def main():
print("Loading dataset")
os.chdir('../../../..')
dfs = [pd.read_csv('data/empty.csv', header=None) for k in range(28)]
# iris pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data', header=None)
dfs[0] = pd.read_csv('data/datasets_metrics/iris.csv', header=None)
# wine pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/wine/wine.data', header=None)
dfs[1] = pd.read_csv('data/datasets_metrics/wine.csv', header=None)
# glass pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/glass/glass.data', header=None)
dfs[2] = pd.read_csv('data/datasets_metrics/glass.csv', header=None)
# breast cancer wincosin pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/breast-cancer-wisconsin.data', header=None)
dfs[3] = pd.read_csv('data/datasets_metrics/breast-cancer-wisconsin.csv', header=None)
# wdbc pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/wdbc.data', header=None)
dfs[4] = pd.read_csv('data/datasets_metrics/wdbc.csv', header=None)
# liver disorders pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/liver-disorders/bupa.data', header=None)
dfs[5] = pd.read_csv('data/datasets_metrics/bupa.csv', header=None)
# contraceptive method choice pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/cmc/cmc.data', header=None)
dfs[6] = pd.read_csv('data/datasets_metrics/cmc.csv', header=None)
# tiroide pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/thyroid-disease/new-thyroid.data', header=None)
dfs[7] = pd.read_csv('data/datasets_metrics/new-thyroid.csv', header=None)
# dematology pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/dermatology/dermatology.data', header=None)
dfs[8] = pd.read_csv('data/datasets_metrics/dermatology.csv', header=None)
# egyptian skools http://www.dm.unibo.it/~simoncin/EgyptianSkulls.html
dfs[9] = df = pd.read_csv('data/datasets_metrics/egyptian-skulls.csv', header=None)
# heart statlog pd.read_csv('http://archive.ics.uci.edu/ml/machine-learning-databases/statlog/heart/heart.dat', header=None)
dfs[10] = df = pd.read_csv('data/datasets_metrics/heart.csv', header=None)
# ionosphere
dfs[11] = df = pd.read_csv('data/datasets_metrics/ionosphere.csv', header=None)
# vehicle
dfs[12] = df = pd.read_csv('data/datasets_metrics/vehicle.csv', header=None)
# balance scale pd.read_csv('http://archive.ics.uci.edu/ml/machine-learning-databases/balance-scale/balance-scale.data', header=None)
dfs[13] = df = pd.read_csv('data/datasets_metrics/balance-scale.csv', header=None)
# sonar
dfs[14] = df = pd.read_csv('data/datasets_metrics/sonar.csv', header=None)
# zoo pd.read_csv('http://archive.ics.uci.edu/ml/machine-learning-databases/zoo/zoo.data', header=None)
dfs[15] = df = pd.read_csv('data/datasets_metrics/zoo.csv', header=None)
# isolet5
dfs[16] = df = pd.read_csv('data/datasets_metrics/isolet5.csv', header=None)
# movement libras
dfs[17] = df = pd.read_csv('data/datasets_metrics/libras.csv', header=None)
# cleveland http://archive.ics.uci.edu/ml/machine-learning-databases/heart-disease/processed.cleveland.data
dfs[18] = df = pd.read_csv('data/datasets_metrics/cleveland.csv', header=None)
# australian
dfs[19] = df = pd.read_csv('data/datasets_metrics/australian.csv', header=None)
dfs[20] = df = pd.read_csv('data/shapesets/compound.csv', header=None)
dfs[21] = df = pd.read_csv('data/shapesets/flame.csv', header=None)
dfs[22] = df = pd.read_csv('data/shapesets/jain.csv', header=None)
dfs[23] = df = pd.read_csv('data/shapesets/r15.csv', header=None)
