def select_RPs(self):
N = self.X.shape[0]
if self.rp_number <= 1: self.rp_number = int(self.rp_number * N)
# fcm = FCM(n_clusters=self.rp_number, random_state=self.random_state, max_iter=50)
fcm = FCM(k=self.rp_number,
random_state=self.random_state,
max_iter=50)
fcm.fit(self.X)
# c = fcm.u.argmax(axis=1)
c = fcm.labels_
# homongenious_clusters
homongenious_clusters = np.where(
np.bincount(
np.unique(np.vstack((c, self.y.argmax(
axis=1))), axis=1)[0, :]) == 1)[0]
# centers = fcm.centers[homongenious_clusters,:]
centers = fcm.cluster_centers_[homongenious_clusters, :]
# if all clusters are heterogenious, use all clusters
if len(centers) == 0:
# centers = fcm.centers
centers = fcm.cluster_centers_
# get most closest samples from centers
rp_id = cdist(centers, self.X).argmin(axis=1)
self.rp_X = self.X[rp_id, :]
self.rp_y = self.y[rp_id, :]
self.D_x = cdist(self.X, self.rp_X)
self.D_y = cdist(self.y, self.rp_y)
def simulate(self):
scales = np.random.gamma(1, 10, self.M)**-1
all_P = np.random.poisson(100*self.N)
self.x = np.zeros((all_P, self.M))
self.assignments = np.zeros((all_P, 2), dtype=int)
for i in range(all_P):
z = list(np.random.multinomial(1, self.beta)).index(1)
x = np.random.multivariate_normal(self.mu[z], self.Sigma[z])
self.x[i] = x
self.assignments[i,1] = z
# cluster x into N groups
fuzzy_kmeans = FuzzyKMeans(k=self.N, m=2)
fuzzy_kmeans.fit(self.x)
self.pi = np.zeros((self.K, self.N))
self.P = np.zeros(self.N)
self.x_hat = np.zeros((self.N, self.K, self.M))
for i in range(all_P):
n = list(np.random.multinomial(1, fuzzy_kmeans.fuzzy_labels_[i])).index(1)
self.assignments[i,0] = n
self.x_hat[n][self.assignments[i,1]] += self.x[i]
self.P[n] += 1
self.pi[self.assignments[i,1],n] += 1
# normalize pi
self.pi = (self.pi.T / self.pi.sum(1).T).T
self.y = self.draw(self.x_hat.sum(1) / self.P[:,np.newaxis], scales)
self.x = self.itransform(self.x)
def __init__(self,
n_clusters=8,
m=1,
init='k-means++',
n_init=10,
max_iter=300,
tol=1e-4,
precompute_distances='auto',
verbose=0,
random_state=None,
copy_x=True,
n_jobs=None,
algorithm='auto'):
FuzzyKMeans.__init__(self,
k=n_clusters,
m=m,
max_iter=max_iter,
random_state=random_state,
tol=tol)
KMeans.__init__(self,
n_clusters=n_clusters,
init=init,
n_init=n_init,
max_iter=max_iter,
tol=tol,
precompute_distances=precompute_distances,
verbose=verbose,
random_state=random_state,
copy_x=copy_x,
n_jobs=n_jobs,
algorithm=algorithm)
def __init__(self, data, c):
self.data = data
self.n_clusters = c
# Fuzzy index set to 2 in order to be aligned to the paper
if c is None: self.fuzzy_means = FuzzyKMeans(m=2)
else:
self.fuzzy_means = FuzzyKMeans(self.n_clusters, m=2)
def plot_compare_mu(self, feature1, feature2):
## Run Other Models and plot results
# Fuzzy K-Means
mdl = FuzzyKMeans(k=self.settings.K)
mdl.fit(self.data.obs[:])
# K-Means
kmeans = KMeans(k=self.settings.K)
kmeans.fit(self.data.obs[:])
# Gaussian Mixture Model
gmm = GaussianMixture(n_components=self.settings.K)
gmm.fit(self.data.obs[:])
# Plot Results
plt.clf()
plt.plot(self.data.obs[:, feature1],
self.data.obs[:, feature2],
'ko',
