def cluster(train, val, type, number_of_clusters, plot_folder, classes):
# todo this should be a class
if type == "spectral_clustering":
clustering_model = SpectralClustering(n_clusters=number_of_clusters,
assign_labels="discretize",
random_state=0).fit(
train["data"])
elif type == "kmeans":
clustering_model = KMeans(n_clusters=number_of_clusters,
random_state=0).fit(train["data"])
else:
raise NotImplementedError
# compute metrics
accuracies = {}
random_array = np.random.randint(9, size=train["labels"].shape)
centroids = find_centroids(number_of_clusters, train,
clustering_model.labels_)
test_classifications = cluster_test(val, centroids)
visualize_clustering(train, clustering_model.labels_, type + "_training",
plot_folder, number_of_clusters, centroids)
visualize_clustering(val, np.asarray(test_classifications),
type + "_validation", plot_folder, number_of_clusters,
centroids)
accuracies["random_score"] = homogeneity_score(train["labels"],
random_array)
accuracies["v_measure_score"] = v_measure_score(train["labels"],
clustering_model.labels_)
accuracies["homogeneity_score"] = homogeneity_score(
train["labels"], clustering_model.labels_)
accuracies["completeness_score"] = completeness_score(
train["labels"], clustering_model.labels_)
accuracies["silhouette_score"] = silhouette_score(train["data"],
clustering_model.labels_)
accuracies["purity_score"], accuracies[
"contingency_matrix"] = purity_score(train["labels"],
clustering_model.labels_)
accuracies["v_measure_score_test"] = v_measure_score(
val["labels"], test_classifications)
accuracies["homogeneity_score_test"] = homogeneity_score(
val["labels"], test_classifications)
accuracies["completeness_score_test"] = completeness_score(
val["labels"], test_classifications)
accuracies["silhouette_score_test"] = silhouette_score(
val["data"], test_classifications)
accuracies["purity_score_test"], accuracies[
"contingency_matrix_test"] = purity_score(val["labels"],
test_classifications)
return accuracies
def get_clustering_metrics(train_data,
cluster_labels,
ground_truth_labels=None):
clustering_metric_dict = dict({})
clustering_metric_dict['silhouette_score'] = silhouette_score(
train_data, cluster_labels, random_state=42)
clustering_metric_dict[
'calinski_harabasz_score'] = calinski_harabasz_score(
train_data, cluster_labels)
clustering_metric_dict['davies_bouldin_score'] = davies_bouldin_score(
train_data, cluster_labels)
if ground_truth_labels is not None:
clustering_metric_dict['v_measure_score'] = v_measure_score(
ground_truth_labels, cluster_labels)
clustering_metric_dict[
'fowlkes_mallows_score'] = fowlkes_mallows_score(
ground_truth_labels, cluster_labels)
clustering_metric_dict['homogeneity_score'] = homogeneity_score(
ground_truth_labels, cluster_labels)
clustering_metric_dict[
'normalized_mutual_info_score'] = normalized_mutual_info_score(
ground_truth_labels, cluster_labels)
clustering_metric_dict['adjusted_rand_score'] = adjusted_rand_score(
ground_truth_labels, cluster_labels)
clustering_metric_dict['completeness_score'] = completeness_score(
ground_truth_labels, cluster_labels)
return clustering_metric_dict
def show_result(self, prediction, msg):
new_line(50)
print(msg)
new_line(50)
real = self.train_labels
print "Confusion Matrix: "
print str(confusion_matrix(real, prediction))
homo_score = homogeneity_score(real, prediction)
complete_score = completeness_score(real, prediction)
v_score = v_measure_score(real, prediction)
rand_score = adjusted_rand_score(real, prediction)
mutual_info = adjusted_mutual_info_score(real, prediction)
print("Homogeneity Score: %0.3f" % homo_score)
print("Completeness Score: %0.3f" % complete_score)
print("V-measure: %0.3f" % v_score)
print("Adjusted Rand Score: %0.3f" % rand_score)
print("Adjusted Mutual Info Score: %0.3f\n" % mutual_info)
return {
'Homogeneity': homo_score,
'Completeness': complete_score,
'V-measure': v_score,
'RAND': rand_score,