dfs[24] = df = pd.read_csv('data/shapesets/d31.csv', header=None)
dfs[25] = df = pd.read_csv('data/shapesets/spiral.csv', header=None)
dfs[26] = df = pd.read_csv('data/shapesets/pathbased.csv', header=None)
dfs[27] = df = pd.read_csv('data/shapesets/agregation.csv', header=None)
# hill-valley
# diabetes
# olive
# crud oil
# musk version 1
# landsat satellite
# heart disease len(dfs)
for eachdataset in range(0, len(dfs)):
df = dfs[eachdataset]
df = df.drop(len(df.columns) - 1, 1)
x = df[df.apply(lambda x: sum([x_ == '?' for x_ in x]) == 0, axis=1)]
x = x.iloc[:, :].values.astype(float)
print("Nomalizing dataset so that all dimenions are in the same scale")
std = MinMaxScaler()
x = std.fit_transform(x)
x = x[np.random.permutation(len(x))]
SIMULATIONS = 30
df = []
# name = ['KMeans', 'FCMeans']
name = ['FCMeans']
# name = ['KPSO', 'CPSO', 'PSC']
for i in range(len(name)):
metrics = []
mean = []
std = []
rng = range(2, 10)
# 27
for metricNumber in range(0, 26):
for k in rng:
print("Number of Clusters = " + str(k) + "\n")
for j in range(SIMULATIONS):
print("Run ====> " + str(j))
# if (name[i] == 'KPSO'):
# clf = KPSO(n_clusters=k, swarm_size=15, n_iter=500, w=0.72, lb_w=0.4, c1=1.49, c2=1.49)
# elif (name[i] == 'CPSO'):
# clf = CPSO(n_clusters=k, swarm_size=15, n_iter=500, w=0.72, lb_w=0.4, c1=1.49, c2=1.49)
# elif (name[i] == 'PSC'):
# clf = PSC(minf=0, maxf=1, swarm_size=k, n_iter=500, w=0.95, v_max=0.01)
if (name[i] == 'KMeans'):
clf = KMeans(n_clusters=k)
elif (name[i] == 'FCMeans'):
clf = FCMeans(n_clusters=k, n_iter=1000)
elif (name[i] == 'PSOC'):
clf = PSOC(n_clusters=k, swarm_size=15, n_iter=500, w=0.72, c1=1.49, c2=1.49)
clf.fit(x)
if not os.path.isdir("results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i]))+"/"):
os.makedirs("results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i]))+"/")
sn = "results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i])+"/")
sn = sn + "dataset_{0}".format(str(eachdataset))
sn = sn + "_metric_{0}".format(str(metricNumber))
sn = sn + "_algorithm_{0}".format(str(name[i]))
sn = sn + "_k_{0}".format(str(k))
sn = sn + "_simulation_{0}".format(str(j))
savecentroids = pd.DataFrame(clf.centroids)
savecentroids = savecentroids.transpose()
savecentroids.to_csv(sn+"_centroids.csv")
clusters = {}
for c in clf.centroids:
clusters[c] = []
for xi in x:
dist = [np.linalg.norm(xi - clf.centroids[c]) for c in clf.centroids]
class_ = dist.index(min(dist))
clusters[class_].append(xi)
# precisa inverter?
file = open(sn+"_clusters.csv", 'w')
file.write(str(len(clf.centroids)) + '\n')
for c in range(len(clf.centroids)):
file.write(str(len(clusters[c])) + '\n')
for xi in range(len(clusters[c])):
file.write(str(clusters[c][xi][0]))
for xij in range(1, len(clusters[c][xi])):
file.write(' ' + str(clusters[c][xi][xij]))
file.write('\n')
file.close()
# if not os.path.exists('gap/' + name[i] + '/' + str(k) + '-centroids'):
# os.makedirs('gap/' + name[i] + '/' + str(k) + '-centroids')
# centroids.to_csv('gap/' + name[i] + '/' + str(k) + '-centroids/centroid-simulation' + str(j) + '.csv', index=False)
# metrics.append(Metrics.Metrics.inter_cluster_statistic(clf))
# metrics.append(Metrics.Metrics.cluster_separation_crisp(data=x, clf=clf))
# metrics.append(Metrics.Metrics.cluster_separation_fuzzy(data=x, clf=clf, m=2.0))
# metrics.append(Metrics.Metrics.abgss(data=x, clf=clf))
# metrics.append(Metrics.Metrics.edge_index(data=x, clf=clf, number_neighbors=4))
