def fit(self, X, y):
if self.inferStds:
# compute stds from data
self.centers, self.stds = km.k_means(X, self.k)
else:
# use a fixed std
self.centers, _ = km.k_means(X, self.k)
dMax = max([
np.abs(c1 - c2) for c1 in self.centers for c2 in self.centers
])
self.stds = np.repeat(dMax / np.sqrt(2 * self.k), self.k)
# training
for epoch in range(self.epochs):
for i in range(X.shape[0]):
# forward pass
a = np.array([
self.rbf(X[i], c, s)
for c, s, in zip(self.centers, self.stds)
])
F = a.T.dot(self.w) + self.b
loss = (y[i] - F).flatten()**2
# print('Loss: {0:.2f}'.format(loss[0]))
# backward pass
error = -(y[i] - F).flatten()
# online update
#chu = np.dot(a , error)
self.w = self.w - self.lr * a * error
self.b = self.b - self.lr * error
def compare_dist(iters=100): pts = gen_data() cs = [] dist = [] for i in range(iters): c,_ = kmeans.k_means(pts, 3) cpp,_ = kmeans.k_means(pts, 3, kmeans.k_means_pp_initialization) d = kmeans.compute_avg2_distortion(pts, c) dpp = kmeans.compute_avg2_distortion(pts, cpp) dist.append([d,dpp]) cs.append(np.array([c,cpp])) return np.array(dist), np.array(cs)
def validar_agrupamento(df):
# Calcula kmeans quando K=2, K=3, K=4
for i in range(2, 5):
# Remove a última coluna, pois ela é a classe, vou utilizar ela para o algoritmo kmeans
data_sem_classe = df.drop(df.columns[-1], axis=1, inplace=False)
# Chama a função para realizar o kmeans
data = k_means(data_sem_classe, i)
# Adiciona a coluna "Classe" de volta ao Dataframe
data['Classe'] = df[df.columns[-1]]
# Agrupa por 'Classe' e 'Grupo'
grouped_df = data.groupby(["Classe", "Grupo"])
# Converte o objeto para Dataframe e calcula a contagem em cada grupo
data_contagem = pd.DataFrame(
grouped_df.size().reset_index(name="Contagem"))
# Guarda em uma variável todos os tipos de classe, removendo aquelas repetidas por estarem em mais de um grupo
classeList = sorted(set(data_contagem['Classe'].values.tolist()))
# Monta uma estrutura de Dataframe para calcular a pureza
data_estrutura = montar_estrutura(data_contagem, classeList, i)
# Calcula a pureza
data_pureza = calcula_pureza(data_estrutura, i)
def main():
clusters = args.clusters
if clusters not in [4, 6, 8, 10]:
raise ValueError("Number of clusters must be 4, 6, 8, or 10!")
data_path = './data/Mall_Customers.csv'
attr = ['Gender', 'Age', 'Annual Income (k$)', 'Spending Score (1-100)']
f1, f2 = 'Age', 'Spending Score (1-100)'
df = pd.read_csv(data_path)
data = df[[f1, f2]].iloc[:, :].to_numpy()
if f1 == 'Gender':
gender = data[:, 0]
gender[gender == 'Male'] = 5.
gender[gender == 'Female'] = 10.
print(gender)
data[:, 0] = gender
print(f'Shape of the data: {data.shape}')
km = k_means(num_clusters=clusters, tol=1e-4)
est_centroid, history_centroids, predict_labels, loss, num_iter = km.get_cluster(
data)
plotting(predict_labels, data, clusters, est_centroid, f1, f2)
def outer_criteria(X, k_range, reference, iterations=200):
n = len(X)
RS = dict() # Rand Statistic
FM = dict() # Fowlkes-Mallows
for k in k_range:
TP, FN, FP, TN = (0, ) * 4
if k < 2:
continue
_, clusters = kmeans.k_means(X, k, iterations)
# Compute TP, FN, FP, TN.
for i in range(n):
for j in range(i + 1, n):
if clusters[i] == clusters[j] and reference[i] == reference[j]:
TP += 1
elif clusters[i] != clusters[j] and reference[i] == reference[
j]:
FN += 1
elif clusters[i] == clusters[
j] and reference[i] != reference[j]:
FP += 1
else:
TN += 1
RS[k] = (TP + TN) / n
precision = TP / (TP + FP)
recall = TP / (TP + FN)
FM[k] = np.sqrt(precision * recall)
best_rs = max(RS, key=RS.get)
best_fm = max(FM, key=FM.get)
return RS, best_rs, FM, best_fm
def k_means_test():
X = np.concatenate((np.random.normal(5, 30,
100), np.random.normal(150, 44, 100),
np.random.normal(300, 20, 100)))
Y = np.concatenate((np.random.normal(5, 30,
100), np.random.normal(150, 44, 100),
np.random.normal(300, 20, 100)))
data = np.column_stack((X, Y))
clusters, result, centroids = k_means(data, 3, euclidean_dist)
cluster_points = []
for cluster in clusters:
Xaux = [X[i] for i in cluster]
Yaux = [Y[i] for i in cluster]
cluster_points.append((Xaux, Yaux))
plt.figure()
colors = [
'green', 'red', 'blue', 'yellow', 'gray', 'lightblue', 'pink', 'black'
]
i = 0
for cluster in cluster_points:
X, Y = cluster
plt.scatter(X, Y, c=colors[i])
i = (i + 1) % len(colors)
plt.show()
def spectral_cluster(W, k):
L = get_laplacian(W)
Lambda, v = np.linalg.eig(L)
length = len(Lambda)
eigen_pairs = [(Lambda[i], v[:, i]) for i in range(length)]
eigen_pairs = sorted(eigen_pairs, reverse=False, key=lambda k: k[0])
temp = np.column_stack((eigen_pairs[i][1] for i in range(k)))
result = kmeans.k_means(temp.T, 2, 5)
return result
def main():
print "\n>> Since, matplotlib has inconsistent behaviour, I am saving all the generated plots in the directory './images'\n"
if not os.path.exists('./images'):
os.mkdir('./images')
print "> Running K-Means for blob_data\n\n"
blob_data = np.genfromtxt('./hw5_blob.csv', delimiter=',')
for k in [2, 3, 5]:
clusters = k_means(blob_data, k, "blob", True)
print "> Running K-Means for circle_data\n\n"
circle_data = np.genfromtxt('./hw5_circle.csv', delimiter=',')
for k in [2, 3, 5]:
clusters = k_means(circle_data, k, "circle", True)
print "> Running Kernel K-Means for circle_data\n\n"
kernel_kmeans.main()
print "> Running EM Algorithm for blob_data\n\n"
em_algorithm.main()
def compress_image(im_path, k, init_f=uniform_mode_dist_init):
"""
returns the original and compressed version of an image together
with time profile data
Arguments:
im_path: string
k: int
init_f: function of 2d array x 1d array x int x function -> 2d array
Output:
image: numpy 2d numerical array
compressed_image numpy 2d numerical array
mse: float
time_profile: dict of string -> float
"""
# Read image
image = cv2.imread(im_path)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Store original shape
original_shape = image.shape
# Reshape into a numpy 2d array
image = image.reshape([image.shape[0] * image.shape[1], 3])
# Run k-means
t0 = time.time()
c_means, clusters, mse, time_profile = k_means(image, k, rgb_distance,
init_f, pixel_to_str,
str_to_pixel)
t1 = time.time()
time_profile['k_means'] = t1 - t0
# Create compressed image
compressed_image = np.zeros(image.shape, dtype=np.uint8)
for i in range(compressed_image.shape[0]):
compressed_image[i] = c_means[clusters[i]].astype(np.uint8)
# Return to original shape
image = image.reshape(original_shape).astype(np.uint8)
compressed_image = compressed_image.reshape(original_shape).astype(
np.uint8)
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
compressed_image = cv2.cvtColor(compressed_image, cv2.COLOR_RGB2BGR)
aid = ptp_idm(image, compressed_image)
return image, compressed_image, mse, aid, time_profile
def main(argv):
config_path = args.conf
num_anchors = args.anchors
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
data = {
'train': {
'image_folder': config['train']['train_image_folder'],
'annot_folder': config['train']['train_annot_folder'],
},
'valid': {
'image_folder': config['valid']['valid_image_folder'],
'annot_folder': config['valid']['valid_annot_folder'],
}
}
video_folder_list, video_annot_list = data_preparation(data['train'],
FOR_YOLO=True)
grid_w = config['model']['input_size'] / 32
grid_h = config['model']['input_size'] / 32
cell_w = 1280.0 / grid_w
cell_h = 720.0 / grid_h
# run k_mean to find the anchors
annotation_dims = []
for video_annot in video_annot_list:
labels = np.loadtxt(video_annot, delimiter=',')
for label in labels:
relative_w = label[2] / cell_w
relative_h = label[3] / cell_h
if math.isnan(relative_w) or math.isnan(relative_h):
# print("NaN annotations! {}".format(basename(video_annot)))
1
else:
annotation_dims.append(map(float, (relative_w, relative_h)))
annotation_dims = np.array(annotation_dims)
centroids, cluster_assignment = k_means(annotation_dims, num_anchors)
# write anchors to file
print '\naverage IOU for', num_anchors, 'anchors:', '%0.2f' % avg_IOU(
annotation_dims, centroids)
print_anchors(centroids)
def main():
#data, label = datasets.load_iris().data, datasets.load_iris().traget
#data = load_data('F:\machine learning\PythonCode\dataset\cluster\spectral_cluster_data.txt')
data, label = load_data_with_label('F:\machine learning\PythonCode\dataset\cluster\Spiral.txt')
data = normal_data(data)
#显示原来的数据分布
plt.ion()
#kmeans.mat_plot_sample(data)
mat_plot_cluster_sample(data, label, 'original sample')
# k-means聚类，对比实验
k_dist1, k_centers1, cluster_group = kmeans.k_means(data, 3)
kmeans.mat_plot_k_means_sample(k_dist1, k_centers1)
# 谱聚类，第三个参数为聚类是采用的算法，有'NJW'、'self-tuning'、'sl',默认为'sl'
spectral_cluster(data, 3)
spectral_cluster(data, 3, 'NJW')
spectral_cluster(data, 3, 'self-tuning')
plt.ioff()
plt.show()
def training_model(number_of_templates, using_kmeans, read_feature_from_file=0):
"""
training the model using DTW to recognize the records
:param number_of_templates: int, using number_of_templates to train the model
:param using_kmeans: 1 if using k-means to templates to generate some states
:param read_feature_from_file:1 then read mfcc feature from file,0 then compute mfcc feature from records
:return: DTW_obj:a DTW object,the trained model
:return: inputs:list,every element of it is a template using for test
"""
templates = []
inputs = []
for digit in xrange(0, 10):
temp = get_templates_and_inputs(digit, number_of_templates, read_feature_from_file)
# print 'temp',temp[0]
# print 'k_means(temp[0], number_of_states=5)',kmeans.k_means(temp[0], number_of_states=5)
templates.extend(
kmeans.k_means(temp[0], number_of_states=5)[0] if using_kmeans and number_of_templates == 5 else temp[0])
inputs.extend(temp[1])
DTW_obj = DTW.DTW(templates)
return DTW_obj, inputs
def spectral_clustering(data, k, gamma=0.1):
print("Spectral clustering...")
