def main():
lb, im = mnist.read(dataset="training")
lbt, imt = mnist.read(dataset="testing")
im = im.reshape((len(im),len(im[0])*len(im[0])))
imt = imt.reshape((len(imt),len(imt[0])*len(imt[0])))
im.astype(float)
imt.astype(float)
# validation set
imv = im[:500]
lbv = lb[:500]
imtt = im[500:]
lbtt = lb[500:]
tag = np.zeros(len(lbv),dtype=int)
if READ:
sigma = []
mu = []
for i in range(10):
ts = np.loadtxt('sigma'+str(i)+'.out')
tm = np.loadtxt('mu'+str(i)+'.out')
sigma.append(ts)
mu.append(tm)
sigma = np.array(sigma)
mu = np.array(mu)
sigma_inv = []
for i in range(len(sigma)):
sigma_inv.append(np.linalg.inv(sigma[i]))
sigma_inv = np.array(sigma_inv)
else:
data = zip(lbtt, imtt)
datat = zip(lbv, imv)
# group data into 10 groups
data_grouped = [list(v) for l,v in groupby(sorted(data, key=lambda x:x[0]), lambda x:x[0])]
sigma, mu = cov.cov(data_grouped, 10000)
sigma = np.array(sigma)
mu = np.array(mu)
sigma_inv = []
for i in range(len(sigma)):
sigma_inv.append(np.linalg.inv(sigma[i]))
sigma_inv = np.array(sigma_inv)
#print np.linalg.slogdet(sigma[0])
pi = class_prob(lb)
cnt = 0
for imt_data in imv:
p = -sys.maxint
ind = -1
pj = class_gauss(imt_data, sigma, sigma_inv, mu)
for j in range(len(pj)):
temp = math.log(pi[j])+pj[j]
if (p < temp):
p = temp
ind = j
print lbv[cnt],':',ind
if lbv[cnt] == ind:
tag[cnt] = 1
cnt += 1
print 'hit:',sum(tag)
print 'total:', len(tag)
print 'miss rate:', 1.0-float(sum(tag))/len(tag)
def main():
f = 0.05
th = 6.27344414456
#f = 0.1
#th = 9.25216525759
#f = 0.15
#th = 12.6067380436
#f = 0.2
#th = 14.6211445211
lb, im = mnist.read(dataset="training")
lbt, imt = mnist.read(dataset="testing")
tested = len(lbt)
tag = np.zeros(tested,dtype=int)
im = im.reshape((len(im),len(im[0])*len(im[0])))
imt = imt.reshape((len(imt),len(imt[0])*len(imt[0])))
if READ:
sigma = []
mu = []
for i in range(10):
ts = np.loadtxt('sigma'+str(i)+'.out')
tm = np.loadtxt('mu'+str(i)+'.out')
sigma.append(ts)
mu.append(tm)
sigma = np.array(sigma)
mu = np.array(mu)
sigma_inv = []
for i in range(len(sigma)):
sigma_inv.append(np.linalg.inv(sigma[i]))
sigma_inv = np.array(sigma_inv)
else:
data = zip(lb, im)
datat = zip(lbt, imt)
# group data into 10 groups
data_grouped = [list(v) for l,v in groupby(sorted(data, key=lambda x:x[0]), lambda x:x[0])]
sigma, mu = cov.cov(data_grouped, 10000)
sigma = np.array(sigma)
mu = np.array(mu)
sigma_inv = []
for i in range(len(sigma)):
sigma_inv.append(np.linalg.inv(sigma[i]))
sigma_inv = np.array(sigma_inv)
#print np.linalg.slogdet(sigma[0])
pi = class_prob(lb)
cnt = 0
ab = 0
for imt_data in imt:
p = -sys.maxint
p2 = -sys.maxint
ind = -1
ind2 = -1
pj = class_gauss(imt_data, sigma, sigma_inv, mu)
for j in range(len(pj)):
temp = math.log(pi[j])+pj[j]
if (p < temp):
p = temp
ind = j
elif (p2 < temp):
p2 = temp
ind2 = j
if (p-p2) < th:
ab += 1
print lbt[cnt],':',ind, ':', p, '(abstain!)'
else:
print lbt[cnt],':',ind, ':', p
if lbt[cnt] == ind:
tag[cnt] = 1
cnt += 1
print 'hit:',sum(tag)
print 'total:', len(tag)
print 'abstain:', ab
print 'miss rate:', 1.0-float(sum(tag))/(len(tag)-ab)
# -*- coding: utf-8 -*- """ Created on Mon Apr 2 22:18:07 2018 @author: Administrator """ import a, b, c, d, e, f, cov a.x1() b.x2() c.x3() d.x4() e.x5() f.x6() cov.cov()