# metrics.append(Metrics.Metrics.cluster_connectedness(data=x, clf=clf, number_neighbors=4))
# metrics.append(Metrics.Metrics.intra_cluster_statistic(clf))
# metrics.append(Metrics.Metrics.ball_hall(data=x, clf=clf))
# metrics.append( Metrics.Metrics.j_index(data=x, clf=clf, m=2.0) )
# metrics.append( Metrics.Metrics.total_within_cluster_variance(data=x, clf=clf) )
# metrics.append(Metrics.Metrics.classification_entropy(data=x, clf=clf, m=2.0))
# metrics.append(Metrics.Metrics.intra_cluster_entropy(data=x, clf=clf))
# metrics.append(Metrics.Metrics.variance_based_ch(data=x, clf=clf))
# metrics.append(Metrics.Metrics.hartigan(data=x, clf=clf))
# metrics.append(Metrics.Metrics.xu(data=x, clf=clf))
# metrics.append(Metrics.Metrics.rl(data=x, clf=clf))
# metrics.append(Metrics.Metrics.wb(data=x, clf=clf))
# metrics.append(Metrics.Metrics.xie_beni(data=x, clf=clf, m=2.0))
c = clf.centroids[0][0]
# metrics.append(Metrics.Metrics.i_index(data=x, clf=clf, centroidUnique=c))
# metrics.append(Metrics.Metrics.dunn_index(data=x, clf=clf))
# metrics.append(Metrics.Metrics.davies_bouldin(data=x, clf=clf))
# metrics.append(Metrics.Metrics.cs_index(data=x, clf=clf))
# metrics.append(Metrics.Metrics.silhouette(data=x, clf=clf))
# metrics.append(Metrics.Metrics.min_max_cut(data=x, clf=clf))
# metrics.append(Metrics.Metrics.gap_statistic(data=x, clf=clf))
metrics.append(Metrics.clustering_evaluation("{0}".format(str(metricNumber)), centroids=clf.centroids, data=x, clf=clf, m=2.0, number_neighbors=2, centroidUnique=c))
mean.append(np.mean(metrics))
std.append(np.std(metrics))
df.append([name[i], k, np.mean(metrics), np.std(metrics)])
metrics = []
# plt.subplot(130 + (i + 1))
plt.clf()
plt.title(str(name[i]) + ' - Metric')
plt.errorbar(rng, mean, yerr=std, marker='o', ecolor='b', capthick=2, barsabove=True)
plt.xlabel('Clusters')
plt.ylabel('Metric')
saveName = "results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i])+"/")
saveName = saveName + "dataset_{0}".format(str(eachdataset))
saveName = saveName + "_metric_{0}".format(str(metricNumber))
saveName = saveName + "_algorithm_{0}".format(str(name[i]))
plt.savefig(saveName+".pdf")
df = pd.DataFrame(df)
df.columns = ['ALGORITHM', 'CLUSTERS', 'MEAN', 'STD']
df.to_csv(saveName+".csv")
mean = []
std = []
plt.tight_layout()
def main(parameters_simulation=None):
data_file = open(parameters_simulation)
param_simultations = json.load(data_file)
pathname_dataset = param_simultations["pathname_dataset"]
pathname_dir_results = param_simultations["pathname_dir_results"]
num_simulations = param_simultations["NUN_SIMULATIONS"] # 30
num_trials = param_simultations["NUM_TRIALS"] # 50
num_iterations = param_simultations["NUM_ITERATIONS"] # 500
swarm_size = param_simultations["SWARM_SIZE"] # 15
name_classifier = param_simultations["ALGORITHMS"]
clustering_metrics = param_simultations["EVALUATION_METRICS"]
criteria = []
table_criteria = {}
for idx_alg in range(len(clustering_metrics)):
value = clustering_metrics[idx_alg]
criteria.append(value['criteria'])
table_criteria[value['criteria']] = (
clustering_metrics[idx_alg]['name'],
clustering_metrics[idx_alg]['file'])
time.sleep(2)
pathname_output = pathname_dir_results + '/2017_LA-CCI_ClusteringSimulation'
currDirectory = (pathname_output + '-ID-' +
datetime.now().strftime('%d-%b-%Y-%Hh:%Mm:%Ss'))
pathname_output = currDirectory
if not os.path.exists(pathname_output):
os.makedirs(pathname_output)
else:
raise Exception("This simulation cannot execute!")
