def kkmeans():
# 20孩子，每个孩子10天数据
raw_data = '[[1, 8, 20, 19, 4, 8, 3, 0, 0, 1], [18, 18, 1, 19, 4, 13, 6, 19, 4, 1], [4, 7, 10, 3, 6, 4, 10, 20, 11, 10], [12, 5, 4, 14, 7, 10, 16, 2, 9, 17], [7, 19, 14, 17, 11, 15, 19, 6, 8, 6], [17, 7, 3, 5, 7, 20, 1, 16, 13, 3], [19, 11, 10, 0, 17, 2, 14, 15, 5, 6], [4, 14, 18, 9, 19, 19, 1, 18, 20, 7], [20, 15, 8, 3, 12, 1, 12, 6, 0, 10], [18, 16, 17, 6, 0, 9, 9, 11, 2, 8], [2, 2, 9, 3, 19, 18, 1, 16, 9, 20], [15, 15, 13, 19, 11, 7, 20, 8, 14, 6], [1, 20, 1, 17, 4, 3, 13, 4, 2, 18], [0, 16, 18, 20, 16, 14, 8, 20, 5, 14], [11, 1, 7, 17, 17, 11, 10, 14, 6, 16], [8, 12, 15, 8, 5, 18, 19, 1, 13, 4], [17, 20, 13, 9, 11, 0, 16, 8, 16, 15], [3, 2, 12, 8, 8, 5, 7, 8, 20, 3], [20, 2, 2, 13, 4, 20, 0, 4, 14, 11], [20, 3, 12, 9, 14, 18, 17, 7, 5, 7]]'
data = eval(raw_data)
kmeans = Kmeans(data, 3)
kmeans.get_k_rand()
kmeans.compare_to_k()
kmeans.get_k_avarage()
cl = kmeans.compare_to_k2()
while True:
if len(cl[0]["data"]["rows"]) < 4 or len(
cl[1]["data"]["rows"]) < 4 or len(cl[2]["data"]["rows"]) < 4:
kmeans = Kmeans(data, 3)
kmeans.get_k_rand()
kmeans.compare_to_k()
kmeans.get_k_avarage()
cl = kmeans.compare_to_k2()
data1 = kmeans_helper(cl[0])
data2 = kmeans_helper(cl[1])
data3 = kmeans_helper(cl[2])
data_all = kmeans_helper(cl[0] + cl[1] + cl[2])
pprint.pprint(data1)
pprint.pprint(data2)
pprint.pprint(data3)
pprint.pprint(data_all)
def initialization(self):
# init basic matrices
self.W = np.zeros((self._data_dimension, self._num_bases))
self.H = np.zeros((self._num_bases, self._num_samples))
self.G = np.zeros((self._num_samples, self._num_bases))
#####
# initialize using k-means
km = Kmeans(self.data[:, :],
num_bases=self._num_bases,
show_progress=self._show_progress)
km.initialization()
km.factorize()
assign = km.assigned
num_i = np.zeros(self._num_bases)
for i in range(self._num_bases):
num_i[i] = len(np.where(assign == i)[0])
self.G[range(len(assign)), assign] = 1.0
self.G += 0.01
self.G /= np.tile(np.reshape(num_i[assign], (-1, 1)), self.G.shape[1])
self.H.T[range(len(assign)), assign] = 1.0
self.H += 0.2 * np.ones((self._num_bases, self._num_samples))
self.W = np.dot(self.data[:, :], self.G)
def init_h(self):
if not hasattr(self, 'H'):
# init basic matrices
self.H = np.zeros((self._num_bases, self._num_samples))
# initialize using k-means
km = Kmeans(self.data[:, :],
num_bases=self._num_bases,
seed=self.seed)
km.factorize(niter=10)
assign = km.assigned
num_i = np.zeros(self._num_bases)
for i in range(self._num_bases):
num_i[i] = len(np.where(assign == i)[0])
self.H.T[range(len(assign)), assign] = 1.0
self.H += 0.2 * np.ones((self._num_bases, self._num_samples))
if not hasattr(self, 'G'):
self.G = np.zeros((self._num_samples, self._num_bases))
self.G[range(len(assign)), assign] = 1.0
self.G += 0.01
self.G /= np.tile(np.reshape(num_i[assign], (-1, 1)),
self.G.shape[1])
if not hasattr(self, 'W'):
self.W = np.dot(self.data[:, :], self.G)
def create_splits(self, X):
# get shape of dataset
N, D = X.shape
# thresholds is set of K-Means of each feature
self.thresholds = []
for d in range(D):
# reshape (n,) to (n,1)
feature = X[:, d]
feature = np.reshape(feature, [feature.size, 1])
# Initialize K-Means model
k_means = Kmeans(k=k)
min_err = np.inf
min_err_means = None
for i in range(50):
k_means.fit(feature)
error = k_means.error(feature)
if error < min_err:
min_err = error
min_err_means = k_means.means
self.thresholds.append(min_err_means)
def select_next(iterval):
""" select the next best data sample using robust map
or simply the max iterval ... """
if self._robust_map:
