def main():
#read input data
my_data = np.genfromtxt('regression_data.txt', dtype=float, delimiter=',')
# create input and output numpy arrays
x = my_data[:,0]
y = my_data[:,1]
# create regression class object
reg = Regression()
# set learning rate
reg.set_learning_rate(0.001)
# set maximum iterations
reg.set_max_iterations(20000)
# set l1 and l2 penalty
reg.set_l1_penalty(0.1)
reg.set_l2_penalty(0.1)
# set tolerance
reg.set_tolerance(1e-5)
# fit a polynomial regression model
theta, cost, it = reg.polynomial_regression(x, y, 6)
print "Regression coefficients :" + str(theta)
print "Minimum cost function: " + str(cost)
print "Iterations taken: " + str(it)
# predict values for new input
z = np.linspace(-2, 2, 4/0.01)
prediction = reg.predict(z)
# plot
fig = plt.figure()
plt.plot(x,y,'.', label='Input data')
plt.plot(z,prediction,'r-', label='Prediction')
plt.legend(loc=4)
fig.suptitle('Polynomial Regression Fit')
plt.xlabel('x (input)')
plt.ylabel('y (predicted)')
plt.savefig('fit_values.eps')
plt.show()
def predict(xlabel,ylabel,x_val,x,y):
meanx=np.mean(x)
stddevx=np.std(x)
meany=np.mean(y)
stddevy=np.std(y)
x=featureNormalize(x)
y=featureNormalize(y)
reg = Regression()
reg.set_learning_rate(0.01)
reg.set_max_iterations(10000)
reg.set_l1_penalty(0.1)
reg.set_l2_penalty(0.1)
reg.set_tolerance(1e-5)
theta, cost, it = reg.polynomial_regression(x, y, 5)
z = np.linspace(-1.9, 2.1, 4/0.01)
prediction = reg.predict(z)
x=np.array(x)*stddevx+meanx
z=np.array(z)*stddevx+meanx
y=np.array(y)*stddevy+meany
prediction=np.array(prediction)*stddevy+meany
x_val=(x_val-meanx)/stddevx
y_val=reg.predict([x_val])
x_val=[x_val]
x_val=np.array(x_val)*stddevx+meanx
y_val=np.array(y_val)*stddevy+meany
fig = plt.figure(figsize=(4,4))
plt.plot(x,y,'.', label='Input data')
plt.plot(z,prediction,'r-', label='Best fit curve')
plt.plot(x_val,y_val,'gx',label='Predicted Data')
plt.legend(loc=4)
title=xlabel + " vs " + ylabel
plt.title(title,size=10)
plt.xticks(size=8)
plt.yticks(size=8)
plt.close('all')
return([y_val,fig])
def predict2(x_val,x,y):
meanx=np.mean(x)
stddevx=np.std(x)
meany=np.mean(y)
stddevy=np.std(y)
x=featureNormalize(x)
y=featureNormalize(y)
reg = Regression()
reg.set_learning_rate(0.001)
reg.set_max_iterations(10000)
reg.set_l1_penalty(0.1)
reg.set_l2_penalty(0.1)
reg.set_tolerance(1e-5)
theta, cost, it = reg.polynomial_regression(x, y, 5)
x_val=(x_val-meanx)/stddevx
y_val=reg.predict([x_val])
y_val=np.array(y_val)*stddevy+meany
return(y_val)
