Last active
August 29, 2015 14:03
-
-
Save leonaburime/58a858f551bdf10937be to your computer and use it in GitHub Desktop.
Gradient Descent Algorithm
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from __future__ import division | |
import numpy as np | |
import math, pdb | |
from sklearn import linear_model | |
#http://stackoverflow.com/questions/17784587/gradient-descent-using-python-and-numpy | |
def genData(numPoints, bias, variance): | |
x = np.zeros(shape=(numPoints, 1)) | |
y = np.zeros(shape=numPoints) | |
# basically a straight line | |
for i in range(0, numPoints): | |
# bias feature | |
x[i] = i | |
# our target variable | |
y[i] = (i + bias) + np.random.uniform(0, 1) * variance | |
return x, y | |
X, y = genData(100, 25, 10) | |
def prependOnesColumn(X): | |
# Lets do some preprocessing to add a column of ones for intercept(x^0) values | |
# This series of steps is being done so if you have a nx1 or nxm array it will prepend a column of ones | |
num_of_rows =X.shape[0] # First get the dimensions | |
new_matrix = np.insert( X.flatten('F'), 0, np.ones(num_of_rows)) # Next flatten out the matrix and insert ones at the zeroth index | |
X = new_matrix.reshape( num_of_rows, int(X.size/num_of_rows + 1), order='F') # Our new X with a prepended column of ones | |
return X | |
#A much simpler, faster, and waaay more efficient way of doing linear regression. | |
def normalEquation(X,y): | |
X, y = np.array(X), np.array(y) | |
X = prependOnesColumn(X) | |
#Normal Equation - Solve for x at http://mathworld.wolfram.com/NormalEquation.html | |
A_inv = np.linalg.inv( np.dot( X.T, X) ) | |
B = np.dot( X.T, y) | |
theta = np.dot(A_inv, B) | |
print 'Thetas are ' + str( theta ) | |
class linearRegression: | |
learning_rate = 0.0005 | |
max_iterations = 1000 | |
def __init__(self): | |
pass | |
def hypothesis(self, X, thetas): | |
return np.dot(X, thetas) | |
def getCost(self, X, Y, thetas): | |
m = X.shape[0] | |
cost = (1.0/(2*m))*np.sum( (self.hypothesis(X,thetas)-Y)**2) | |
return cost | |
def solve(self, X=X, Y=y ): | |
X = prependOnesColumn(X) | |
num_of_rows, num_of_columns = X.shape[0], X.shape[1] # Number of rows | |
thetas = np.ones( num_of_columns,)# Create and initialize thetas | |
cost = [0]*num_of_columns | |
max_iterations = 100000 | |
for i in range(max_iterations): | |
error = (self.hypothesis(X,thetas).T - Y.T) | |
j_derivative = np.dot(error, X) | |
thetas -= (self.learning_rate/num_of_rows) * j_derivative | |
print 'Cost ', self.getCost(X,y, thetas) | |
print 'Thetas ' , thetas |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment