X = df.values
y = bdata.target
# need to scale the features (use zero mean scaling)
X_norm,mu,sigma = utils.feature_normalize(X)
# add intercept term to X_norm
XX = np.vstack([np.ones((X.shape[0],)),X_norm.T]).T
print 'Running gradient descent ..'
# set up model and train
linear_reg3 = LinearReg_SquaredLoss()
J_history3 = linear_reg3.train(XX,y,learning_rate=0.01,num_iters=5000,verbose=False)
# Plot the convergence graph and save it in fig5.pdf
plot_utils.plot_data(range(len(J_history3)),J_history3,'Number of iterations','Cost J')
plt.savefig('fig5.pdf')
# Display the computed theta
print 'Theta computed by gradient descent: ', linear_reg3.theta
########################################################################
# ======= Part 3: Predict on unseen data with model ======= ===========#
########################################################################
y = bdata.target
# need to scale the features (use zero mean scaling)
X_norm,mu,sigma = utils.feature_normalize(X)
# add intercept term to X_norm
XX = np.vstack([np.ones((X.shape[0],)),X_norm.T]).T
print X_norm
print 'Running gradient descent ..'
# set up model and train
linear_reg3 = LinearReg_SquaredLoss()
J_history3 = linear_reg3.train(XX,y,learning_rate=0.01,num_iters=5000,verbose=False)
# Plot the convergence graph and save it in fig5.pdf
plot_utils.plot_data(range(len(J_history3)),J_history3,'Number of iterations','Cost J')
plt.savefig('fig5a.pdf')
# Display the computed theta
print 'Theta computed by gradient descent: ', linear_reg3.theta
########################################################################
# ======= Part 3: Predict on unseen data with model ======= ===========#