k = np.argsort(iterval)[::-1]
d_sub = self.data[:, k[:self._robust_nselect]]
self.sub.extend(k[:self._robust_nselect])
# cluster d_sub
kmeans_mdl = Kmeans(d_sub, num_bases=self._robust_cluster)
kmeans_mdl.factorize(niter=10)
# get largest cluster
h = np.histogram(kmeans_mdl.assigned,
range(self._robust_cluster + 1))[0]
largest_cluster = np.argmax(h)
sel = pdist(
kmeans_mdl.W[:, largest_cluster:largest_cluster + 1],
d_sub)
sel = k[np.argmin(sel)]
else:
sel = np.argmax(iterval)
return sel
def main():
# 1.读取数据
dataDF = getDF()
# 2.测试最佳K值，第一次出现明显拐角处便是最佳K值
km = Kmeans()
km.searchK(SAVAPATH,dataDF,2,12) # 查看保存的图片，选择最佳K值
def quantize(self, img):
"""
Quantizes an image into 2^b clusters
Parameters
----------
img : a (H,W,3) numpy array
Returns
-------
quantized_img : a (H,W,1) numpy array containing cluster indices
Stores
------
colours : a (2^b, 3) numpy array, each row is a colour
"""
H, W, _ = img.shape
pixels = img.reshape((-1, 3))
model = Kmeans(2**self.b)
model.fit(pixels)
quantized_img = model.predict(pixels).reshape((H, W, 1))
self.colours = model.means
return quantized_img
def main():
# Reading the training data
path_train = './data/EMGaussian.data'
path_test = './data/EMGaussian.test'
data = dp.parse_data_wo_labels(path_train, 2, delimiter=' ')
data_test = dp.parse_data_wo_labels(path_test, 2, delimiter=' ')
# Initialization with K-means
best_kmean_model = None
min_distortion = float("inf")
distortions = []
for i in xrange(NB_INITIALIZATION_RETRIES):
kmean_model = Kmeans(data, NB_CLUSTERS, MAX_K_MEAN_ITER)
kmean_model.run()
distortions.append(kmean_model.distortion)
if kmean_model.distortion < min_distortion:
best_kmean_model = kmean_model
min_distortion = kmean_model.distortion
# Showing the distortions
plt.plot(range(1, NB_INITIALIZATION_RETRIES + 1), distortions)
plt.xlabel("Initialization number")
plt.ylabel("Distortion")
plt.title("Running few Kmeans and measuring the distortion for each")
plt.show()
# Plotting the result
best_kmean_model.plot()
# Case where the covariance matrix is proportional to identity
run_em_model(data, data_test, best_kmean_model, sigma_prop_identity=True)
# General Case
run_em_model(data, data_test, best_kmean_model, sigma_prop_identity=False)
def kmeansselect(self):
kmeans_mdl = Kmeans(self.data, num_bases=self._nsub)
kmeans_mdl.initialization()
kmeans_mdl.factorize()
# pick data samples closest to the centres
idx = dist.vq(kmeans_mdl.data, kmeans_mdl.W)
return idx
def quantize(self, img):
b = self.b
C, R, D = img.shape
self.img = img
X = np.reshape(img, (C * R, D))
model = Kmeans(k=pow(2, b))
model.fit(X)
self.model = model
return model.means
def main():
dataset1 = np.genfromtxt(r'../data/new_dataset_1.txt',
dtype=float,
delimiter='\t')
dataset2 = np.genfromtxt(r'../data/cho.txt', dtype=float, delimiter='\t')
km1 = Kmeans(dataset1[:, 2:], dataset1[:, 1], 3)
km2 = Kmeans(dataset2[:, 2:], dataset2[:, 1], 10)
ic1 = km1.initial_centroids(3, 5, 9)
#ic1 = km1.initial_random_centroids(5)
ic2 = km2.initial_random_centroids(5)
# km1.centroids = km1.init_centroids = np.loadtxt(r'../log/cho_ground_centroids.txt')
# specify iteration as parameter here
km1.kmeans_algorithm()
km2.kmeans_algorithm()
extr_index_validation1 = ExternalIndex(km1.ground_truth_clusters,
km1.clusters)
extr_index_validation2 = ExternalIndex(km2.ground_truth_clusters,
km2.clusters)
print('Rand Index of dataset1 clusters :',
extr_index_validation1.rand_index())
print('Jaccard Coefficient of dataset1 clusters :',
extr_index_validation1.jaccard_coefficient())
print('Rand Index of dataset2 clusters :',