markersize=2,
label='Data') # Observed Data
plt.plot(self.inference_mu[:, feature1,
len(self.iterations) - 1],
self.inference_mu[:, feature2,
len(self.iterations) - 1],
'b*',
markersize=6,
label='NDMs Inferred') # Inferred Centers
plt.plot(self.data.mu[:, feature1],
self.data.mu[:, feature2],
'ro--',
markersize=6,
label='True') # True Centers
plt.plot(self.data.mu[[-1, 0], feature1], self.data.mu[[-1, 0],
feature2],
'r--') # Complete True Polygon
plt.plot(mdl.cluster_centers_[:, feature1],
mdl.cluster_centers_[:, feature2],
'ks',
markersize=6,
label='Fuzzy K-Means')
plt.plot(kmeans.cluster_centers_[:, feature1],
kmeans.cluster_centers_[:, feature2],
'kv',
markersize=6,
label='K-Means')
plt.plot(gmm.means_[:, feature1],
gmm.means_[:, feature2],
'kh',
markersize=4,
label='GMM')
plt.legend()
plt.savefig(os.path.join(self.settings.outdir, 'Comparison_Graph.png'))
class FuzzyMeansAlgorithm:
def __init__(self, data, c):
self.data = data
self.n_clusters = c
# Fuzzy index set to 2 in order to be aligned to the paper
if c is None: self.fuzzy_means = FuzzyKMeans(m=2)
else:
self.fuzzy_means = FuzzyKMeans(self.n_clusters, m=2)
def model_name(self):
title = "Fuzzy-Means"
return title
def clusterize(self):
print(" * Clustering data with {}...".format(self.model_name()))
return self.fuzzy_means.fit(self.data)
def get_model(self):
return self.fuzzy_means
if opts.minibatch:
km = MiniBatchKMeans(n_clusters=true_k,
init='k-means++',
n_init=1,
init_size=1000,
batch_size=1000,
verbose=opts.verbose)
else:
km = KMeans(n_clusters=true_k,
init='k-means++',
max_iter=100,
n_init=1,
verbose=opts.verbose)
#############
fuzzy_kmeans = FuzzyKMeans(k=true_k, m=1.01) #, max_iter=1000)
X1 = np.asarray(X.todense())
fuzzy_kmeans.fit(X1)
def print_metrics(km):
print("Homogeneity: %0.3f" % metrics.homogeneity_score(labels, km.labels_))
print("Completeness: %0.3f" %
metrics.completeness_score(labels, km.labels_))
print("V-measure: %0.3f" % metrics.v_measure_score(labels, km.labels_))
print("Adjusted Rand-Index: %.3f" %
metrics.adjusted_rand_score(labels, km.labels_))
print("Silhouette Coefficient: %0.3f" %
metrics.silhouette_score(X, km.labels_, sample_size=1000))
def get_FuzzyKmeans(dataframe,K):
mdl = FuzzyKMeans(k=K)
mdl.fit(dataframe)
return mdl.labels_
"""
import numpy as np
from sklearn_extensions.fuzzy_kmeans import KMedians, FuzzyKMeans, KMeans
from sklearn.datasets.samples_generator import make_blobs
np.random.seed(0)
batch_size = 45
centers = [[1, 1], [-1, -1], [1, -1]]
n_clusters = len(centers)
X, labels_true = make_blobs(n_samples=1200, centers=centers, cluster_std=0.3)
kmeans = KMeans(k=3)
kmeans.fit(X)
kmedians = KMedians(k=3)
kmedians.fit(X)
fuzzy_kmeans = FuzzyKMeans(k=3, m=2)
fuzzy_kmeans.fit(X)
print('KMEANS')
print(kmeans.cluster_centers_)
print('KMEDIANS')
print(kmedians.cluster_centers_)
print('FUZZY_KMEANS')
print(fuzzy_kmeans.cluster_centers_)
sentimentScore = []
sentimentScore.extend(pos_tweets.flatten())
sentimentScore.extend(neg_tweets.flatten())
shuffle(sentimentScore)
sentimentScore = np.array(sentimentScore)
alldata = sentimentScore.reshape(-1, 1)
