def choose_hyper(sampx, order, sampy, polyx, polyy, alpha, sig2):
sigma_star, y_prime = do_regression(grand_order(sampx, order),
sampy, polyx, polyy, order, alpha,
float(sig2))
best_alpha = alpha
least_mean_error = data_reading.MeanSquareError(y_prime, polyy)
alphas = linspace(0.1, 11, 100)
for alp in alphas:
sigma_star, y_prime = do_regression(grand_order(sampx, order), sampy,
polyx, polyy, order, alp,
float(sig2))
mean_error = data_reading.MeanSquareError(y_prime, polyy)
if mean_error <= least_mean_error:
least_mean_error = mean_error
best_alpha = alp
return best_alpha
def choose_hyper(matFi, sampy, polyx, polyy, order):
least_error = 100
target_alpha = 1
alphas = np.linspace(-3, 2, 1000)
for alp in alphas:
y_prime = grand_order(polyx, order) * do_regression(matFi, sampy, alp)
error = data_reading.MeanSquareError(y_prime, polyy)
if error <= least_error:
target_alpha = alp
least_error = error
return target_alpha
def choose_hyper(phi, y, polyx, polyy, order):
w = do_regression(phi, y, lambd=10)
y_prime = (grand_order(polyx, order) * w)
least_error = 100
target_alpha = 1
alphas = np.linspace(-20, -18, 10)
for alp in alphas:
w = do_regression(phi, y, alp)
y_prime = (grand_order(polyx, order) * w)
error = data_reading.MeanSquareError(y_prime, polyy)
if error <= least_error:
target_alpha = alp
least_error = error
return target_alpha
X, y = grand_order(poly_data['sampx'][0], order), poly_data['sampy']
polyx, polyy = poly_data['polyx'][0], poly_data['polyy']
count_data, count_keys = data_reading.readMatFile("count_data.mat")
target_alpha = choose_hyper(X, y, polyx, polyy, order)
w = do_regression(X, y, target_alpha)
y_prime = grand_order(polyx, order) * w
fig = plt.figure("LASSO")
ax = fig.add_subplot(111)
ax.plot(poly_data['sampx'][0],
y,
color='r',
linestyle='',
marker='*',
label="sample")
# regression line
# ax.plot(polyx,y_prime,color='g',linestyle='-',marker='',label="predict")
ax.plot(polyx,
y_prime,
color='g',
linestyle='-',
label="lambda = " + str(target_alpha))
print "least mean error: " + str(
data_reading.MeanSquareError(y_prime,
polyy)) + " at " + str(target_alpha)
# poly points
ax.plot(polyx, polyy, color='b', linestyle='-', marker='', label="target")
ax.legend()
plt.show()
result = np.zeros((4, 5))
for j in range(1000):
MSEs = []
for percent in percents:
try:
mse = []
train_length = int(percent * total_length)
a = np.random.randint(0, total_length, size=[1,
train_length])[0]
sampx_sub = sampx[a]
sampy_sub = sampy[a]
phi = grand_order(sampx_sub, order)
# LS
LS_y_prime = grand_order(polyx, order) * LSregression(
phi.T, sampy_sub, order)
mse.append(data_reading.MeanSquareError(LS_y_prime, polyy))
# RLS
RLS_target_alpha = RLS_hyper(phi.T, sampy_sub, polyx, polyy,
order)
RLS_y_prime = grand_order(polyx, order) * RLSregression(
phi.T, sampy_sub, RLS_target_alpha)
mse.append(data_reading.MeanSquareError(RLS_y_prime, polyy))
#LASSO
LASSO_target_alpha = LASSO_hyper(phi, sampy_sub, polyx, polyy,
order)
LASSO_y_prime = grand_order(polyx, order) * LASSOregression(
phi, sampy_sub, LASSO_target_alpha)
mse.append(data_reading.MeanSquareError(LASSO_y_prime, polyy))
#RR
RR_y_prime = grand_order(polyx, order) * RRregression(
D = order + 1
n = len(y)
f = np.array(np.append(np.zeros((1,D)),np.ones((1,n))))
In = np.eye(n)
X = np.zeros((n+D,1))
matA = np.vstack((np.hstack((-phiT, -In)), np.hstack((phiT, -In))))
fun = lambda x: np.dot(f.transpose(), x)
cons = ({'type': 'ineq', 'fun': lambda x: b-np.dot(matA, x)})
res = op.minimize(fun, X, constraints=cons, method="COBYLA")
return res.x[0:D]
if '__main__' == __name__:
poly_data , poly_keys = data_reading.readMatFile("poly_data.mat")
order = 5
phiT, y = grand_order(poly_data['sampx'][0],order), poly_data['sampy']
polyx, polyy= poly_data['polyx'][0], poly_data['polyy']
w = do_regression(phiT,y,order)
y_prime = (grand_order(polyx,order) * w)
fig = plt.figure("RR")
ax = fig.add_subplot(111)
ax.plot(poly_data['sampx'][0],y,color='r',linestyle='',marker='*',label="sample")
# # regression line
ax.plot(polyx,y_prime,color='g',linestyle='-',marker='',label="predict")
print "mean error: " + str(data_reading.MeanSquareError(y_prime,polyy))
# poly points
ax.plot(polyx,polyy,color='b',linestyle='-',label="target")
ax.legend()
plt.show()
def do_regression(phi,sampy,order=1):
w = ((np.linalg.inv(phi * phi.T)) * phi) * sampy
# part of painting
return w
if __name__ == "__main__":
poly_data , poly_keys = data_reading.readMatFile("poly_data.mat")
order = 5
sampx = poly_data['sampx'][0]
sampy = poly_data['sampy']
polyx = poly_data['polyx'][0]
polyy = poly_data['polyy']
w = do_regression(grand_order(sampx,order).T,sampy,order)
targety = grand_order(polyx,order) * w
fig = plt.figure("LS")
ax = fig.add_subplot(111)
# sample points
ax.plot(poly_data['sampx'][0],sampy,color='r',linestyle='',marker='*',label="sample")
# regression line
ax.plot(polyx,targety,color='g',linestyle='-',marker='',label="predict")
# poly points
ax.plot(polyx,polyy,color='b',linestyle='-',marker='',label="target")
print data_reading.MeanSquareError(targety,polyy)
ax.legend()
plt.show()