extr_index_validation2.rand_index())
print('Jaccard Coefficient of dataset2 dataset clusters :',
extr_index_validation2.jaccard_coefficient())
plot1 = Visualization(dataset1.data[:, 2:], km1.clusters, dataset1.data[:,
1])
plot2 = Visualization(dataset2.data[:, 2:], km2.clusters, dataset2.data[:,
1])
plot1.plot(r'../log/td1.jpg')
plot2.plot(r'../log/cho2.jpg')
# gene_cluster_matched = km1.cluster_validation()
# print('Genes that matched in clusters: ', gene_cluster_matched)
return
def main():
km = Kmeans(tc.init_board_gauss(nb_points, nb_classe, mini, maxi,
ecart_min, ecart_max),
nb_cluster=nb_cluster,
cpu=cpu,
methode_dist=methode_dist,
adr=img_dir)
km.run_global(choose_nb_graph=True, grphq=True)
km.save(km_path)
print("\n{}".format(km))
return None
def _init_kmeans(self):
"""
Initialize using k-means.
Uses random intialization for k-means. This is a really bad idea.
"""
data = self.data
k = self.k
# Estimate the means of the mixture components, using k-means
km = Kmeans(data, k)
return km.cluster.T, km.label
def quantize(self, X):
N, D, C = X.shape
X_reshaped = np.reshape(X, (N * D, C))
print(X_reshaped)
model = Kmeans(np.power(2, self.b))
model.fit(X_reshaped)
model.predict(X_reshaped)
y = np.reshape(model.predict(X_reshaped), (N, D))
self.means = model.means
self.y = y
self.X = X
def main(_argv):
probki_string, nazwy_atr, czy_atr_symb = wczytaj_baze_probek_z_tekstem(
'spirala.txt', 'spirala-type.txt')
probki = probki_str_na_liczby(probki_string, (0, 1))
grupy, osrodki = Kmeans(probki, FLAGS.groups, FLAGS.iterations, progress)
fig = plt.figure(1)
anim = animation.FuncAnimation(fig,
Animate,
frames=len(progress),
repeat=False,
interval=500)
chart.show()
def create_splits(self, X):
model = Kmeans(3)
N, D = X.shape
splits = np.empty((D * model.k, ))
for d in range(D):
conSplit = X[:, d]
conSplit = np.array(conSplit).transpose()
model.fit(conSplit)
for i in range(model.k):
splits[d] = model.means[i, ]
self.thresholds = np.unique(self.means)
def main():
df = bdd.date_dir(path)
idx, mtx = bdd.df2np(df)
del df
km = Kmeans(mtx, nb_cluster=nb_cluster, cpu=cpu, methode_dist=methode_dist, adr=img_dir, index=idx)
km.run_global(grphq=True, choose_nb_graph=True)
km.save(km_path)
print("\n{}".format(km))
return None
def quantize(self, img):
"""
Quantizes an image into 2^b clusters
Parameters
----------
img : a (H,W,3) numpy array
Returns
-------
quantized_img : a (H,W) numpy array containing cluster indices
Stores
------
colours : a (2^b, 3) numpy array, each row is a colour
"""
H, W, D = img.shape
# model = KMeans(n_clusters=2**self.b, n_init=3)
model = Kmeans(k=2**self.b)
X = np.reshape(img, (H * W, 3))
model.fit(X)
y = model.predict(X)
print(y.shape)
# self.y=y
# self.center=model.means
# Reshape 2D-matrix to 3D-img
# quantized_img = img
# X=np.reshape(img,(H*W,3))
# model.fit(X)
# y=model.predict(X)
# m=y.shape
# print(m)
# quantized_img=y
self.colours = np.zeros((2**self.b, 3), dtype='uint8')
# ,dtype='uint8')
for i in range(2**self.b):
# img[i, :] = quantized_img[i]
self.colours[i, :] = model.means[i, :]
img = np.zeros((H * W), dtype='uint8')
for i in range(H * W):
img[i] = y[i]
img = np.reshape(img, (H, W))
quantized_img = img
# TODO: fill in code here
# raise NotImplementedError()
return quantized_img
def create_splits(self, X):
#k value obtained via elbow method
N, D = X.shape
splits = []
for i in range(D):
model = Kmeans(k=10)
#all values in an example
vec = X[:, i].reshape(N, 1)
model.fit(vec)
threshs = model.means
splits.append(np.squeeze(threshs))
self.thresholds = splits