'Mutual': mutual_info
}
def compute_V_measure(clusters, classes):
class_list, cluster_list = [], []
# not_found_id = 1000000
clustered_but_unaligned = 0
for read in clusters:
if read in classes:
class_list.append(classes[read])
cluster_list.append(clusters[read])
else:
# print("Read was clustered but unaligned:", read)
clustered_but_unaligned += 1
# added the unprocessed reads to the measure
not_clustered = set(classes.keys()) - set(clusters.keys())
highest_cluster_id = max(clusters.values())
highest_cluster_id += 1
for read in not_clustered:
class_list.append(classes[read])
cluster_list.append(highest_cluster_id)
highest_cluster_id += 1
v_score = v_measure_score(class_list, cluster_list)
compl_score = completeness_score(class_list, cluster_list)
homog_score = homogeneity_score(class_list, cluster_list)
ari = adjusted_rand_score(class_list, cluster_list)
print("Not included in clustering but aligned:", len(not_clustered))
print("V:", v_score, "Completeness:", compl_score, "Homogeneity:",
homog_score)
print(
"Nr reads clustered but unaligned (i.e., no class and excluded from V-measure): ",
clustered_but_unaligned)
return v_score, compl_score, homog_score, clustered_but_unaligned, ari
def calc_homogenity_comp_vmeas_training(df, y_train):
# user_input_df['predict'] = y1
# confidence_counter = -1
# for c in candidates:
# confidence_counter += 1
# adj = c.getSource()
# noun = c.getTarget()
# candidate_df = user_input_df.loc[(user_input_df['adj'] == adj) & (user_input_df['noun'] == noun)]
# print(candidate_df["adj"][confidence_counter])
# print(candidate_df["noun"][confidence_counter])
# if candidate_df["class"][confidence_counter] != 2:
# truelabels.append(candidate_df["class"][confidence_counter])
# predictlabels.append(candidate_df["predict"][confidence_counter])
# print("truelables:",truelabels)
# print("predictlabels:",predictlabels)
# homogenity_scr = homogeneity_score(truelabels,predictlabels)
# vmeasure_scr = v_measure_score(truelabels,predictlabels)
# completness_scr =completeness_score(truelabels,predictlabels)
# print("homogenity_scr={},vmeasure_scr={},completness_scr={}".format(homogenity_scr,vmeasure_scr,completness_scr))
truelabels = df['class']
predictlabels = y_train
homogenity_scr = homogeneity_score(truelabels, predictlabels)
vmeasure_scr = v_measure_score(truelabels, predictlabels)
completness_scr = completeness_score(truelabels, predictlabels)
print("truelables:", truelabels)
print("predictlabels:", predictlabels)
print("homogenity_scr={},vmeasure_scr={},completness_scr={}".format(
homogenity_scr, vmeasure_scr, completness_scr))
def evaluate_bins(self):
self.labels = []
newcolors = []
for bin in self.bins:
for b in bin:
self.labels.append(self.bins.index(bin))
if self.colors != None:
newcolors.append(self.colors[b])
self.colors = newcolors
labels = numpy.array(self.labels)
colors = numpy.array(self.colors)
points = []
for bin in self.bins:
for b in bin:
start_lat = self.data[b]['trip_start_location'][1]
start_lon = self.data[b]['trip_start_location'][0]
end_lat = self.data[b]['trip_end_location'][1]
end_lon = self.data[b]['trip_end_location'][0]
path = [start_lat, start_lon, end_lat, end_lon]
points.append(path)
if self.colors != None:
a = metrics.silhouette_score(numpy.array(points), labels)
b = homogeneity_score(colors, labels)
c = completeness_score(colors, labels)
print 'number of bins is ' + str(len(self.bins))
print 'silhouette score is ' + str(a)
print 'homogeneity is ' + str(b)
print 'completeness is ' + str(c)
print 'accuracy is ' + str(((a+1)/2.0 + b + c)/3.0)
def compareAB(A, B, X):
#measures the similarity of the two assignments, ignoring permutations and with chance normalization
ars = metrics.adjusted_rand_score(A, B)
ars_str = '%17.3f' % ars
# each cluster contains only members of a single class
hs = homogeneity_score(A, B)
hs_str = '%17.3f' % hs
#all members of a given class are assigned to the same cluster