# Number of samples
n = data.shape[0]
# Adjacency matrix
W = rbf_kernel_gram_matrix(data, gamma)
# Build Degree matrix
D = compute_degrees(W)
# Graph Laplacian
L = D - W
# Normalized Cut
D_inv_sqrt = scipy.linalg.fractional_matrix_power(D, -1 / 2)
L_sym = D_inv_sqrt.dot(L).dot(D_inv_sqrt)
# Compute the first k eigenvectors of L_sym
eig_values, eig_vectors = LA.eig(L_sym)
# T contains first k eigenvectors of normalized Laplacian
T = np.zeros((n, k))
for i in range(k):
T[:, i] = eig_vectors[:, i]
# Resubstitude matrix H and normalize by its rows.
H = D_inv_sqrt.dot(T)
H /= LA.norm(H, axis=1, ord=2)[:, np.newaxis]
# Cluster data points in eigenspace
centroids, cluster_assignment = k_means(H, k)
# Show the data points at the same point in eigenspace.
# discrete_h = np.zeros_like(H)
# c, discrete_h[:, 0] = k_means(H[:, 0].reshape(-1, 1), k)
# c, discrete_h[:, 1] = k_means(H[:, 1].reshape(-1, 1), k)
# centroids, cluster_assignment = k_means(discrete_h, k)
return cluster_assignment
def inner_criteria(X, k_range, iterations=200):
DB = dict() # Davies-Bouldin
CH = dict() # Calinski-Harabasz
for k in k_range:
if k < 2:
continue
centroids, clusters = kmeans.k_means(X, k, iterations)
DB[k] = DaviesBouldin(X, centroids, clusters)
CH[k] = CHIndex(X, centroids, clusters)
ch_list = list(CH.items())
db_best = min(DB, key=DB.get)
ch_best = 0
delta = sys.maxsize
for k in range(1, len(CH) - 1):
temp = ch_list[k + 1][1] - 2 * ch_list[k][1] + ch_list[k - 1][1]
if temp < delta:
delta = temp
ch_best = ch_list[k][0]
return DB, db_best, CH, ch_best
def training_model(number_of_templates,
using_kmeans,
read_feature_from_file=0):
"""
training the model using DTW to recognize the records
:param number_of_templates: int, using number_of_templates to train the model
:param using_kmeans: 1 if using k-means to templates to generate some states
:param read_feature_from_file:1 then read mfcc feature from file,0 then compute mfcc feature from records
:return: DTW_obj:a DTW object,the trained model
:return: inputs:list,every element of it is a template using for test
"""
templates = []
inputs = []
for digit in xrange(0, 10):
temp = get_templates_and_inputs(digit, number_of_templates,
read_feature_from_file)
# print 'temp',temp[0]
# print 'k_means(temp[0], number_of_states=5)',kmeans.k_means(temp[0], number_of_states=5)
templates.extend(
kmeans.k_means(temp[0], number_of_states=5)[0]
if using_kmeans and number_of_templates == 5 else temp[0])
inputs.extend(temp[1])
DTW_obj = DTW.DTW(templates)
return DTW_obj, inputs
f = open( 'docs.json', 'r' )
for doc in f.readlines():
#Remove the '\n' at the end of doc
doc = doc.strip('\n')
doc_dict = json.loads( doc )
Doc_List.append( doc_dict )
#obtain Title_List
Title_List.append( doc_dict[ 'title' ] )
f.close()
#Calculate tf-idf of each word in each doc:
#doc_collection's structure: [doc_vector1,doc_vector2,...]
#doc_vector's structure:{word1:tf-idf val, word2:tf-idf val,...}
doc_collection = TF_IDF()
#this block will generate value_collection. Its structure is: [[1,2,0,1,2,...],[0,2,1,3,0...],...]
value_collection = []
for doc_vector in doc_collection:
doc_value = []
for word in doc_vector:
doc_value.append( doc_vector[ word ] )
value_collection.append( doc_value )
#print value_collection
#K Means is coming!
k = raw_input( "Please input a value for k: " )
k = int( k )
KMeans = kmeans.k_means( k )
KMeans.K_MEANS( value_collection, Title_List )
def serial_kmeans(data, k, dist): return kmeans.k_means(data,k,dist)
dataSet = []
# 打开数据文件
fileIn = open('testSet.txt')
# 读取每一行，使用中间的tab进行切割
for line in fileIn.readlines():
# 切割
lineArr = line.strip().split('\t')
# 存入数据列表(有序)，转化为float数据类型，数据文件的每行根据tab分割为两部分，使用下标来访问
dataSet.append([float(lineArr[0]), float(lineArr[1])])
# --------------------------
# **************************
# step 2: clustering...
# 第二步：聚类...
print "第二步：聚类..."
# 使用mat函数把列表(数组)转换成矩阵
dataSet = mat(dataSet)
# 参数k为聚类中心数目
k = 4
# 执行kmeans文件中的k_means函数，传入矩阵和聚类中心数目k参数，分别赋值
centroids, clusterAssment = kmeans.k_means(dataSet, k)
# **************************
# ++++++++++++++++++++++++++
# step 3: show the result...
# 第三步：显示结果...
print "第三步：显示结果..."
