def test_distance_calculation():
# Start afresh again
# Let us now test how our distance behaves
# Get the first sample
N = extract_samples(X, y, 1)
Samp_Size = 10000
rand = [numpy.random.choice(N.shape[0], Samp_Size, replace = True) for i in \
xrange(Samp_Size)]
N = N[rand, :]
scaler = preprocessing.StandardScaler().fit(N)
N_transform = scaler.transform(N)
Ref, Tree = initialize_calculation(T = None, Data = N_transform, gsize = 3, \
par_train = 0)
# Get the other sample
T = extract_samples(X, y, 4)
rand = [numpy.random.choice(N.shape[0], Samp_Size, replace = True) for i in \
xrange(Samp_Size)]
T = T[rand, :]
T_transform = scaler.transform(T)
Test, Tree = initialize_calculation(T = Tree, Data = T_transform, gsize = 3,\
par_train = 1)
Tmp = traditional_MTS(Ref, Test, par=0)
import matplotlib.pyplot as plt
plt.plot(Tmp)
plt.show()
def distance_Iteration():
## Class wise distance evaluation
from Library_Paper_one import traditional_MTS
# # Loop to evaluate distance value for all the classes.
for p in tqdm(xrange(1, C + 1)):
N = np.loadtxt(path_store + 'C' + str(p) + dataset_type + '.csv',
delimiter=',')
scaler = preprocessing.StandardScaler().fit(N)
N_transform = scaler.transform(N)
Ref, Tree = initialize_calculation(T = None, Data = N_transform, gsize = 3,\
par_train = 0)
T = np.loadtxt(path_store + 'C' + str(p) + dataset_type + '.csv',
delimiter=',')
T_transform = scaler.transform(T)
Test, Tree = initialize_calculation(T = Tree, Data = T_transform, gsize = 3,\
par_train = 1)
Data = []
for i in tqdm(xrange(T_iterations)):
rand = np.random.choice(Test.shape[0], Samp_Size, replace=True)
T = Test[rand, :]
Tmp = traditional_MTS(Ref, T, par=0)
Data.append(Tmp.mean())
np.savetxt((path_store + 'R' + str(p) + dataset_type + '.csv'),
np.reshape(np.array(Data), [
-1,
]),
delimiter=',')
def distance_Iteration():
## Class wise distance evaluation
from Library_Paper_one import traditional_MTS
# # Loop to evaluate distance value for all the classes.
for p in tqdm(xrange(1, C + 1)):
N = extract_samples(X, y, p)
np.savetxt(("../" + 'C' + str(p) + dataset_type + '.csv'),
np.array(N),
delimiter=',')
scaler = preprocessing.StandardScaler().fit(N)
N_transform = scaler.transform(N)
Ref, Tree = initialize_calculation( T = None, Data = N_transform, gsize = 2,\
par_train = 0, output_dimension = 4 )
Data = []
np.savetxt(("../" + 'C_dimred' + str(p) + dataset_type + '.csv'),
np.array(Ref),
delimiter=',')
for i in tqdm(xrange(T_iterations)):
T = np.array(subsample(N_transform, ratio=0.06))
T_dim, Tree = Test, Tree = initialize_calculation(T = Tree, Data = T, gsize = 2,\
par_train = 1, output_dimension = 4)
Tmp = traditional_MTS(Ref, T_dim, par=0)
Data.append(Tmp.mean())
np.savetxt(("../" + 'R' + str(p) + dataset_type + '.csv'),
np.reshape(np.array(Data), [
-1,
]),
delimiter=',')
def Classification_Faults_Bearing(N, T):
p_value_0 = []
p_value_1 = []
p_value_2 = []
p_value_3 = []
for ola in xrange(4):
print "ola --", ola
Test = np.array(T[ola].copy())
print "Shape -- ", Test.shape
D_M = np.zeros((Test.shape[0], 4))
print "D_M shape", D_M.shape
D_M[:, 0] = traditional_MTS(N[0], Test, 0).reshape((Test.shape[0], ))
D_M[:, 1] = traditional_MTS(N[1], Test, 0).reshape((Test.shape[0], ))
D_M[:, 2] = traditional_MTS(N[2], Test, 0).reshape((Test.shape[0], ))
D_M[:, 3] = traditional_MTS(N[3], Test, 0).reshape((Test.shape[0], ))
Label = fisher(D_M, N[0].shape[0], N[1].shape[0], N[2].shape[0],
N[3].shape[0])
c = [0, 0, 0, 0]
c[0] = list(Label).count(0)
c[1] = list(Label).count(1)
c[2] = list(Label).count(2)
c[3] = list(Label).count(3)
print "Counted detection", "(", c[0], ",", c[1], ",", c[2], ",", c[
3], ")"
if ola == 0:
p_value_0.append(
(c[1] + c[2] + c[3]) / float(c[0] + c[1] + c[2] + c[3]))
if ola == 1:
p_value_1.append(
(c[0] + c[2] + c[3]) / float(c[0] + c[1] + c[2] + c[3]))
if ola == 2:
p_value_2.append(
(c[1] + c[0] + c[3]) / float(c[0] + c[1] + c[2] + c[3]))
if ola == 3:
p_value_3.append(
(c[1] + c[2] + c[0]) / float(c[0] + c[1] + c[2] + c[3]))
P = []
P.append(sum(p_value_0) / float(len(p_value_0)))
P.append(sum(p_value_1) / float(len(p_value_1)))
P.append(sum(p_value_2) / float(len(p_value_2)))
P.append(sum(p_value_3) / float(len(p_value_3)))