cs = completeness_score(A, B)
cs_str = '%17.3f' % cs
vms = metrics.v_measure_score(A, B)
vms_str = '%17.3f' % vms
# geometric mean of the pairwise precision and recall
fowlkes_mallows_score = metrics.fowlkes_mallows_score(A, B)
fms_str = '%17.3f' % fowlkes_mallows_score
sc = metrics.silhouette_score(X, B, metric='euclidean')
sc_str = '%17.3f' % sc
my_str = ars_str + "&" + hs_str + "&" + cs_str + "&" + vms_str + "&" + fms_str + "&" + sc_str
return my_str
def main():
m = 2000 # number of points
n = 50 # Number of dimensions
k = 30 # Number of latent clusters
np.random.seed(3)
X, z_true = draw_points(m, n, k=k)
show_points(X, z_true, title="True")
S = fully_connected_similarity(X)
# Unnormalized spectral clustering
# A = spectral_clustering(S, k)
# Normalized spectral clustering according to Shi and Malik (2000)
# A = spectral_clustering(S, k, normalization=LaplacianNorm.symmetric, generalized_eigenproblem=True)
# Normalized spectral clustering according to Ng, Jordan, and Weiss (2002)
A = spectral_clustering(S,
k,
normalization=LaplacianNorm.symmetric,
norm_rows=True)
show_points(X, A, title="Spectral Clustering")
complete_score = completeness_score(z_true, A)
homog_score = homogeneity_score(z_true, A)
print("Completeness score: %s" % complete_score)
print("Homogeneity score: %s" % homog_score)
def print_scores(labels, predicted, svd):
print "Homogeneity: " + str(homogeneity_score(labels, predicted))
print "completeness: " + str(completeness_score(labels, predicted))
print "V-measure: " + str(v_measure_score(labels, predicted))
print "RAND score: " + str(adjusted_rand_score(labels, predicted))
print "Mutual Info: " + str(adjusted_mutual_info_score(labels, predicted))
ret = []
ret.append(homogeneity_score(labels, predicted))
ret.append(completeness_score(labels, predicted))
ret.append(v_measure_score(labels, predicted))
ret.append(adjusted_rand_score(labels, predicted))
ret.append(adjusted_mutual_info_score(labels, predicted))
if svd:
svd_all.append(ret)
else:
nmf_all.append(ret)
return homogeneity_score(labels, predicted)
def calculate_scores(self): x, c, labels = self.x, self.c, self.labels self.v_measure = v_measure_score(c, labels) self.complete = completeness_score(c, labels) self.adjusted_mutual = adjusted_mutual_info_score(c, labels) self.adjusted_rand = adjusted_rand_score(c, labels) self.silhouette = silhouette_score(x, c) self.purity, self.partial_purity = self.__purity__()
def calculate_scores(self):
x, c, labels = self.x, self.c, self.labels
self.v_measure = v_measure_score(c, labels)
self.complete = completeness_score(c, labels)
self.adjusted_mutual = adjusted_mutual_info_score(c, labels)
self.adjusted_rand = adjusted_rand_score(c, labels)
self.silhouette = silhouette_score(x, c)
self.purity, self.partial_purity = self.__purity__()
def five_measure_scores(label_true, label_pred):
print("Homogeneity_score = %f" % homogeneity_score(label_true, label_pred))
print("Completeness_score = %f" %
completeness_score(label_true, label_pred))
print("Adjusted_rand_score = %f" %
adjusted_rand_score(label_true, label_pred))
print("V_measure_score = %f" % v_measure_score(label_true, label_pred))
print("Adjusted_mutual_info_score = %f" %
adjusted_mutual_info_score(label_true, label_pred))
def v_measure(cluster_labels, true_labels):
h_score = homogeneity_score(true_labels, cluster_labels)
c_score = completeness_score(true_labels, cluster_labels)
v_score = v_measure_score(true_labels, cluster_labels)
print("Homogeneity Score: %.6f" % h_score)
print("Completeness Score: %.6f" % c_score)
print("V Measure Score: %.6f" % v_score)
return h_score, c_score, v_score
def print_scores(labels, predicted):
print "Contingency: "
print str(confusion_matrix(labels, predicted))
ret = []
ret.append(homogeneity_score(labels, predicted))
ret.append(completeness_score(labels, predicted))
ret.append(v_measure_score(labels, predicted))
ret.append(adjusted_rand_score(labels, predicted))