def quantize_image(self, img):
# w, h, d = img.shape
w, h, d = original_shape = tuple(img.shape)
resized_image = np.reshape(img, (w * h, d))
model = Kmeans(k=(2**self.b))
model.fit(resized_image)
labels = model.predict(resized_image)
self.means = getattr(model, "means")
print("Cluster Assignments")
print(labels)
return labels
def main():
df = bdd.concat_dir(path)
df = bdd.drop_profile(df, drop_var)
df = bdd.bdd2bow(df)
idx, mtx = bdd.df2np(df)
col = df.columns.values.astype(str)
del df
km = Kmeans(mtx, nb_cluster=nb_cluster, cpu=cpu, methode_dist=methode_dist, adr=img_dir, index=idx)
km.run_global(choose_nb_graph=True)
bdd.print_means_words(km, col)
km.save(km_path)
print("\n{}".format(km))
return None
def main():
KMEANS = leerModelo()
if (KMEANS == None):
print('NO EXISTE UN MODELO')
spotify = SpotifyPro()
df = spotify.iniciar(idPlaylist='1QP6tyANnZZ9bRTfQG4X7a')
k = Kmeans(df) # contiene red y datasets
if (len(df)):
k.importarDatos()
if (k.red != None):
guardarModelo(k)
else:
print('Ya existe un modelo')
def dequantize_image(self, img):
w, h, d = img.shape
resized_image = np.reshape(img, (w * h, d))
original_image = np.zeros(img.shape)
model = Kmeans(k=(2**self.b))
model.fit(resized_image)
labels = model.predict(resized_image)
self.means = getattr(model, "means")
label_idx = 0
for i in range(w):
for j in range(h):
original_image[i][j] = self.means[labels[label_idx]]
label_idx += 1
return original_image
def closure_1_3_1():
k = 4
best_model = None
min_error = np.inf
for i in range(50):
model = Kmeans(k)
model.fit(X)
error = model.error(X)
if error < min_error:
min_error = error
best_model = model
plt.figure()
utils.plot_2dclustering(X, best_model.predict(X))
fname = os.path.join("..", "figs",
"kmeans_outliers_best_model.png")
plt.savefig(fname)
print("\nFigure saved as '%s'" % fname)
def test_kmeans(self):
exact_labels = []
label_1 = "Iris-setosa"
label_2 = "Iris-versicolor"
label_3 = "Iris-virginica"
for item in self.data["label"]:
if item == label_1:
exact_labels.append(0)
elif item == label_2:
exact_labels.append(1)
else:
exact_labels.append(2)
k = 3
kmeans = Kmeans(k)
X_train, X_test, y_train, y_test = train_test_split(self.features,
exact_labels,
test_size=0.33,
random_state=42)
kmeans.load_data(X_train.to_numpy().tolist())
kmeans.train()
labels = kmeans.predict(X_test.to_numpy().tolist())
accurate_sum = 0
for i in range(len(labels)):
if labels[i] == y_test[i]:
accurate_sum += 1
print("Akurasi K-Means: ", accurate_sum / len(labels))
kmeans_sklearn = KMeans(n_clusters=3)
kmeans_sklearn.fit(X_train)
sklearn_accurate_sum = 0
for i in range(len(labels)):
if kmeans_sklearn.labels_[i] == y_test[i]:
sklearn_accurate_sum += 1
print("Akurasi K-Means sklearn: ", sklearn_accurate_sum / len(labels))
def _init_parameters(self, X, method='kmeans'):
"""
初始化高斯分布的参数。
如果 method == 'kmeans'，那么使用kmeans进行初始化；
如果 method == 'random'，那么进行随机初始化。
"""
n = X.shape[0]
self.Guass = [Guass_distribution(dim=self.dim) for i in range(self.m)]
if method is 'kmeans':
try:
kmeans = Kmeans()
labels, centroids = kmeans.main(X,
k=self.m,
t=100,
c_strategy='kmeans')
except:
centroids, labels = vq.kmeans2(X,
self.m,
minit='points',
iter=1000)
clusters = [[j for j in range(n) if labels[j] == i]
for i in range(self.m)]
elif method is 'random':
time_seed = int(time.time())
np.random.seed(time_seed)
clusters = [[] for i in range(self.m)]
centroids = random.sample(list(range(n)), self.m) # 随机生成m个中心
for i in range(n):
ci = np.argmin([la.norm(X[i] - X[c]) for c in centroids])
clusters[ci].append(i)
else:
raise ValueError("Unknown method type!")