time.sleep(1)
timestamp = datetime.now().strftime('%a, %d %b %Y at %Hh:%Mm:%Ss')
if not os.path.exists(pathname_output + '/timestamp.txt'):
file_timestamp = open(pathname_output + '/timestamp.txt', 'a')
file_timestamp.write('This simulation started on ' + timestamp)
file_timestamp.close()
print("This simulation started on " + timestamp)
else:
raise Exception("This simulation cannot execute!")
print("Loading dataset")
df = pd.read_excel(io=pathname_dataset, sheetname='Plan1')
df.drop(['ID', 'Nome', 'E-mail'], 1, inplace=True)
print pathname_dataset
x = df.iloc[:, :].values.astype(float)
print("Normalizing dataset so that all dimenions are in the same scale")
min_max_scaler = MinMaxScaler()
x = min_max_scaler.fit_transform(x)
indices_attributes_4p = np.array([1, 7, 8, 10, 12, 13, 16])
indices_attributes_4p = indices_attributes_4p - 1
indices_attributes_5p = np.array([1, 7, 8, 10, 12, 13])
indices_attributes_5p = indices_attributes_5p - 1
x = x[:, indices_attributes_4p]
# x = np.array(x)
# The term clf means classifier
# name_classifier = ['PSOC']
# name_classifier = ['PSC']
KList = range(2, 10)
idx_success_simulation = {}
idx_trial = {}
idx_fail_simulation = {}
check_sim_its_over = {}
for idx_alg in range(len(name_classifier)):
idx_success_simulation[name_classifier[idx_alg]] = {}
idx_trial[name_classifier[idx_alg]] = {}
idx_fail_simulation[name_classifier[idx_alg]] = {}
check_sim_its_over[name_classifier[idx_alg]] = {}
for k in KList:
idx_success_simulation[name_classifier[idx_alg]][k] = 0
idx_trial[name_classifier[idx_alg]][k] = 0
idx_fail_simulation[name_classifier[idx_alg]][k] = 0
check_sim_its_over[name_classifier[idx_alg]][k] = False
metrics_list_sim_by_algorithm_and_k = {}
mean_metric_by_algorithm_and_k = {}
std_metric_by_algorithm_and_k = {}
for m in range(len(criteria)):
mean_metric_by_algorithm_and_k[criteria[m]] = {}
std_metric_by_algorithm_and_k[criteria[m]] = {}
for m in range(len(criteria)):
metrics_list_sim_by_algorithm_and_k[criteria[m]] = {}
for idx_alg in range(len(name_classifier)):
metrics_list_sim_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]] = {}
for k in KList:
metrics_list_sim_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]][k] = []
for m in range(len(criteria)):
min_value_mean = {}
max_value_mean = {}
min_value_std = {}
max_value_std = {}
for m in range(len(criteria)):
min_value_mean[criteria[m]] = np.inf
max_value_mean[criteria[m]] = -np.inf
min_value_std[criteria[m]] = np.inf
max_value_std[criteria[m]] = -np.inf
#First the simulation
#(idx_success_simulation < num_simulations) and (idx_trial < num_trials)
finished = False
idx_fig = 1
Round = 1
while not finished:
list_results_by_metric = {}
for m in range(len(criteria)):
list_results_by_metric[criteria[m]] = []
for idx_alg in range(len(name_classifier)):
for m in range(len(criteria)):
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]] = np.zeros((len(KList), ))
std_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]] = np.zeros((len(KList), ))
for k in KList:
if not check_sim_its_over[name_classifier[idx_alg]][k]:
create_dir_algorithms_and_k(idx_alg, k, name_classifier,
pathname_output)
pathname_output_clusters = pathname_output + '/data/' + name_classifier[idx_alg] + '/' \
+ '/clusters/' + str(k) + '-centroids'
pathname_output_metrics = pathname_output + '/data/' + name_classifier[idx_alg] + '/' \
+ '/metrics/' + str(k) + '-centroids'
pathname_output_evolution_success = pathname_output + '/data/' + name_classifier[idx_alg] + '/' \