ret.append(adjusted_mutual_info_score(labels, predicted))
print "Homogeneity: " + str(homogeneity_score(labels, predicted))
print "completeness: " + str(completeness_score(labels, predicted))
print "V-measure: " + str(v_measure_score(labels, predicted))
print "RAND score: " + str(adjusted_rand_score(labels, predicted))
print "Mutual Info: " + str(adjusted_mutual_info_score(labels, predicted))
return ret
def get_landmarking(dataset_name, df):
start = time.time()
record = {'dataset': dataset_name.split('.')[0]}
results = []
n_samples = int(len(df)*0.1) if len(df) > 400 else min(df.shape[0], 40)
data = df.sample(n=n_samples, replace=False)
labels = get_dbscan(data)
k = len(np.unique(labels))
labels2 = get_Kmeans(data, k, 40)
full_tree = DecisionTreeClassifier()
full_tree.fit(data, labels)
worst_attr = np.argmin(full_tree.feature_importances_)
X_train, X_test, y_train, y_test = train_test_split(data, labels, test_size=0.3)
best_stump = DecisionTreeClassifier(max_depth=1)
random_stump = DecisionTreeClassifier(splitter="random", max_depth=1)
worst_stump = DecisionTreeClassifier(max_depth=1)
elite_knn = KNeighborsClassifier(n_neighbors=1)
one_knn = KNeighborsClassifier(n_neighbors=1,
algorithm="auto",
weights="uniform",
p=2,
metric="minkowski")
nb = GaussianNB()
lda = LinearDiscriminantAnalysis()
best_stump.fit(X_train, y_train)
random_stump.fit(X_train, y_train)
worst_stump.fit(X_train.iloc[:, worst_attr].values.reshape(-1, 1), y_train)
elite_knn.fit(X_train, y_train)
one_knn.fit(X_train, y_train)
# lda.fit(X_train, y_train)
nb.fit(X_train, y_train)
record['LM1'] = np.log2(df.shape[0])
record['LM2'] = np.log2(df.shape[1])
record['LM3'] = accuracy_score(best_stump.predict(X_test), y_test)
# record['LM4'] = f1_score(best_stump.predict(X_test), y_test, average='weighted')
record['LM5'] = accuracy_score(random_stump.predict(X_test), y_test)
# record['LM6'] = f1_score(random_stump.predict(X_test), y_test, average='weighted')
# record['LM7'] = model.inertia_
record['LM8'] = accuracy_score(elite_knn.predict(X_test), y_test)
# record['LM9'] = f1_score(elite_knn.predict(X_test), y_test, average='weighted')
# record['LM10'] = accuracy_score(lda.predict(X_test), y_test)
# record['LM11'] = f1_score(lda.predict(X_test), y_test, average='weighted')
record['LM12'] = accuracy_score(nb.predict(X_test), y_test)
# record['LM13'] = f1_score(nb.predict(X_test), y_test, average='weighted')
record['LM14'] = accuracy_score(one_knn.predict(X_test), y_test)
# record['LM15'] = f1_score(one_knn.predict(X_test), y_test, average='weighted')
record['LM16'] = accuracy_score(worst_stump.predict(X_test.iloc[:, worst_attr].values.reshape(-1, 1)), y_test)
# record['LM17'] = f1_score(worst_stump.predict(X_test.iloc[:, worst_attr].values.reshape(-1, 1)), y_test, average='weighted')
record['LM18'] = adjusted_rand_score(labels, labels2)
record['LM19'] = adjusted_mutual_info_score(labels, labels2)
record['LM20'] = completeness_score(labels, labels2)
record['LM21'] = fowlkes_mallows_score(labels, labels2)
end = time.time()
return record, (df.shape[0], df.shape[1], end-start)
def calc_performance_score(self, algo_type: str, predicted, y_train):
homo_score = homogeneity_score(y_train, predicted)
complete_socre = completeness_score(y_train, predicted)
adjusted_mute_info_score = adjusted_mutual_info_score(
y_train, predicted)
print(algo_type + ' homo_score ' + "{:.2f}".format(homo_score))
print(algo_type + ' complete_socre ' + "{:.2f}".format(complete_socre))
print(algo_type + ' adjusted_mute_info_score ' +
"{:.2f}".format(adjusted_mute_info_score))
def get_homogeneity_and_completeness(self, clusters, category):
labels = getattr(self.scb.nodes,
'to_' + category)(range(len(self.scb.nodes)))
keys = dict()
for i, label in enumerate(labels):
if label not in keys:
keys[label] = len(keys)
labels[i] = keys[label]
hs = homogeneity_score(labels, clusters)
cs = completeness_score(labels, clusters)