# 执行kmeans文件中的显示图表函数
kmeans.showCluster(dataSet, k, centroids, clusterAssment)
# ++++++++++++++++++++++++++
reader = csv.reader(f) #iterator object
vote_topic = next(reader) #acc. to csv data
headers = next(reader) #acc. to csv data
for person, state, district, vote, name, party in reader:
senator = Senator(name, party, state)
accumulated_record[senator].append(vote_value[vote])
pprint(accumulated_record, width=500)
record = {
senator: tuple(votes)
for senator, votes in accumulated_record.items()
} #type: Dict[Senator,VoteHistory]
pprint(record, width=500)
#use kmeans to locate the cluster centroids from patterns of votes, assign each senetor to the nearest cluster
centroids = k_means(data=record.values(), num_centroids=3)
clustered_votes = assign_data(centroids=centroids, data=record.values())
pprint(clustered_votes)
#build a reverse mapping from a vote history to a list of senators that voted that way
votes_to_senators = defaultdict(
list) #type: DefaultDict[VoteHistory, List[Senator]]
for senator, vote_history in record.items():
votes_to_senators[vote_history].append(senator)
#we know that come senators might have the same voting history,
#also check if all 100 are present even after rearranging by using assertion
assert sum([len(cluster)
for cluster in votes_to_senators.values()]) == NUM_SENATORS
#display clusters and members of each clusters
def cluster_k(db_path, file_name, file_name_p, n_clusters):
score = kmeans.k_means(dump_path=db_path, file_name=file_name, file_name_p=file_name_p, n_clusters=int(n_clusters))
print '{0}:{1}'.format(n_clusters, score)
vote_value = {'Nay': -1, 'Not Voting': 0, 'Yea': 1} # type: Dict[str, VoteValue]
accumulated_record = defaultdict(list) # type: DefaultDict[Senator, List[VoteValue]]
for filename in glob.glob('congress_data/*.csv'):
with open(filename) as f:
reader = csv.reader(f)
vote_topic = next(reader)
headers = next(reader)
for person, state, district, vote, name, party in reader:
senator = Senator(name, party, state)
accumulated_record[senator].append(vote_value[vote])
# Transform record into plain dict mapping a senator to a tuple of vote values
record = {senator: tuple(votes) for senator, votes in accumulated_record.items()} # type: Dict[Senator, Tuple[VoteValue, ...]]
# Use k-means to locate the cluster centroids and assign senators to the nearest cluster
centroids = k_means(record.values(), k=3, iterations=50)
clustered_votes = assign_data(centroids, record.values())
# Build a reverse mapping from a pattern of votes to senators who voted that way
votes_to_senators = defaultdict(list) # type: DefaultDict[Tuple[VoteValue, ...], List[Senator]]
for senator, votes in record.items():
votes_to_senators[votes].append(senator)
assert sum(map(len, clustered_votes.values())) == 100
# Display the clusters and the members of each cluster
for i, votes_in_cluster in enumerate(clustered_votes.values(), start=1):
print(f'=========== Voting Cluster #{i} ===========')
party_totals = Counter() # type: Counter
for votes in set(votes_in_cluster):
for senator in votes_to_senators[votes]:
party_totals[senator.party] += 1
for i in range(0, len(test_dataset_np)):
if test_dataset_np[i][0] == 1:
test_count[0] += 1
if test_dataset_np[i][0] == 4:
test_count[1] += 1
if test_dataset_np[i][0] == 8:
test_count[2] += 1
#print(train_count)
train_dataset_np_order = train_dataset_np[
train_dataset_np[:, 0].argsort()] #The number:1:1005, 4:652, 8:542
test_dataset_np_order = test_dataset_np[test_dataset_np[:, 0].argsort()]
#print(train_dataset_np_order)
result = k_means(train_dataset_np_order, 3, 1000)
print(result[1])
#A method to assign class labels to each of your clusters：
acc = Cal_Accuracy_testdata(result[1], train_count)
print("The acc of training is :", acc)
#Predict labels on the zip.test data:
labels = testdata_classify(test_dataset_np_order, result[2])
print("Print labels", labels)
#Acc:
acc_test = Cal_Accuracy_testdata(labels, test_count)
print("The acc of test is :", acc_test)
#PCA:
def get_transform_relationship_FSM_table(filename, states_in_each_word=5, using_continuous_feature=0, **kwargs):
"""
从有限状态机中获取模板,状态转移列表,起始状态,每个nonemitting state所对应的emitting state 在模板中连续
:param filename: 有限状态机文件名
:param states_in_each_word: 每个单词有多少states组成
:param kwargs:
:return: template,states的模板
transform_list,第i个元素为列表,代表第i个state可以从哪些states进入
begin_list,t=0时可进入的states
word_in_template,第i个表示template中第i个元素代表的word
"""
start_states, terminal_states, nonemitting_transform_list = get_info_FSM_table(filename)
number_of_nonemitting_states = len(nonemitting_transform_list)
template = []
digit_template = [] # 第i个值表示数字i的template
number_of_frames_in_each_state = [] # 第i个元素代表state i的frame数
number_of_frames_in_each_state_for_digit = [] # 第i个元素代表数字i中每个state拥有的frame数量
covariance_matrix_in_each_state = []
mean_in_each_state = []
covariance_matrix_for_each_digit = []
mean_for_each_digit = []
word_in_template = [] # 第i个表示template中第i个元素代表的word
if using_continuous_feature:
digit_template = SR.get_digit_feature_from_continuous_speech()
for digit in xrange(0, 10):
temp = SR.get_templates_and_inputs(digit, number_of_templates=10)
if not using_continuous_feature:
result = kmeans.k_means(temp[0], number_of_states=5)
digit_template.extend(result[0])
covariance_matrix_for_each_digit.append(result[2])
mean_for_each_digit.append(result[3])
temp_number_of_frames_in_each_state = [0 for i in xrange(states_in_each_word)]
for number_of_frames_in_each_state_in_each_template in result[5]:
for i in xrange(states_in_each_word):
