def KNNclassifier(training, test, tLabels, k, d, *args):
'''Implements the k-nearest neighbors classifier, using a training data set
and the test data set to be labeled. Receives k by argument, as well as the
distance function to be used. Any other arguments that might be needed by
the distance function are stored in *args'''
# Saving dimensions
q = len(training)
n = len(training[0]) - 1
j = len(test)
trainingRDD = sc.parallelize(training)
labels = []
for i in test:
dist = trainingRDD.map(lambda x: euclidean_distance(x, i)).collect()
k_labels = []
# Getting labels of k-nearest neighbors
for i in range(k):
nNeighbor = min(dist)
nnIndex = dist.index(nNeighbor)
k_labels.append(tLabels[nnIndex])
dist.remove(nNeighbor)
labels.append(stat.mode(k_labels))
return labels
def bisect(lines):
distance = []
for index, l in enumerate(lines):
if index != 0:
distance.append(l[0] - lines[index - 1][0])
max_distance = max(distance)
index_distance = distance.index(max_distance)
av_distance = max_distance // 2
lines.append([lines[index_distance][0] + av_distance, lines[index_distance][0] + av_distance])
return lines
def threesect(lines):
distance = []
for index, l in enumerate(lines):
if index != 0:
distance.append(l[0] - lines[index - 1][0])
max_distance = max(distance)
index_distance = distance.index(max_distance)
#45 - this in length three character + epsilon
if max_distance >= 45:
av_distance = max_distance // 2
lines.append([lines[index_distance][0] + av_distance, lines[index_distance][0] + av_distance])
lines.append([lines[index_distance][0] + 2 * av_distance, lines[index_distance][0] + 2 * av_distance])
else:
lines = bisect(lines)
lines = bisect(lines)
return lines
def filtering(self, image):
# 各関数の呼び出し
self.modeling()
intensity, intensity_all = self.calcLikelihood(image)
self.x_means(intensity, intensity_all)
self.normalize()
# リストの用意
self.X = [''] * self.cluster
self.Y = [''] * self.cluster
self.W = [''] * self.cluster
self.H = [''] * self.cluster
self.bx = np.zeros((self.cluster, 1))
self.by = np.zeros((self.cluster, 1))
self.bw = np.zeros((self.cluster, 1))
self.bh = np.zeros((self.cluster, 1))
self.px = [''] * self.cluster
# リサンプリング
self.resampling()
for i in range(self.cluster):
self.X[i] = self.x_kmeans[i][0][:, 0][self.sample[i]]
self.Y[i] = self.x_kmeans[i][0][:, 1][self.sample[i]]
self.W[i] = self.x_kmeans[i][0][:, 2][self.sample[i]]
self.H[i] = self.x_kmeans[i][0][:, 3][self.sample[i]]
# 対象推定
for i in range(self.cluster):
for j in range(len(self.X[i])):
self.bx[i][0] += float(self.X[i][j]) * float(
self.weights[i][j])
for j in range(len(self.Y[i])):
self.by[i][0] += float(self.Y[i][j]) * float(
self.weights[i][j])
for j in range(len(self.W[i])):
self.bw[i][0] += float(self.W[i][j]) * float(
self.weights[i][j])
for j in range(len(self.H[i])):
self.bh[i][0] += float(self.H[i][j]) * float(
self.weights[i][j])
self.px[i] = [
self.bx[i][0], self.by[i][0], self.bw[i][0], self.bh[i][0]
] # 各クラスタのバウンディングボックスのx,y,w,h
# ハンガリアン法
if self.frame_count > 1 and self.cluster > 1:
id = self.hungarian() # ハンガリアン関数の呼び出し
self.pre_px = np.array([x[:] for x in self.px
]) # コピー self.px → self.pre_px
self.pre_id = id[:] # コピー id → self.pre_id
self.flag_count = 0
# クラスタが一つの場合
elif self.frame_count > 1 and self.cluster == 1:
self.flag_count += 1
p1 = np.array(self.px)
p2 = self.pre_px
print "p1:{}".format(len(p1))
print "p2:{}".format(len(p2))
distance = []
# ユークリッド距離を比較し、一番ユークリッド距離が小さいidとする
for px in p2:
distance.append(np.linalg.norm(p1[0] - px))
id = [self.pre_id[distance.index(min(distance))]]
# flag_countが5以上になったら、追跡対象が一匹になったと判断し、idを保存する
if self.flag_count > 5:
self.pre_px = np.array([x[:] for x in self.px])
self.pre_id = id[:]
# 1フレーム目
else:
id = range(self.cluster)
# フレームアウト
average, id = self.frameout(id)
return average, id
wav_file = '/number/eight.wav'
print(str(8) + ' making spectrogram arr')
eight = graph_spectrogram(wav_file)
wav_file = '/number/nine.wav'
print(str(9) + ' making spectrogram arr')
nine = graph_spectrogram(wav_file)
userInput = input()
print('user input making spectrogram arr')
example = graph_spectrogram(userInput)
one = find_similarity(example, one)
two = find_similarity(example, two)
three = find_similarity(example, three)
four = find_similarity(example, four)
five = find_similarity(example, five)
six = find_similarity(example, six)
seven = find_similarity(example, seven)
eight = find_similarity(example, eight)
nine = find_similarity(example, nine)
zero = find_similarity(example, zero)
print('\n minimal gap distance arr')
distance = [zero, one, two, three, four, five, six, seven, eight, nine]
print(distance)
print(distance.index(min(distance)))
# train_data = train_data[0:10,:]
labels = np.zeros(train_data.shape[0])
# print labels.shape
for cluster in number_of_clusters:
directory_path = './' + str(cluster) + '_centroids_images/'
centroid_index = random.sample(range(1, train_data.shape[0]), cluster)
centroid = train_data[centroid_index]
# print centroid[0].shape
for iterations in range(0, iteration):
print "Iteration " + str(iterations)
for idx, data in enumerate(train_data):
dist = []
for center in centroid:
dist.append(np.linalg.norm(data - center))
labels[idx] = dist.index(min(dist))
print labels
#Update Clusterss
for cluster_number in range(0, cluster):
index = []
for idx, label in enumerate(labels):
if label == cluster_number:
index.append(idx)
temp_data = train_data[index]
print 'Updating cluster ' + str(cluster_number)
centroid[cluster_number] = np.mean(temp_data, axis=0)
print centroid
np.savez(str(cluster) + '_centroids.npz', centroids=centroid)
vis.visualize(cluster)
print("")
print(
"-------------------------------------------ici commence la correspondance-------------------------------------------"
)
for file in fichiers:
if not os.path.isdir("./" + dir_features + "/" + file):
#print(file)
with open("./" + dir_features + "/" + file, "rb") as fic:
data = pickle.load(fic)
dist = []
for c in centres:
d = DistanceHu(data[2], c)
dist.append(d)
minima = min(dist)
#print(dist)
kindice = dist.index(minima)
groupe = labels[kindice]
if (groupe == part):
print("voici son fichier de correspondance:", file,
"et son groupe est ", groupe)
break
#while i <nbr_img-1:
#print("voici le probleme:",all_files[i])
# histB=histogramme(rep, all_files[i])
# img2 = cv2.imread('images/' + all_files[i], 0)
#fonction de calcul da la distance d'histogramme
# img2_couleur = cv2.imread('images/' + all_files[i], 1)
# matrix_co=coocurrence(img2_couleur)
# i=i+1