return {'homogeneity': hs, 'completeness': cs}
def km(clusters, dats, Y):
NMI = defaultdict(dict)
INL = defaultdict(dict)
CMS = defaultdict(dict)
SIL = defaultdict(dict)
for i, dat in enumerate(dats):
for cluster in clusters:
km = KMeans(n_clusters=cluster, random_state=0).fit(dat)
cluster_labels = km.labels_
NMI[i][cluster] = normalized_mutual_info_score(Y, cluster_labels)
INL[i][cluster] = km.inertia_
CMS[i][cluster] = completeness_score(Y, cluster_labels)
SIL[i][cluster] = silhouette_score(dat, cluster_labels)
plt.plot(clusters, NMI[0].values(), 'bx-', color='C0')
plt.plot(clusters, NMI[1].values(), 'bx-', color='C1')
plt.plot(clusters, NMI[2].values(), 'bx-', color='C2')
plt.plot(clusters, NMI[3].values(), 'bx-', color='C3')
plt.plot(clusters, NMI[4].values(), 'bx-', color='C4')
plt.legend(['PCA', 'ICA', 'RP', 'RF', 'Original'])
plt.xlabel('k')
plt.title('Normalized Mutual Information on K-Means')
plt.show()
plt.plot(clusters, INL[0].values(), 'bx-', color='C0')
plt.plot(clusters, INL[1].values(), 'bx-', color='C1')
plt.plot(clusters, INL[2].values(), 'bx-', color='C2')
plt.plot(clusters, INL[3].values(), 'bx-', color='C3')
plt.plot(clusters, INL[4].values(), 'bx-', color='C4')
plt.legend(['PCA', 'ICA', 'RP', 'RF', 'Original'])
plt.xlabel('k')
plt.title('Elbow Method on K-Means')
plt.show()
plt.plot(clusters, CMS[0].values(), 'bx-', color='C0')
plt.plot(clusters, CMS[1].values(), 'bx-', color='C1')
plt.plot(clusters, CMS[2].values(), 'bx-', color='C2')
plt.plot(clusters, CMS[3].values(), 'bx-', color='C3')
plt.plot(clusters, CMS[4].values(), 'bx-', color='C4')
plt.legend(['PCA', 'ICA', 'RP', 'RF', 'Original'])
plt.xlabel('k')
plt.title('Completeness on K-Means')
plt.show()
plt.plot(clusters, SIL[0].values(), 'bx-', color='C0')
plt.plot(clusters, SIL[1].values(), 'bx-', color='C1')
plt.plot(clusters, SIL[2].values(), 'bx-', color='C2')
plt.plot(clusters, SIL[3].values(), 'bx-', color='C3')
plt.plot(clusters, SIL[4].values(), 'bx-', color='C4')
plt.legend(['PCA', 'ICA', 'RP', 'RF', 'Original'])
plt.xlabel('k')
plt.title('Silhouette on K-Means')
plt.show()
def clusterEvaluation(trueY, fittedY):
result = dict()
## NMI denotes normalized mutual information
## ARS denotes adjusted rand score
## HS stands for homogeneity_score, 1 means perfect
## VM represents v_measure_score ranging [0, 1], 1.0 is perfectly complete labeling
## SS represents silhouette_score
result['NMI'] = normalized_mutual_info_score(trueY, fittedY)
result['ARS'] = adjusted_rand_score(trueY, fittedY)
result['HS'] = homogeneity_score(trueY, fittedY)
result['CS'] = completeness_score(trueY, fittedY)
result['VM'] = v_measure_score(trueY, fittedY)
return result
def print_five_measures(target, predicted):
print('homogeneity score:')
print(homogeneity_score(target, predicted))
print('completeness score:')
print(completeness_score(target, predicted))
print('V-measure:')
print(v_measure_score(target, predicted))
print('adjusted rand score:')
print(adjusted_rand_score(target, predicted))
print('adjuted mutual info score:')
print(adjusted_mutual_info_score(target, predicted))
def main():
start = timeit.default_timer()
# Chuyển dữ liệu ảnh sang bitmap và scale
X, Y = load(scale = 50*100 )
# Chạy Birch và gán nhãn
pred_label = process(X, labels_num = 32)
# Tính accuracy với nhãn thật
accuracy(pred_label, Y, labels_num = 32)
print("Completeness :", completeness_score(Y, pred_label) * 100, "%")
print("Homogeneity :", homogeneity_score(Y, pred_label) * 100, "%")
stop = timeit.default_timer()
print("Time :", stop - start, "s")
def evaluate(clusters, typedict):
"""Given the predicted clusters and type dictionary, this function calculates homogeneity, completeness, and V-measure assuming the gold tags are the most frequent tags for each type in the type dict
input:
clusters (dict of int:Cluster): Clusters by id
typedict (dict of str:Word): Word by wordform
return:
(float): homogeneity score