temp_number_of_frames_in_each_state[i] += number_of_frames_in_each_state_in_each_template[i]
number_of_frames_in_each_state_for_digit.append(temp_number_of_frames_in_each_state)
# print number_of_frames_in_each_state
# 获取number_of_emitting_states_begin_from_nonemitting_states
number_of_emitting_states_begin_from_nonemitting_states = [] # 第i个元素表示源于第i个nonemitting_state的edge包含的states个数
for nonemitting_state in nonemitting_transform_list:
# nonemitting_state形如{1: [2, 3, 4, 5, 6, 7, 8, 9], 3: []}
temp = []
for edge_list in nonemitting_state.values():
temp.extend(edge_list)
for i in temp:
# 获取template
covariance_matrix_in_each_state.extend(covariance_matrix_for_each_digit[i])
mean_in_each_state.extend(mean_for_each_digit[i])
number_of_frames_in_each_state.extend(number_of_frames_in_each_state_for_digit[i])
template.extend(digit_template[i])
word_in_template.append(i)
number_of_emitting_states = len(set(temp))
number_of_emitting_states_begin_from_nonemitting_states.append(number_of_emitting_states)
# 获取begin_states_index_for_each_nonemitting_state
begin_states_index_for_each_nonemitting_state = [] # 第i个元素表示源于第i个nonemitting_state的edge包含的states之前有多少states
for i in xrange(number_of_nonemitting_states):
begin_states_index_for_each_nonemitting_state.append(
sum(number_of_emitting_states_begin_from_nonemitting_states[:i]))
# 获取emitting_out_list,第i个元素为一个列表,代表第i个nonemitting state可以进入的states
emitting_out_list = [[] for i in xrange(number_of_nonemitting_states)]
for i, nonemitting_state in enumerate(nonemitting_transform_list):
# nonemitting_state形如{1: [2, 3, 4, 5, 6, 7, 8, 9], 3: []}
begin_index = begin_states_index_for_each_nonemitting_state[i]
number_of_edges = sum(map(len, nonemitting_state.values()))
for j in xrange(number_of_edges):
emitting_out_list[i].append(states_in_each_word * (begin_index + j))
changed = True
# 考虑无条件跳转的情况
while changed:
new_emitting_out_list = emitting_out_list[:]
for i, nonemitting_state in enumerate(nonemitting_transform_list):
for key in nonemitting_state.keys():
if nonemitting_state[key] == []:
new_emitting_out_list[i].extend(new_emitting_out_list[key])
changed = (new_emitting_out_list != emitting_out_list)
emitting_out_list = new_emitting_out_list
# 获取emitting_in_list,第i个元素为一个列表,代表第i个nonemitting state可以由哪些states进入
emitting_in_list = [[] for i in xrange(number_of_nonemitting_states)]
for i, nonemitting_state in enumerate(nonemitting_transform_list):
# nonemitting_state形如{1: [2, 3, 4, 5, 6, 7, 8, 9], 3: []}
begin_index = begin_states_index_for_each_nonemitting_state[i]
j = 0
for key in nonemitting_state.keys():
for value in nonemitting_state[key]:
emitting_in_list[key].append(states_in_each_word * (begin_index + j))
j += 1
changed = True
# 考虑无条件跳转的情况
while changed:
new_emitting_in_list = emitting_in_list[:]
for i, nonemitting_state in enumerate(nonemitting_transform_list):
for key in nonemitting_state.keys():
if nonemitting_state[key] == []:
new_emitting_in_list[key].extend(new_emitting_in_list[i])
changed = (new_emitting_in_list != emitting_in_list)
emitting_in_list = new_emitting_in_list
# 计算transform_list,每个元素为一个列表,表示可以进入这个state的states
transform_list = [[] for i in
xrange(sum(number_of_emitting_states_begin_from_nonemitting_states) * states_in_each_word)]
for i in xrange(number_of_nonemitting_states):
for next_state in emitting_out_list[i]:
for cur_state in emitting_in_list[i]:
transform_list[next_state].append(cur_state + 4)
for i, element in enumerate(transform_list):
element.append(i)
if i % states_in_each_word > 0:
element.append(i - 1)
if i % states_in_each_word > 1:
element.append(i - 2)
begin_states = emitting_out_list[0]
return template, transform_list, begin_states, word_in_template, number_of_frames_in_each_state, covariance_matrix_in_each_state, mean_in_each_state
vote_value: Dict[str, VoteValue] = {'Nay': -1, 'Not Voting': 0, 'Yea': 1}
accumulated_record: DefaultDict[Senator, List[VoteValue]] = defaultdict(list)
for filename in glob.glob('congress_data/*.csv'):
with open(filename, encoding='utf-8') as f:
reader = csv.reader(f)
vote_topic = next(reader)
headers = next(reader)
for person, state, district, vote, name, party in reader:
senator = Senator(name, party, state)
accumulated_record[senator].append(vote_value[vote])
# Transform record into plain dict mapping a senator to a tuple of vote values
record: Dict[Senator, Tuple[VoteValue, ...]] = {senator: tuple(votes) for senator, votes in accumulated_record.items() }
# Use k-means to locate the cluster centroids and assign senators to the nearest cluster
centroids = k_means(record.values(), k=3, iterations=50)
clustered_votes = assign_data(centroids, record.values())
# Build a reverse mapping from a pattern of votes to senators who voted that way
votes_to_senators: DefaultDict[Tuple[VoteValue, ...], List[Senator]] = defaultdict(list)
for senator, votes in record.items():
votes_to_senators[votes].append(senator)
assert sum(len(cluster) for cluster in clustered_votes.values()) == 100
# Display the clusters and the members of each cluster
for i, votes_in_cluster in enumerate(clustered_votes.values(), start=1):
print(f'=========== Voting Cluster #{i} ===========')
party_totals: Counter = Counter()
for votes in set(votes_in_cluster):
for senator in votes_to_senators[votes]:
party_totals[senator.party] += 1
import seaborn as sns
import numpy as np
import pandas as pd
from skimage import io
#import imp
#kmean = imp.load_source('./kmeans.py')
#import kmean
import kmeans as keman
pic = io.imread('./data/bird_small.png') / 255.