#### We are using FuzzyKMeans from skfuzzy because fuzzyCMeans of skfuzzy needs 2 dimensional data
#### and we have one-dimensional data.
####http://pythonhosted.org/scikit-fuzzy/auto_examples/plot_cmeans.html
#### Fuzzy KMeans
#print('FUZZY_KMEANS')
fuzzy_kmeans = FuzzyKMeans(k=6, m=2)
fuzzy_kmeans.fit(alldata)
print np.sort((np.array(fuzzy_kmeans.cluster_centers_).flatten()))
# print(fuzzy_kmeans.cluster_centers_)
# print (kmeans.labels_)
# print (kmeans.cluster_centers_)
# print (kmeans.n_clusters)
### contains predicted centroids for all 6 emoticons, will be printed in fuzzy_output.txt
my_cluster_centers = np.array(fuzzy_kmeans.cluster_centers_).flatten().tolist()
my_cluster_centers = sorted(my_cluster_centers)
negative_centers = np.array(my_cluster_centers[:3]).flatten()
positive_centers = np.array(my_cluster_centers[3:]).flatten()
#print positive_centers, negative_centers
## cluster all data into 6 groups depending on which centroid its closest to
data1 = np.random.multivariate_normal(miu1, cov1, size=n) data2 = np.random.multivariate_normal(miu2, cov2, size=n) data3 = np.random.multivariate_normal(miu3, cov3, size=n) data = np.concatenate((data1, data2, data3)) np.random.shuffle(data) print(data.shape) plt.xlim(-40, 50) plt.ylim(-40, 50) start_time = time.time() fuzzy = FuzzyKMeans(k = 3) fuzzy.fit(data) centers = fuzzy.cluster_centers_ end_time = time.time() print((end_time - start_time)) plt.plot(data[:, 0], data[:, 1], marker="o", linestyle="") plt.plot(centers[:, 0], centers[:, 1], marker="o", linestyle="",color='red') plt.show()
# create kmeans object
kmeans = KMeans(k=5)
# fit kmeans object to data
kmeans.fit(points)
# print location of clusters learned by kmeans object
print('K-means cluster centroids:\n', kmeans.cluster_centers_)
###########################################################################################################################
# create kmedians object
kmedians = KMedians(k=5)
# fit kmeans object to data
kmedians.fit(points)
# print location of clusters learned by kmedians object
print('K-medians cluster centroids:\n', kmedians.cluster_centers_)
###########################################################################################################################
# create kmeansf (fuzzy) object
kmeansf = FuzzyKMeans(k=5)
# fit kmeans object to data
kmeansf.fit(points)
# print location of clusters learned by kmeans fuzzy object
print('K-means fuzzy cluster centroids:\n', kmeansf.cluster_centers_)
###########################################################################################################################
fig = plt.figure()
colors = ['#4EACC5', '#FF9C34', '#4E9A06', '#8B0000', '#FFEF00']
objects = (kmeans, kmedians, kmeansf)
X = points
for i, obj in enumerate(objects):
ax = fig.add_subplot(1, len(objects), i + 1)
for k, col in zip(range(obj.k), colors):
clustering_df = df.drop(columns=["Unnamed: 0", "Player", "final_team", "Pos"])
# we keep the interesting value
df_fcm = df[[
'Player', 'TRB', 'PTS', 'AST', "DWS", '3PA', "OWS", "USG%", "Height"
]]
# we keep the players name for later
players_name = df_fcm["Player"]
# we remove the player column for the computation
df_fcm = df_fcm.loc[:, (df_fcm.columns != "Player")]
# Computation
nb_cluster_fuzzy = 35
fuzzy_kmeans = FuzzyKMeans(k=nb_cluster_fuzzy, m=1.1)
fuzzy_kmeans.fit(df_fcm)
fuzzy_clusters = pd.DataFrame(fuzzy_kmeans.fuzzy_labels_)
# we add the players name back
fuzzy_clusters = pd.concat([players_name, fuzzy_clusters], axis=1)
nb_max_players_per_cluster_fcm = 3
final_clusters = pd.DataFrame()
for i in range(nb_cluster_fuzzy):
# lets keep the coresponding col of membership degree
sets = fuzzy_clusters[["Player", i]]
# lets sort
from sklearn_extensions.fuzzy_kmeans import FuzzyKMeans mdl = FuzzyKMeans(k=7, m=2) mdl.fit(np.concatenate([x_km, x_km_test], axis=0)) train_fcm_pred = list(mdl.labels_)[0:x_km.shape[0]] test_fcm_pred = list(mdl.labels_)[x_km.shape[0]:]
from sklearn_extensions.fuzzy_kmeans import FuzzyKMeans
###Fonction utilisée pour ouvrir un ComputedCluster et reconstuire un dataframe avec les informations sur les joueurs (équipe,PER, etc..)