(float): completeness score
(float): V measure"""
# The instructor completed this function in 7 line including the return
golds = []
preds = []
# Your code here
return homogeneity_score(golds, preds), completeness_score(
golds, preds), v_measure_score(golds, preds, beta=2.0)
def __computeKmeansMetrics(self, data, predictedLabels, gsLabels, title, basePath, phase4Results):
metrics = dict()
metrics["davies_bouldin_score"] = clusteringMetrics.davies_bouldin_score(data, predictedLabels)
metrics["adjusted_rand_score"] = clusteringMetrics.adjusted_rand_score(gsLabels, predictedLabels)
metrics["completeness_score"] = clusteringMetrics.completeness_score(gsLabels, predictedLabels)
metrics["purity_score"] = purity_score(gsLabels, predictedLabels)
confusionMatrixMapped = clusteringMappingMetric(predictedLabels, gsLabels)
confusionMatrix = confusion_matrix(gsLabels, predictedLabels)
kdf = pd.DataFrame.from_dict(metrics, orient='index', columns=[title])
phase4Results = phase4Results.join(kdf)
np.savetxt(basePath / f"{title}_kmeans_confusionMapping.csv", confusionMatrixMapped, delimiter=",", fmt='%i')
np.savetxt(basePath / f"{title}_kmeans_confusion.csv", confusionMatrix, delimiter=",", fmt='%i')
return phase4Results
def k_means_clustering(training_data,
target_labels,
title='Contingency Matrix',
n_clusters=20,
random_state=0,
max_iter=1000,
n_init=30):
start = time.time()
km = KMeans(n_clusters=n_clusters,
random_state=random_state,
max_iter=max_iter,
n_init=n_init)
km.fit(training_data)
print("Finished clustering in %f seconds" % (time.time() - start))
cm = contingency_matrix(target_labels, km.labels_)
# reorder to maximize along diagonal
rows, cols = linear_sum_assignment(cm, maximize=True)
new_cm = cm[rows[:, np.newaxis], cols]
print("Show Contingency Matrix:")
plot_contingency_table_20(new_cm, title=title)
print("Report 5 Measures for K-Means Clustering")
homogeneity = homogeneity_score(target_labels, km.labels_)
completeness = completeness_score(target_labels, km.labels_)
v_measure = v_measure_score(target_labels, km.labels_)
adjusted_rand_index = adjusted_rand_score(target_labels, km.labels_)
adjusted_mutual_info = adjusted_mutual_info_score(target_labels,
km.labels_)
print("Homogeneity Score: %f" % homogeneity)
print("Completeness Score: %f" % completeness)
print("V-Measure Score: %f" % v_measure)
print("Adjusted Rand Index: %f" % adjusted_rand_index)
print("Adjusted Mutual Information: %f" % adjusted_mutual_info)
results = {
"homogeneity": homogeneity,
"completeness": completeness,
"v_measure": v_measure,
"adjusted_rand_index": adjusted_rand_index,
"adjusted_mutual_info": adjusted_mutual_info
}
return results, km
def test():
model.eval()
z = model.encode(data.x, data.train_pos_edge_index)
# Cluster embedded values using k-means.
kmeans_input = z.cpu().numpy()
kmeans = KMeans(n_clusters=7, random_state=0).fit(kmeans_input)
pred = kmeans.predict(kmeans_input)
labels = data.y.cpu().numpy()
completeness = completeness_score(labels, pred)
hm = homogeneity_score(labels, pred)
nmi = v_measure_score(labels, pred)
auc, ap = model.test(z, data.test_pos_edge_index, data.test_neg_edge_index)
return auc, ap, completeness, hm, nmi
def evaluate(self):
eval_result_dict = {}
eval_result_dict['ami'] = adjusted_mutual_info_score(
self.data['true_y'], self.data['pred_y'])
eval_result_dict['rand'] = adjusted_rand_score(self.data['true_y'],
self.data['pred_y'])
eval_result_dict['comp'] = completeness_score(self.data['true_y'],
self.data['pred_y'])
eval_result_dict['fow'] = fowlkes_mallows_score(
self.data['true_y'], self.data['pred_y'])
eval_result_dict['hom'] = homogeneity_score(self.data['true_y'],
self.data['pred_y'])
eval_result_dict['nmi'] = normalized_mutual_info_score(
self.data['true_y'], self.data['pred_y'])
eval_result_dict['v_score'] = v_measure_score(self.data['true_y'],
self.data['pred_y'])
return eval_result_dict
def compute_result(self, loss, preds, targets, stage):