#io.imshow(pic)
data = pic.reshape(128 * 128, 3)
#k-mean
C, centroids, cost = keman.k_means(pd.DataFrame(data), 16, epoch=10, n_init=3)
compressed_pic = centroids[C].reshape(
(128, 128, 3)) #a[b]中每个元素为b[i]为索引，对应a中的每行
#sklearn KMeans
#from sklearn.cluster import KMeans
#model = KMeans(n_clusters=16, n_init=100, n_jobs=-1)
#model.fit(data)
#centroids = model.cluster_centers_
#C = model.predict(data)
#compressed_pic = centroids[C].reshape((128,128,3))
fix, ax = plt.subplots(1, 2)
ax[0].imshow(pic)
ax[1].imshow(compressed_pic)
plt.show()
NC=list(range(3, 4))#numOfModels+1)) # list of numbers of clusters
accuracies = np.zeros((numOfEPS, len(NC)))
clusteringResult = {}
for numOfClusters in NC:
clusteringResult[numOfClusters] = []
with open("clustering_result.txt", "w") as fp:
for numOfClusters in NC:
# clustering into c groups
print("Clustering: {} clusters".format(numOfClusters))
# kmeans = KMeans(n_clusters=numOfClusters, random_state=0).fit(predVec)
# for c in range(numOfClusters):
# clusteringResult[numOfClusters].append(np.where(kmeans.labels_ == c)[0])
# print(np.where(kmeans.labels_ == c)[0])
assignments = k_means(predVec, numOfClusters, "L2", "ZerosFarAway")
fp.write("## number of clusters: "+str(numOfClusters)+"\n")
for c in range(numOfClusters):
cluster = np.where(assignments==c)[0]
clusteringResult[numOfClusters].append(cluster)
print(cluster)
fp.write("\t"+str(cluster)+"\n")
fp.write("\n")
def vote1(participants):
'''
Input:
participants: a list of opinions. Each element in the list is a numpy array, N X 2.
N is the number of events. The second dimension contains (opinion/label, confidence)
Output:
voteResult : a numapy array NX2 represents opinion and confidence across N events
reader = csv.reader(f)
vote_topic = next(reader)
header = next(reader)
for person, state, district, vote, name, party in reader:
senator = Senator(name, party, state)
accumulated_record[senator].append(vote_value[vote])
# Transform the record into a plain dict that maps to a tuple of votes.
record = {
senator: tuple(votes)
for senator, votes in accumulated_record.items()
} # type: Dict[Senator, VoteHistory]
# Use k-means to locate the cluster centroids from pattern of votes, assign
# each senator to the nearest cluster.
centroids = k_means(record.values(), k=3)
clustered_votes = assign_data(centroids, record.values())
# Build a reverse mapping from a vote history to a list of senators who voted
# that way.
votes_to_senators: DefaultDict[VoteHistory, List[Senator]] = defaultdict(list)
for senator, votehistory in record.items():
votes_to_senators[votehistory].append(senator)
assert sum(len(cluster)
for cluster in votes_to_senators.values()) == NUM_SENATORS
# Display the clusters and the members (senators) of each cluster.
for i, votes_in_cluster in enumerate(clustered_votes.values(), start=1):
print(f"==================== Voting Cluster #{i} ====================")
party_totals: Counter[str] = Counter()
for votes in set(votes_in_cluster):
if __name__ == "__main__":
data = sio.loadmat('ExtYaleB10.mat')
train_sample = data['train']
test_sample = data['test']
x_train_full = np.column_stack(train_sample[0, i][:, :,
j].reshape(192 * 168, 1)
for i in range(10) for j in range(50))
x_test_full = np.column_stack(test_sample[0, i][:, :,
j].reshape(192 * 168, 1)
for i in range(10) for j in range(14))
I = np.identity(10)
y_train = np.column_stack(I[:, i] for i in range(10) for j in range(50))
y_test = np.column_stack(I[:, i] for i in range(10) for j in range(14))
print("spectral_clustering computing")
v = spectral.spectral_clustering(x_train_full, 10, 1, 10)
for i in range(v.shape[1]):
v[:, i] = v[:, i] / (la.norm(v[:, i]))
final_c, final_z = km.k_means(np.asarray(v.T), 10, 10)
y_train = np.zeros(500, )
cost = 0
for i in range(10):
for j in range(50):
y_train[i * 50 + j] = i
for i in range(500):
for j in range(10):
if final_z[i, j] == 1:
if y_train[i] != j:
cost += 1
print("Errors of spectral clustering", cost)
with open(filename) as f:
reader = csv.reader(f)
vote_topic = next(reader)
headers = next(reader)
for person, state, district, vote, name, party in reader:
senator = Senator(name, party, state)
accumulated_record[senator].append(vote_value[vote])
# Transform record into a plain dict maps a senator to a tuple of vote values
record = {
senator: tuple(votes)
for senator, votes in accumulated_record.items()
}
# Show-off our talent with k-means advances machine learning like skynet or the HAL5000 or Deep Blue
centroids = kmeans.k_means(list(record.values()), k=2, iterations=50)
clustered_votes = kmeans.assign_data(centroids, list(
record.values())) # type: Dict[Tuple[VoteValue], List[Tuple[VoteValue]]
# Build a reverse mapping. Given a voting record, get the list of senators who voted that way
votes_to_senators = collections.defaultdict(list)
for senator, votes in record.items():
votes_to_senators[votes].append(senator)
# Display the clusters and the people who voted that way
for i, votes_in_clusters in enumerate(clustered_votes.values(), start=1):
print(
f'======================= Voting Cluster #{i} ========================='
)
for votes in set(votes_in_clusters):
for senator in votes_to_senators[votes]:
if __name__ == '__main__':
data_dir = 'data'
results_dir = 'results'
os.makedirs(results_dir, exist_ok=True)
clusters_range = range(2, 10)
# Inner criterias' block.
image_path = os.path.join(data_dir, 'policemen.jpg')
image = np.array(Image.open(image_path), dtype=np.uint8)
new_image = image.reshape(image.shape[0] * image.shape[1], image.shape[2])
db, best_db, ch, best_ch = inner_criteria(new_image, clusters_range)