for i in range(self.m):
guass = self.Guass[i]
data = X[clusters[i]]
guass.init(data)
guass.weight = len(clusters[i]) / n
def sliding_window_three_months(df, date):
starting_date_obj = datetime.datetime.strptime(
date, "%Y-%m-%d")
preprocess_obj = preprocess()
monthly_data = preprocess_obj.get_three_monthly_candlestick_data(
df, date)
kmeans = Kmeans(monthly_data)
e = kmeans.get_clusters()
print('original', e)
ctut = []
ctlt = []
ctbl = []
ctc = []
for i in range(0, len(e)):
ctut.append(e[i][0])
ctlt.append(e[i][1])
ctbl.append(e[i][2])
ctc.append(e[i][3])
candlestickst = candlestickState(ctut, ctlt, ctbl, ctc)
return candlestickst
def main():
#In our dataset, our customers are id'd 0 to 999. Adjust this if dataset changes number of clients
customerValue = [0] * 1000
pointsFile = open('../data/points.json')
breakfastMovementFile = open('../data/breakfast.json')
breakfastPurchaseFile = open('../data/breakfastbuy.json')
lunchMovementFile = open('../data/lunch.json')
lunchPurchaseFile = open('../data/lunchbuy.json')
dinnerMovementFile = open('../data/dinner.json')
dinnerPurchaseFile = open('../data/dinnerbuy.json')
breakfastPurchaseData = json.load(breakfastPurchaseFile)["data"]
for purchase in breakfastPurchaseData:
customerValue[int(purchase["Client ID"])] += int(
float(purchase["Price"]))
lunchPurchaseData = json.load(lunchPurchaseFile)["data"]
for purchase in lunchPurchaseData:
customerValue[int(purchase["Client ID"])] += int(
float(purchase["Price"]))
dinnerPurchaseData = json.load(dinnerPurchaseFile)["data"]
for purchase in dinnerPurchaseData:
customerValue[int(purchase["Client ID"])] += int(
float(purchase["Price"]))
dataset = []
print("Loading dataset...")
#Change the line below to use a different dataset
lunchMovementsData = json.load(lunchMovementFile)["data"]
for customer in lunchMovementsData:
#Data is in x,y. We want row,col
dataset.append([
int(customer["X"]), -int(customer["Y"]) + 55,
customerValue[int(customer["ID"])]
])
print("Done!")
Kmeans(4, dataset)
def clusterpixels(infile, k, steps):
im = np.array(Image.open(infile))
dx = int(im.shape[0] / steps)
dy = int(im.shape[1] / steps)
# 计算每个组件的图像特征
features = []
for x in range(steps):
for y in range(steps):
R = np.mean(im[x * dx:(x + 1) * dx, y * dy:(y + 1) * dy,
0]) #行、列、颜色通道
G = np.mean(im[x * dx:(x + 1) * dx, y * dy:(y + 1) * dy, 1])
B = np.mean(im[x * dx:(x + 1) * dx, y * dy:(y + 1) * dy, 2])
features.append([R, G, B])
features = np.array(features, 'f') # 将特征值变换为数组矩阵形式
# 聚类， k是聚类数目
centroids, variance, iternum = Kmeans(features, k)
code, distance = vq(features, centroids) #进行矢量量化，使用获得的聚类标签画图
codeim = code.reshape(steps, steps) #给数组一个新的形状而不改变其数据
codeim = imresize(codeim, im.shape[:2],
'nearest') #imresize() 方法,用来指定新图像的大小
return codeim
def kmeans(imagens, segmentadas, path):
k = Kmeans(imagens)
qtd = 0
for i in imagens:
# Leitura Imagem
img = imread(imagens[qtd][0])
#img = cv2.resize(img, (segmentadas[qtd][2], segmentadas[qtd][1]))
res2 = k.kmeans_seg(img, 2) / 255
res3 = k.kmeans_seg(img, 3) / 255
res9 = k.kmeans_seg(img, 5) / 255
cv2.imwrite("res2.png", res2)
cv2.imwrite("res3.png", res3)
cv2.imwrite("res9.png", res9)
fig = plt.figure(figsize=(9,3), dpi=200)
k.add_image(fig, img, 1, 4, 1, 'original')
k.add_image(fig, res2, 1, 4, 2, 'k=2')
k.add_image(fig, res3, 1, 4, 3, 'k=3')
k.add_image(fig, res9, 1, 4, 4, 'k=5')