def import_cluster(epsilon=0.9, minPoints=2, NoiseProp=0.59):
pathCluster = "../ComputedClusters/" + "epsilon_" + str(
epsilon) + "_MinPoints_" + str(minPoints) + "_NoiseProp_" + str(
NoiseProp) + ".csv"
df = pd.read_csv(pathCluster, ';')
df.rename(columns={'0': 'Player', '1': 'Cluster'}, inplace=True)
df_PS = pd.read_csv('../csv/players_stats.csv', ',')
df_join = df.join(df_PS.set_index('Player'), on='Player')
return (df_join)
df = import_cluster()
df_noise = df[df.Cluster == -1] #On garde uniquement le bruit
df_noise = df_noise.reset_index(drop=True)
player_names = pd.DataFrame(df_noise["Player"])
#player_names = player_names.loc[~player_names.index.duplicated(keep='first')]
df = df_noise[['TRB', 'PTS', 'AST', 'DWS', 'TS%', "3PA", "OWS", "USG%"]]
fuzzy_kmeans = FuzzyKMeans(k=5, m=2)
fuzzy_kmeans.fit(df)
clusters = pd.DataFrame(fuzzy_kmeans.fuzzy_labels_)
res = pd.concat([player_names, clusters], axis=1)
print('FUZZY_KMEANS')
print(fuzzy_kmeans.cluster_centers_)
fout.create_dataset("pf/z", data=pf_z)
fout.create_dataset("pf/components", data=pf_components.T)
# K-Means
from sklearn.cluster import KMeans
kmeans = KMeans(n_clusters=args.K)
k_z = kmeans.fit_predict(obs)
k_components = kmeans.cluster_centers_
fout.create_dataset("kmeans/z", data=k_z)
fout.create_dataset("kmeans/components", data=k_components)
# Fuzzy K-Means
from sklearn_extensions.fuzzy_kmeans import FuzzyKMeans
fkmeans = FuzzyKMeans(k=args.K, m=2)
fkmeans.fit(obs)
fk_z = fkmeans.fuzzy_labels_
fk_components = fkmeans.cluster_centers_
fout.create_dataset("fuzzy_kmeans/z", data=fk_z)
fout.create_dataset("fuzzy_kmeans/components", data=fk_components)
# GMM
from sklearn.mixture import GaussianMixture
gmm = GaussianMixture(n_components=args.K)
gmm.fit(obs)
gmm_z = gmm.predict_proba(obs)
fout.create_dataset("gmm/z", data=gmm_z)
fout.create_dataset("gmm/components", data=gmm.means_)
fout.create_dataset("gmm/covariances", data=gmm.covariances_)
def create_fuzzy(self, number_of_clusters, data):
fuzzy_kmeans = FuzzyKMeans(k=number_of_clusters, m=2, max_iter=100)
fuzzy_kmeans.fit(data)
return fuzzy_kmeans
def fuzzy_means(dados):
fuzzy_kmeans = FuzzyKMeans(k=2, m=1000)
fuzzy_kmeans.fit(dados)
return fuzzy_kmeans
def sklearn(self):
all_data = list()
for x in range(0, len(data_handler_1.param1)):
all_data.append(
list([data_handler_1.param1[x], data_handler_1.param2[x]]))
data = np.array(all_data)
datapd = pd.DataFrame(data)
scores = []
# for k in range(2, 10):
k = 4
fuzzy_kmeans = FuzzyKMeans(k=k, m=4, max_iter=300)
fuzzy_kmeans.fit(datapd)
datapd['labels'] = pd.Series(fuzzy_kmeans.labels_)
score = ss(datapd[[0, 1]], labels=datapd['labels'])
scores.append(score)
#
# datapd.plot.scatter(x=0, y=1, c='labels', colormap='viridis')
# plt.xlabel("Param 1")
# plt.ylabel("Param2")
# plt.title(f'K = {k}, Silhouette score = {score}')
for center in fuzzy_kmeans.cluster_centers_:
plt.plot(center[0], center[1], 'ro')
# for datapd['labels']
group_data = datapd.groupby(pd.Series(fuzzy_kmeans.labels_),
group_keys=datapd['labels'].unique())
# print(datapd)
# print(group_data[1].get_group(1))
# n = [i for i in range(2, 10)]
# plt.figure()
# plt.plot(n, scores)
# plt.xlabel("K")
# plt.ylabel("Silhouette score")
# plt.show()
second_data = DataHandler()
second_data.open_file("files\\sdmt3.txt")
second = list()
for x in range(0, len(data_handler_1.param1)):
second.append(
list([data_handler_1.param1[x], data_handler_1.param2[x]]))
sec_data = np.array(second)
sec_datapd = pd.DataFrame(sec_data)
scores = []
# print(second_data.param1)
max_values = []
min_values = []
for group_label in datapd['labels'].unique():
max_param1 = np.max(group_data[0].get_group(group_label))
max_param2 = np.max(group_data[1].get_group(group_label))
min_param1 = np.min(group_data[0].get_group(group_label))
min_param2 = np.min(group_data[1].get_group(group_label))
tuple_max = (max_param1, max_param2)
tuple_min = (min_param1, min_param2)
max_values.append(tuple_max)
min_values.append(tuple_min)
# print(max_values)
# print(min_values)
# print(sec_datapd)
sec_datapd['labels'] = pd.Series(fuzzy_kmeans.labels_)
sec_datapd.set_value(2, 'labels', 50)
print(sec_datapd)
for i in range(0, 40):
for j in range(len(max_values)):
x, y, z = sec_datapd.values[i]
max1, max2 = max_values[j]
min1, min2 = min_values[j]
# print(x,y)
if x > min1 and x < max1 and y > min2 and y < max2:
sec_datapd.set_value(i, 'labels', j)
sec_score = ss(sec_datapd[[0, 1]], labels=sec_datapd['labels'])
print(sec_datapd)
sec_datapd.plot.scatter(x=0, y=1, c='labels', colormap='viridis')
plt.xlabel("Param 1")
plt.ylabel("Param2")
plt.title(f'K = {k}, Silhouette score = {sec_score}')
plt.show()