best_inner = (best_db + best_ch) // 2
# Save the clustered image.
centroids, clusters = kmeans.k_means(new_image, best_inner, iterations=200)
new_image = np.vstack([centroids[i] for i in clusters
]).astype(np.uint8).reshape(image.shape)
Image.fromarray(new_image).save(
os.path.join(results_dir, '%d-clusters-policemen.jpg' % best_inner))
# Outer criterias' block.
outer_criterias_input = os.path.join(data_dir, 'outer_criterias_input.txt')
data = np.loadtxt(outer_criterias_input, delimiter=' ')
reference, points = data[:, 0], data[:, 1:]
rs, best_rs, fm, best_fm = outer_criteria(points, clusters_range,
reference)
# Draw the results.
fig, ax = plt.subplots(nrows=2, ncols=2)
ax1, ax2, ax3, ax4 = ax.flatten()
def main():
r = 5 # number of random initial
d = 2 # reduced dimension
k = 2 # number of cluster
N = 200 # sample size
part = input("Input part A/B \n")
number = input("Input number 1~6 \n")
if (part == 'A'):
data = sio.loadmat('HW3_Data/dataset1.mat')
Y = data['Y']
if (number == 1):
plt.close()
x1 = Y[0, :]
y1 = Y[1, :]
plt.scatter(x1, y1)
plt.show()
plt.close()
if (number == 2):
plt.close()
x1 = Y[0, :]
y1 = Y[2, :]
plt.scatter(x1, y1)
plt.show()
if (number == 3):
plt.close()
u, y_reduced = pca.pca(Y, 2)
x = y_reduced[0, :]
y = y_reduced[1, :]
plt.scatter(np.asarray(x), np.asarray(y))
plt.show()
if (number == 4):
plt.close()
result = kmeans.k_means(np.matrix(Y), k, r)
x1 = []
y1 = []
x2 = []
y2 = []
U, y_2d = pca.pca(Y, d)
for i in range(N):
if (result[i] == 0):
x1.append(y_2d[0, i])
y1.append(y_2d[1, i])
else:
x2.append(y_2d[0, i])
y2.append(y_2d[1, i])
plt.scatter(x1, y1, color='red')
plt.scatter(x2, y2, color='blue')
plt.show()
if (number == 5):
plt.close()
U, y_2d = pca.pca(Y, d)
result = kmeans.k_means(y_2d, k, r)
x1 = []
y1 = []
x2 = []
y2 = []
for i in range(200):
if (result[i] == 0):
x1.append(y_2d[0, i])
y1.append(y_2d[1, i])
else:
x2.append(y_2d[0, i])
y2.append(y_2d[1, i])
plt.scatter(x1, y1, color='red')
plt.scatter(x2, y2, color='blue')
plt.show()
if (part == 'B'):
data = sio.loadmat('HW3_Data/dataset2.mat')
Y = data['Y']
if (number == 1):
x1 = Y[0, :]
y1 = Y[1, :]
plt.scatter(x1, y1)
plt.show()
if (number == 2):
x1 = Y[0, :]
y1 = Y[2, :]
plt.scatter(x1, y1)
plt.show()
if (number == 3):
u, y_reduced = pca.pca(Y, 2)
x = y_reduced[0, :]
y = y_reduced[1, :]
plt.scatter(np.asarray(x), np.asarray(y))
plt.show()
if (number == 4):
U, y_2d = pca.pca(Y, d)
result = kmeans.k_means(np.matrix(y_2d), k, r)
x1 = []
y1 = []
x2 = []
y2 = []
for i in range(N):
if (result[i] == 0):
x1.append(y_2d[0, i])
y1.append(y_2d[1, i])
else:
x2.append(y_2d[0, i])
y2.append(y_2d[1, i])
plt.scatter(x1, y1, color='red')
plt.scatter(x2, y2, color='blue')
plt.show()
if (number == 5):
kernel = pca.get_kernel(Y)
u = pca.kernel_pca(kernel, d)
y_reduced = np.matrix(kernel * u)
result = kmeans.k_means(y_reduced.T, k, r)
x1 = []
y1 = []
x2 = []
y2 = []
#U, y_2d = pca.pca(Y, d)
y_2d = y_reduced.T
for i in range(N):
if (result[i] == 0):
x1.append(y_2d[0, i])
y1.append(y_2d[1, i])
else:
x2.append(y_2d[0, i])
y2.append(y_2d[1, i])
plt.scatter(x1, y1, color='red')
plt.scatter(x2, y2, color='blue')
plt.show()
if (number == 6):
W = np.matrix(spectral.get_w_matrix(Y, 5, 1))
result = spectral.spectral_cluster(W, 2)
x1 = []
y1 = []
x2 = []
y2 = []
U, x = pca.pca(Y, 2)
y_2d = x
for i in range(200):
if (result[i] == 0):
x1.append(y_2d[0, i])
y1.append(y_2d[1, i])
else:
x2.append(y_2d[0, i])
y2.append(y_2d[1, i])
plt.scatter(x1, y1, color='red')
plt.scatter(x2, y2, color='blue')
plt.show()
else:
idf[j] = math.log(idf[j])
for i in range(len(urls)):
for j in range(len(taglist)):
tf[i][j] *= idf[j]
t = np.array(tf)
t = t.transpose()
mat,trans = pca_min.pca(t,3)
plot([trans[:9],trans[9:17],trans[17:]])
#print mat
#print trans
##m,r,a = kmeans.k_means(np.array(tf),2)
##print r,a
##c0 = np.array([[trans[i][0],trans[i][1],trans[i][2]] for i in range(len(trans)) if a[i]==0])
##c1 = np.array([[trans[i][0],trans[i][1],trans[i][2]] for i in range(len(trans)) if a[i]==1])
##plot([c0,c1])
m,r,a = kmeans.k_means(np.array(trans),3)
for i in range(10):
m1,r1,a1 = kmeans.k_means(np.array(trans),3)
print r1,a1
if(r1<r):
m = m1
r = r1
a = a1
print r,a
c0 = np.array([[trans[i][0],trans[i][1],trans[i][2]] for i in range(len(trans)) if a[i]==0])
c1 = np.array([[trans[i][0],trans[i][1],trans[i][2]] for i in range(len(trans)) if a[i]==1])
c2 = np.array([[trans[i][0],trans[i][1],trans[i][2]] for i in range(len(trans)) if a[i]==2])
