def rbfClassifier(v, v0, k, gamma): X = v.X[:,1:] Y = v.Y U, S = k_neighbour(X, k, True) Fi = rbfTransform(X, U, gamma) w = ln.lstsq(Fi, Y)[0] yy = Fi.dot(w) yy[yy>=0]=1 yy[yy<0]=-1 e_in = hw8.classificationError(Y, yy) ''' Generate E_out ''' X = v0.X[:,1:] Y = v0.Y yy = rbfTransform(X, U, gamma).dot(w) yy[yy>=0]=1 yy[yy<0]=-1 e_out = hw8.classificationError(Y, yy) return e_in, e_out
def rbfClassifier(v, v0, k, gamma):
X = v.X[:, 1:]
Y = v.Y
U, S = k_neighbour(X, k, True)
Fi = rbfTransform(X, U, gamma)
w = ln.lstsq(Fi, Y)[0]
yy = Fi.dot(w)
yy[yy >= 0] = 1
yy[yy < 0] = -1
e_in = hw8.classificationError(Y, yy)
''' Generate E_out '''
X = v0.X[:, 1:]
Y = v0.Y
yy = rbfTransform(X, U, gamma).dot(w)
yy[yy >= 0] = 1
yy[yy < 0] = -1
e_out = hw8.classificationError(Y, yy)
return e_in, e_out
def problem14():
runs = 100
notSep = 0
bit = 0
error = 0
func = SimFunction()
k = 9
gamma = 1.5
e_out = 0
e_out_k = 0
w0 = 0.0
for run in range(runs):
points = 100
v = VSet(Dataset(points), func)
X = v.X[:,1:]
Y = v.Y
try:
U, S = k_neighbour(X, k, True)
# clf = KMeans(k)
# clf.fit(X)
# U = clf.cluster_centers_
Fi = rbfTransform(X, U, gamma)
w = ln.lstsq(Fi, Y)[0]
w0 = max(w0, np.max(np.abs(w)))
# print w
clf = SVC(C=1e6, kernel='rbf', gamma=gamma, coef0=1.)
clf.fit(X, Y)
''' Generate E_out '''
v = VSet(Dataset(1000), func)
X = v.X[:,1:]
Y = v.Y
yy = rbfTransform(X, U, gamma).dot(w)
yy[yy>=0]=1
yy[yy<0]=-1
rbf = hw8.classificationError(Y, yy)
e_out += rbf
kernel = hw8.classificationError(Y, clf.predict(X))
e_out_k += kernel
if kernel < rbf: bit += 1
except:
error += 1
print bit, error, e_out, e_out_k, w0
def problem14():
runs = 100
notSep = 0
bit = 0
error = 0
func = SimFunction()
k = 9
gamma = 1.5
e_out = 0
e_out_k = 0
w0 = 0.0
for run in range(runs):
points = 100
v = VSet(Dataset(points), func)
X = v.X[:, 1:]
Y = v.Y
try:
U, S = k_neighbour(X, k, True)
# clf = KMeans(k)
# clf.fit(X)
# U = clf.cluster_centers_
Fi = rbfTransform(X, U, gamma)
w = ln.lstsq(Fi, Y)[0]
w0 = max(w0, np.max(np.abs(w)))
# print w
clf = SVC(C=1e6, kernel='rbf', gamma=gamma, coef0=1.)
clf.fit(X, Y)
''' Generate E_out '''
v = VSet(Dataset(1000), func)
X = v.X[:, 1:]
Y = v.Y
yy = rbfTransform(X, U, gamma).dot(w)
yy[yy >= 0] = 1
yy[yy < 0] = -1
rbf = hw8.classificationError(Y, yy)
e_out += rbf
kernel = hw8.classificationError(Y, clf.predict(X))
e_out_k += kernel
if kernel < rbf: bit += 1
except:
error += 1
print bit, error, e_out, e_out_k, w0
def problem13():
runs = 1000
notSep = 0
func = SimFunction()
for run in range(runs):
points = 100
v = VSet(Dataset(points), func)
# writeV(points)
# v = readV()
X = v.X[:,1:]
Y = v.Y
clf = SVC(kernel='rbf', C=1e6, gamma=1.5, coef0=1., verbose=False)
clf.fit(X, Y)
# hw8.plotGraph(X, Y, clf, func)
e_in = hw8.classificationError(Y, clf.predict(X))
if e_in > 0.0:
print e_in
notSep += 1
print notSep, notSep/float(runs)
def problem13():
runs = 1000
notSep = 0
func = SimFunction()
for run in range(runs):
points = 100
v = VSet(Dataset(points), func)
# writeV(points)
# v = readV()
X = v.X[:, 1:]
Y = v.Y
clf = SVC(kernel='rbf', C=1e6, gamma=1.5, coef0=1., verbose=False)
clf.fit(X, Y)
# hw8.plotGraph(X, Y, clf, func)
e_in = hw8.classificationError(Y, clf.predict(X))
if e_in > 0.0:
print e_in
notSep += 1
print notSep, notSep / float(runs)