def eigenSpace(matrix):
sampleIndex = np.hstack(
(np.arange(0, 50), np.arange(250, 300), np.arange(700, 750)))
sam_matrix = matrix[sampleIndex, :]
## sam=np.zeros((3,50,30))
## sam[0]=matrix[0:50:1,:]
## sam[1]=matrix[250:300:1,:]
## sam[2]=matrix[700:750:1,:]
eva, eve = functions.pca(sam_matrix)
indsort = np.argsort(-eva)
indeve = np.array([0, 1, 2])
eveSelect = eve[:, indsort[indeve]]
nmat = np.dot(sam_matrix, eveSelect)
#pHist(nmat[50:100:1,2])
pEigenSpace_3d((nmat[0:50:1, :], nmat[50:100:1, :], nmat[100:150:1, :]),
indeve + 1)
pEigenSpace_1((nmat[0:50:1, :], nmat[50:100:1, :], nmat[100:150:1, :]),
indeve + 1)
#we don't care for the appearances of
mapping[id] = 'dc'
all_features[column] = all_features[column].map(mapping)
personal_data_encoding('bin', 3)
personal_data_encoding('card_id', 10)
encoded_feat = pd.get_dummies(all_features)
#after the encoding too many columns are created
#we use pca to reduce the features and identify the most discriminative ones
#we have the best results for 500 components however it takes a lot of time
#thus we have 100 for the script
comp = 100
test_feature = functions.pca(encoded_feat, enc_label, comp)
test_ff = pd.DataFrame(test_feature)
rf = RandomForestClassifier()
tp, fp, tn, fn, true_labels, probs = functions.cross_validation(
rf, test_ff, enc_label, 10, 0.5)
print('Random Forest True Positives %s' % sum(tp))
print('Random Forest False Positives %s' % sum(fp))
fScore, precision, recall = functions.metrics(tp, fp, tn, fn)
print('Random Forest Fscore %s' % fScore)
print('Random Forest Precision %s' % precision)
print('Random Forest Recall %s' % recall)
#ROC Curve
n1, n2, n3, n4, n5])
plt.style.use('dark_background')
fig = plt.figure(figsize=(3,5))
for i in range(len(gray_arr)):
plt.subplot(3, 5, i+1)
plt.title(i)
plt.imshow(gray_arr[i], cmap='gray', vmin=0, vmax=255)
plt.xticks([])
plt.yticks([])
plt.show()
""" Training """
y = []
for i in gray_arr:
y.append(pca(i)[7])
""" Test Images """
# Dragon Queen Test
dqt = grayscale_image('images/Dragon Queen Test.jpg')
ydq = pca(dqt)[7]
# Missandei test
mit = grayscale_image('images/Missandei Test.jpg')
ymi = pca(mit)[7]
# Nichole test
nt = grayscale_image('images/Nichole Test.jpg')
yn = pca(nt)[7]
# Group images
et = np.zeros((19, 1))
etr = np.zeros((19, 1))
acc = np.zeros((19, 1))
e1 = np.zeros((19, 1))
er1 = np.zeros((19, 1))
et1 = np.zeros((19, 1))
etr1 = np.zeros((19, 1))
acc1 = np.zeros((19, 1))
count = 0
for i in range(10, 200, 10):
print(i)
e[count], er[count], et[count], etr[count], acc[count] = fn.pca(
d.trainvector,
d.trainlabels,
d.testvector,
d.testlabels,
nc=i,
clf='Knn')
e1[count], er1[count], et1[count], etr1[count], acc1[count] = fn.pca(
d.trainvector,
d.trainlabels,
d.testvector,
d.testlabels,
nc=i,
clf='NB')
count += 1
print(
etr1.ravel(),