vectorized_regression.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

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


def computeError(intercept, slope, x_data, y_data):
    # you originally were using x_train which was defined as a global
    # above. In general global variables are BAD
    # again, pandas and numpy will handle this broadcast
    y_pred = (intercept + (slope * x_data))
    sum_ = ((y_pred - y_data)**2).sum() # can do .sum here since we're using numpy/pandas
    return sum_/(len(x_data))

    
def costFunction(intercept, slope, y_train, x_train, learning_rate):
    
    #go through each value 
    count = len(x_train)

    # this will broadcast across the dataframe
    y_pred = intercept + (slope * x_train)
    sum0 = ((y_pred - y_train) / count).sum()
    sum1 = (((y_pred - y_train) * x_train) / count).sum()
    
    temp0 = intercept - (learning_rate * sum0)
    temp1 = slope - (learning_rate * sum1)
    
    # don't need to wrap these in [] b/c the comma here defines a tuple
    return temp0, temp1

 
def graphResult(value0, value1, x_train, y_train):
    plt.title('Linear Regression')
    plt.xlabel('X Training Data')
    plt.ylabel('Y Training Data')
    plt.scatter(x_train,y_train)
    x = np.linspace(x_train.min(), x_train.max())
    y_pred = value0 + value1 * x
    plt.plot(x, y_pred)
    plt.show();


def main():
    # changed run() to main() b/c it's more conventional 
    
    start = time.time()
    
    # This is the old data that didn't really have a linear correlation
    # need to specify no header otherwise you will use the first like of data as col headers 
    d = pd.read_csv('data.csv', header=None)
    d.columns = ["X", "y"]
    x_train = d['X']
    y_train = d['y']

    
    value0 = 0
    value1 = 0
    learning_rate = 0.00001
    iterations = 1000
    
    print("Starting gradient descent linear regression with theta0: " 
          + str(value0) + " theta1: " + str(value1))
    print("Working...")
    
    # for each number in iterations run the cost function
    # you need to have a loop here
    for i in range(iterations):
        value0, value1 = costFunction(value0, value1, y_train, x_train, learning_rate)
    
    graphResult(value0, value1, x_train, y_train)
    error = computeError(value0,value1,x_train,y_train)
    
    end = time.time()
    
    print("Finsihed with theta0: " + str(value0) + 
          " theta1: " + str(value1) + " iterations: " 
          + str(iterations) + " time elapsed: " 
          + str(round(end-start, 3)) + "s" 
          + " error: " + str(error)) 
    

# this construct ensures the main function is run when this file is run as
# the primary entry point
if __name__ == "__main__":
    main()