plot([c0,c1,c2])
from scipy import cluster
from sklearn.datasets import load_iris
import kmeans
from matplotlib import pyplot as plt
from os import mkdir
iris = load_iris()
data = iris.data
names = iris.feature_names
np.random.shuffle(data)
k = 3
t = time.time()
centers, center_steps = kmeans.k_means(data, k=k, distance='e')
print(time.time() - t)
for step in range(len(center_steps)):
print("{0}/{1}".format(step + 1, len(center_steps)))
for combination in itertools.combinations(range(data.shape[1]), 2):
step_centers = np.array(center_steps[step])
assignment, cdist = cluster.vq.vq(data[:, combination],
step_centers[:, combination])
if not os.path.exists(names[combination[0]] + "-"
+ names[combination[1]]):
mkdir(names[combination[0]] + "-" + names[combination[1]])
for i in range(k):
data = np.append(data, [step_centers[i]], axis=0)
k = 2
edgelist = pd.read_csv(path + file[0], delimiter=' ', skiprows=1, header=None)
G = nx.from_pandas_edgelist(edgelist,
source=0,
target=1,
create_using=nx.Graph())
adj_matr = nx.to_pandas_adjacency(G, dtype=np.float64)
laplacian = sparse.csgraph.laplacian(adj_matr.values, normed=True)
eigenvalues, eigenvectors = np.linalg.eig(laplacian)
eigenvectors = eigenvectors.astype(np.float64)
centroids, clusters = k_means(eigenvectors,
k,
random_seed=1,
num_iters=10,
plot=False)
print(centroids)
cut_edges = 0
for i in range(edgelist.shape[0]):
if (clusters[edgelist[0][i]] != clusters[edgelist[1][i]]):
cut_edges += 1
counter = collections.Counter(clusters)
smallest = 100000
for key, value in counter.items():
if (value < smallest):
smallest = value
plt.subplot(121)
plt.axis('off')
plt.imshow(img)
plt.subplot(122)
c = np.loadtxt('data/0.txt')
plt.axis('off')
plt.imshow(c)
plt.show()
pos = np.zeros((img.shape[0], img.shape[1], 2))
for i in range(pos.shape[0]):
for j in range(pos.shape[1]):
pos[i, j, :] = [i, j]
pos = pos.reshape((-1, 2))
data = np.hstack((pos, np.reshape(img, (-1, 3))))
# k_means
res = k_means(data, 3, iter_times=20, dist_func=dist)
tag = res[:, -1]
tag = np.reshape(tag, c.shape)
plt.figure(2)
plt.imshow(tag)
plt.axis('off')
plt.show()
# gmm
res = gmm(data, 3, iter_times=20)
tag = res[:, -1]
tag = np.reshape(tag, c.shape)
plt.figure(3)
plt.imshow(tag)
plt.axis('off')
plt.show()
# dbscan
train_data = data[[x_axis, y_axis]].values.reshape((n_data, 2))
#Normalizing data columns so everything exists between -1 and 1
train_data_mean = np.mean(np.abs(train_data), axis=0)
for j in range(train_data.shape[1]):
train_data[:, j] -= train_data_mean[j]
train_data_scale = np.max(train_data, axis=0)
for j in range(train_data.shape[1]):
train_data[:, j] /= train_data_scale[j]
# Enter k_means inputs.
clusters = 1 # Number of clusters into which we want to split our training dataset.
iterations = 50 # maximum number of training iterations.
# Init k_means instance.
k_means = k_means(train_data, clusters)
# Train k_means instance.
#(centroids, nearest_centroid) = k_means.train(iterations)
(centroids, nearest_centroid) = k_means.kmeansOpt()
# Denormalizing column vectors
for j in range(train_data.shape[1]):
train_data[:, j] *= train_data_scale[j]
centroids[:, j] *= train_data_scale[j]
for j in range(train_data.shape[1]):
train_data[:, j] += train_data_mean[j]
centroids[:, j] += train_data_mean[j]
# Plot actual clusters for reference
plt.subplot(2, 2, 3)
print("took: ", time2 - time1)
print("constructing the cost-matrix")
dist_matrix = get_dist_matrix(x, y, z)
time3 = time.time()
print("took: ", time3 - time2)
#print (dist_matrix)
number_of_clusters = 3
show_3d_data(A,
cmap="viridis",
savename="data_k_" + str(number_of_clusters) + "_b_" +
str(A.shape[0]))
random_start = torch.randint(A.shape[0], [number_of_clusters], device=device)
cluster_assign, data = k_means(A.reshape((A.shape[0], x * y * z)), dist_matrix,
number_of_clusters, x, y, z, regularizer)
print("assignment to clusters: ", cluster_assign)
labels = [
"cluster_start_" + str(random_start[i].item())
for i in range(0, data.shape[0])
]
# time_loc_start = time.time()
# wasbary_1 = compute_wasbary(A, verbose=True, method='pot',numItermax=maxiter, reg= reg)
# time_loc_end = time.time()
# print ("computation of barycenter with pot took ", time_loc_end-time_loc_start)
#time_loc_start = time.time()
#wasbary_1 = compute_wasbary(A, verbose=True, method='ipot',numItermax=10,reg=reg)
#time_loc_end = time.time()
#print ("computation of barycenter with ipot took ", time_loc_end-time_loc_start)