# Cluster embedded values using k-means.
kmeans_input = preds.cpu().numpy()
kmeans = KMeans(n_clusters=7, random_state=0).fit(kmeans_input)
pred = kmeans.predict(kmeans_input)
labels = targets.cpu().numpy()
completeness = torch.Tensor([completeness_score(labels, pred)])
hm = torch.Tensor([homogeneity_score(labels, pred)])
nmi = torch.Tensor([v_measure_score(labels, pred)])
# auc, ap = model.test(z, data.test_pos_edge_index, data.test_neg_edge_index)
result = pl.EvalResult(loss)
result.log(f"{stage}_completeness", completeness, prog_bar=True)
result.log(f"{stage}_hm", hm, prog_bar=True)
result.log(f"{stage}_nmi", nmi, prog_bar=True)
return result
def compareAB(A, B):
# measures the similarity of the two assignments, ignoring permutations and with chance normalization
ars = metrics.adjusted_rand_score(A, B)
print("adjusted_rand_score " + str(ars))
# measures the agreement of the two assignments, ignoring permutations
# amis = metrics.adjusted_mutual_info_score(A, B)
# print("adjusted_mutual_info_score " + str(amis))
# each cluster contains only members of a single class
hs = homogeneity_score(A, B)
print("homogeneity_score " + str(hs))
# all members of a given class are assigned to the same cluster
cs = completeness_score(A, B)
print("completeness_score " + str(cs))
vms = metrics.v_measure_score(A, B)
print("v_measure_score " + str(vms))
# geometric mean of the pairwise precision and recall
fowlkes_mallows_score = metrics.fowlkes_mallows_score(A, B)
print("fowlkes_mallows_score " + str(fowlkes_mallows_score))
def main():
start = timeit.default_timer()
# Chuyển dữ liệu ảnh sang bitmap và scale
Test_X, Test_Y, Train_X, Train_Y = load(scale=50 * 100)
print(len(Test_X))
print(len(Test_Y))
print(len(Train_X))
print(len(Train_Y))
# Chạy Kmeans và gán nhãn
pred_label = process(Train_X, Train_Y, Test_X)
# Tính accuracy với nhãn thật
accuracy(pred_label, Test_Y, labels_num=32)
print("Completeness :", completeness_score(Test_Y, pred_label) * 100, "%")
print("Homogeneity :", homogeneity_score(Test_Y, pred_label) * 100, "%")
stop = timeit.default_timer()
print("Time :", stop - start, "s")
def computeMetrics(self, data, trueLabels, predictedLabels):
confusionMatrixes = dict()
metrics = dict()
for algorithmName, labels in predictedLabels.items():
metrics[algorithmName] = dict()
metrics[algorithmName][
"davies_bouldin_score"] = clusteringMetrics.davies_bouldin_score(
data, labels)
metrics[algorithmName][
"adjusted_rand_score"] = clusteringMetrics.adjusted_rand_score(
trueLabels, labels)
metrics[algorithmName][
"completeness_score"] = clusteringMetrics.completeness_score(
trueLabels, labels)
metrics[algorithmName]["purity_score"] = purity_score(
trueLabels, labels)
confusionMatrixes[algorithmName] = clusteringMappingMetric(
labels, trueLabels)
return metrics, confusionMatrixes
def dbscanLibrarySklearn():
X_train, X_test, labels_true, y_test = readData()
# Compute DBSCAN
db = DBSCAN(eps=0.6, min_samples=4).fit(X_train)
core_samples_mask = np.zeros_like(db.labels_, dtype=bool)
