpose)
#(X.Transpose) * X) of [w = (((X.Transpose) * X).invert) * X.Transpose * r]
w_parameters = np.dot(X_transpose, X) #does np.multiply work? probably need shapoing/reshaping
print(‘first dot’)
print(w_parameters)
#(X.Transpose) * X).invert
w_parameters = np.linalg.inv(w_parameters)
print(‘inverted’)
print(w_parameters)
#(((X.Transpose) * X).invert) * X.Transpose
w_parameters = np.dot(w_parameters, X_transpose)
print(‘2nd dot’)
print(w_parameters)
#(((X.Transpose) * X).invert) * X.Transpose * r
#Y_actually_R = np.transpose(Y_actually_R)
w_parameters = np.dot(w_parameters, np.transpose(Y_actually_R)) #np.dot( np.transpose(w_parameters), np.transpose(Y_actually_R))
#two times transpose – redundant. actually, we could avoid both transpose of Y upto this point
print(‘w_matrix’)
print(w_parameters)
w_matrix = w_parameters
#sum of ( rt – w0 – w1x1 – w2x2 ….. wd * xd ) rt = Y_Actually_R[t 1….N][variable_1….d]
#d eqv to m_number_of_variables_cols — i used m for that
#E(w 0 ,w 1 ,…,w d |X )
#Should it be a matrix or just one total sum? I assume one total sum
error_matrix = np.zeros(m_number_of_variables_cols)
error_sum = 0
#calculate error
Y_actually_R = np.transpose(Y_actually_R) #np.array(Y_actually_R)
for n in range(n_number_of_samples_rows):
sum = 0
for m in range(m_number_of_variables_cols):
sum = Y_actually_R[m]
#2nd part ie w1x1 w2x2 wdxd of the equation: sum of ( rt – w0 – w1x1 – w2x2 ….. wd * xd )
wpart = w_matrix[0]
for ii in range(1,m_number_of_variables_cols): # d = m_number_of_variables_cols, sum of w1x1, w2x2 to wdxd
wpart += w_matrix[ii] * X[n,m] #+ w_matrix[2][m] * X[n][m]
#sum = sum – wpart
sum = pow( (sum – wpart), 2)
error_matrix[m] = 0.5 * pow ( (sum – wpart), 2)
error_sum += sum #error_matrix[m] #pow ( (sum – wpart), 2)
error_sum = 0.5 * error_sum
print(‘error matrix if supposed to be = number of variables’)
print(error_matrix)
print(‘Error if supposed to be one number i.e. sum of all errors’)
print(error_sum)