return P
def classification_dimred(N, N1, Test, Y):
D_M = np.zeros((Test.shape[0], 2))
D_M[:, 0] = np.resize(np.array(traditional_MTS(Test, N)),
(Test.shape[0], ))
D_M[:, 1] = np.resize(np.array(traditional_MTS(Test, N1)),
(Test.shape[0], ))
M_1 = N.shape[0]
M_2 = N1.shape[0]
print N.shape
print N1.shape
print Test.shape
# First let us decide on the prior probabilities
pi1 = 1 / float(2)
pi2 = 1 / float(2)
for i in range(0, 1):
Prob_label = []
i = 0
for element in D_M:
P_1 = ((1 + (1 / float(M_1))) * pi1 * exp(-0.5 * element[0])) / (
((1 + (1 / float(M_1))) * pi1 * exp(-0.5 * element[0])) +
((1 + (1 / float(M_2))) * pi2 * exp(-0.5 * element[1])))
P_2 = ((1 + (1 / float(M_2))) * pi2 * exp(-0.5 * element[1])) / (
((1 + (1 / float(M_1))) * pi1 * exp(-0.5 * element[0])) +
((1 + (1 / float(M_2))) * pi2 * exp(-0.5 * element[1])))
lab_max = max(P_1, P_2)
if lab_max == P_1:
Prob_label.append(0)
else:
Prob_label.append(1)
P = np.resize(np.array(Prob_label), (len(Prob_label), ))
#print Prob_label
c = Counter(Prob_label)
pi1 = ((c[0]) / float(c[0] + c[1]))
pi2 = ((c[1]) / float(c[0] + c[1]))
print("The Counter of the labels are", Counter(P))
print("The Counter of the labels are", Counter(Y))
c1 = Counter(Y)
Corr = P[0:c1[0]]
TE = Y[0:c1[0]]
c = Counter(Corr)
print(c[0] / float(c1[0])) * 100
def TypeOneError(Ref, Data):
p_values = []
detect = traditional_MTS(Ref, np.array(Data), 0)
# print "The mean of the MD values are", detect.mean()
p = 4.615
index1 = list([i for i, v in enumerate(detect) if v < p])
p = 6.42
index2 = list([i for i, v in enumerate(detect) if v < p])
p = 9.26
index3 = list([i for i, v in enumerate(detect) if v < p])
p_values.append((len(index1)) / float(len(Data)))
p_values.append((len(index2)) / float(len(Data)))
p_values.append((len(index3)) / float(len(Data)))
return p_values
def TypeTwoError(Ref, Data):
detect = traditional_MTS(Ref, np.array(Data), 0)
p = 4.615
index1 = list([i for i, v in enumerate(detect) if v < p])
p = 6.42
index2 = list([i for i, v in enumerate(detect) if v < p])
p = 9.26
index3 = list([i for i, v in enumerate(detect) if v < p])
p_values = []
pow_values = []
p_values.append((len(index1)) / float(len(Data)))
p_values.append((len(index2)) / float(len(Data)))
p_values.append((len(index3)) / float(len(Data)))
pow_values.append(1 - ((len(index1)) / float(len(Data))))
pow_values.append(1 - ((len(index2)) / float(len(Data))))
pow_values.append(1 - ((len(index3)) / float(len(Data))))
return p_values, pow_values
def test_data_samples(Ref, dat):
t = time.time()
CF = Classification_Distances(Ref, data, 0, 111)
print "The time for GDM-HDR is", (time.time() - t)
print "The length of the values are", len(CF)
count = 0
for element in CF:
if element < 10:
count = count + 1
print count
print "The mean of the distance values are", CF.mean()
t = time.time()
CF = []
CF = traditional_MTS(Ref, data, 0)
print "The time fo MD is", (time.time() - t)
print "The mean of the distance values are", CF.mean()
count = 0
for element in CF:
if element < 10:
count = count + 1
print count
return C
from scipy import stats
# Gather the normal and the test samples from the data
N = extract_samples(train_dataset, train_labels, 1);
T = extract_samples(train_dataset, train_labels, 4);
temp_scalar = preprocessing.StandardScaler(with_mean = True, with_std = True).fit(N)
N = temp_scalar.transform(N)
T = temp_scalar.transform(T)
import time
start = time.time()
# # 1 -- Dimension reduction
# Ref, Tree = initialize_calculation(T = None, Data = Xtr_s, gsize = 2,\
# par_train = 0, output_dimension = 4)
# # Test, Tree = initialize_calculation(T = Tree, Data = Xte_s, gsize = 2,\
# # par_train = 1, output_dimension = 4)
# 2 - Second type of dimension reduction
from distanceHDR import dim_reduction, dim_reduction_test
start = time.time()
Level, Ref = dim_reduction(N, i_dim=N.shape[1], o_dim=5, g_size=2)
# print(stats.describe(Ref))
Test = dim_reduction_test(T, Level, i_dim=T.shape[1], o_dim=5, g_size=2)
print("The time elapsed is", time.time()-start)
print("\nRef", Ref.shape, "Test", Test.shape)
print("\nref-ref", traditional_MTS(Ref, Ref, 0).mean())
print("\nref-test", traditional_MTS(Ref, Test, 0).mean())