core_samples_mask[db.core_sample_indices_] = True
labels = db.labels_
# Number of clusters in labels, ignoring noise if present.
n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
n_noise_ = list(labels).count(-1)
print('Estimated number of clusters: %d' % n_clusters_)
print('Estimated number of noise points: %d' % n_noise_)
print("Homogeneity: %0.3f" % homogeneity_score(labels_true, labels))
print("Completeness: %0.3f" % completeness_score(labels_true, labels))
# Plot result
import matplotlib.pyplot as plt
# Black removed and is used for noise instead.
unique_labels = set(labels)
colors = [plt.cm.Spectral(each)
for each in np.linspace(0, 1, len(unique_labels))]
for k, col in zip(unique_labels, colors):
if k == -1:
# Black used for noise.
col = [0, 0, 0, 1]
class_member_mask = (labels == k)
xy = X_train[class_member_mask & core_samples_mask]
plt.plot(xy[:, 0], xy[:, 1], 'o', markerfacecolor=tuple(col),
markeredgecolor='k', markersize=5)
xy = X_train[class_member_mask & ~core_samples_mask]
plt.plot(xy[:, 0], xy[:, 1], 'o', markerfacecolor=tuple(col),
markeredgecolor='k', markersize=5)
plt.title('Estimated number of clusters: %d' % n_clusters_)
plt.show()
def plot_kmeans_em_v_meansrue(n_clusters):
row_idx = n_clusters - 5
v_measure_array = []
xvalues = all_em_clustering_np_matrix.keys()
for key in xvalues:
kmeans_label = all_kmeans_clustering_np_matrix[key][row_idx]
em_label = all_em_clustering_np_matrix[key][row_idx]
c = completeness_score(kmeans_label, em_label)
h = homogeneity_score(kmeans_label, em_label)
v_measure = 2 * c * h / (c + h)
v_measure_array.append(v_measure)
plt.figure(figsize=(8, 8))
plt.plot(range(len(v_measure_array)), v_measure_array, "o-")
plt.xticks(np.arange(len(v_measure_array)), xvalues)
plt.yticks(np.linspace(0.6, 1.0, 9))
plt.grid(True)
plt.xlabel("\nDatasets",fontsize=14)
plt.ylabel("V-measure score",fontsize=14)
plt.title("Comparison of clustering consistency \nbetween EM and Kmeans on the same dataset",fontsize=14)
plt.show()
from sklearn.metrics.cluster import homogeneity_score print(homogeneity_score([0, 0, 1, 1], [1, 1, 0, 0])) print(homogeneity_score([0, 0, 0, 1, 1, 1], [3, 2, 2, 2, 3, 3])) from sklearn.metrics.cluster import completeness_score print(completeness_score([0, 0, 1, 1], [1, 1, 0, 0])) print(completeness_score([0, 0, 0, 1, 1, 1], [3, 2, 2, 2, 3, 3]))
#X_iso = manifold.Isomap(n_neighbors=5, n_components=2).fit_transform(X_train)
#end = int(round(time.time() * 1000))
#print("--Isomap finished in ", (end-start), "ms--------------")
#print("Done.")
#spectral clustering, fitting and predictions
spectral = cluster.SpectralClustering(n_clusters=10, eigen_solver='arpack', affinity="nearest_neighbors")
#X = spectral.fit(X_iso)
X = spectral.fit(X_spec)
#y_pred = spectral.fit_predict(X_iso)
y_pred = spectral.fit_predict(X_spec)
# clustering evaluation metrics
print(confusion_matrix(y_train, y_pred))
print (completeness_score(y_train, y_pred))
with plt.style.context('fivethirtyeight'):
plt.title("Spectral embedding & spectral clustering on MNIST")
plt.scatter(X_spec[:, 0], X_spec[:, 1], c=y_pred, s=50, cmap=plt.cm.get_cmap("jet", 10))
plt.colorbar(ticks=range(10))
plt.clim(-0.5, 9.5)
plt.show()
def check_clusters(self):
print self.colors
print 'number of clusters is ' + str(self.clusters)
print 'silhouette score is ' + str(self.sil)
print 'homogeneity is ' + str(homogeneity_score(self.colors, self.labels))
print 'completeness is ' + str(completeness_score(self.colors, self.labels))
def evaluate(colors, labels):
b = homogeneity_score(colors, labels)
c = completeness_score(colors, labels)
logging.debug('homogeneity is %d' % b)
logging.debug('completeness is %d' % c)
def evaluate(colors, labels):
b = homogeneity_score(colors, labels)
c = completeness_score(colors, labels)
print 'homogeneity is ' + str(b)
print 'completeness is ' + str(c)