+ '/evolution/success/' + str(k) + '-centroids'
pathname_output_evolution_fail = pathname_output + '/data/' + name_classifier[idx_alg] + '/' \
+ '/evolution/fail/' + str(k) + '-centroids'
if (name_classifier[idx_alg] == 'KMPSOC'):
clf = KMPSOC(n_clusters=k,
swarm_size=swarm_size,
n_iter=num_iterations,
w=0.72,
c1=1.49,
c2=1.49)
elif (name_classifier[idx_alg] == 'PSOC'):
clf = PSOC(n_clusters=k,
swarm_size=swarm_size,
n_iter=num_iterations,
w=0.72,
c1=1.49,
c2=1.49)
elif (name_classifier[idx_alg] == 'PSOCKM'):
clf = PSOCKM(n_clusters=k,
swarm_size=swarm_size,
n_iter=num_iterations,
w=0.72,
c1=1.49,
c2=1.49)
elif (name_classifier[idx_alg] == 'KMeans'):
clf = KMeans(n_clusters=k)
elif (name_classifier[idx_alg] == 'PSC'):
clf = PSC(swarm_size=k,
n_iter=num_iterations,
w=0.95,
c1=2.05,
c2=2.05,
c3=1.0,
c4=1.0,
v_max=0.01)
clf.fit(x)
centroids = clf.centroids
Round = Round + 1
if clf.solution.number_of_effective_clusters == k:
filename = pathname_output_clusters + '/centroid-' + str(k) + \
'-success-simulation-' + str(
idx_success_simulation[name_classifier[idx_alg]][k] + 1) + '.csv'
else:
filename = pathname_output_clusters + '/centroid-' + str(k) \
+ '-fail-simulation-' + str(idx_fail_simulation[name_classifier[idx_alg]][k] + 1) + '.csv'
dataframe_centroids = pd.DataFrame(centroids)
dataframe_centroids.transpose().to_csv(filename,
sep=" ",
index=False)
if clf.solution.number_of_effective_clusters == k:
filename = pathname_output_clusters + '/cluster-' + str(k) \
+ '-success-simulation-' + str(idx_success_simulation[name_classifier[idx_alg]][k] + 1) + '.cluster'
else:
filename = pathname_output_clusters + '/cluster-' + str(k) + \
'-fail-simulation-' + str(idx_fail_simulation[name_classifier[idx_alg]][k] + 1) + '.cluster'
file = open(filename, 'w')
file.write(str(len(clf.centroids)) + '\n')
file.write(
str(clf.solution.number_of_effective_clusters) + '\n')
for c in range(len(clf.centroids)):
if len(clf.solution.clusters[c]) > 0:
file.write(
str(len(clf.solution.clusters[c])) + '\n')
for xi in range(len(clf.solution.clusters[c])):
file.write(str(
clf.solution.clusters[c][xi][0]))
for xij in range(
1, len(clf.solution.clusters[c][xi])):
file.write(
' ' +
str(clf.solution.clusters[c][xi][xij]))
file.write('\n')
file.close()
if clf.solution.number_of_effective_clusters == k:
evol_directory = pathname_output_evolution_success + '/evolution-' + str(k) \
+ '-success-simulation-' + str(idx_success_simulation[name_classifier[idx_alg]][k] + 1)
if not os.path.exists(evol_directory):
os.makedirs(evol_directory)
store_evolution(
evol_directory, clf.debugger, k,
idx_success_simulation[name_classifier[idx_alg]][k]
+ 1)
else:
evol_directory = pathname_output_evolution_fail + '/evolution-' + str(k) + \
'-fail-simulation-' + str(idx_fail_simulation[name_classifier[idx_alg]][k] + 1)
if not os.path.exists(evol_directory):
os.makedirs(evol_directory)
store_evolution(
evol_directory, clf.debugger, k,
idx_fail_simulation[name_classifier[idx_alg]][k] +
1)
if clf.solution.number_of_effective_clusters == k:
for m in range(len(criteria)):
value = Metrics.clustering_evaluation(
criteria=criteria[m],
centroids=centroids,
data=x,
clf=clf)
metrics_list_sim_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]][k].append(value)
for m in range(len(criteria)):
pd.DataFrame(metrics_list_sim_by_algorithm_and_k[
criteria[m]][name_classifier[idx_alg]][k]).to_csv(
pathname_output_metrics + '/' +
table_criteria[criteria[m]][1] + '.csv',
sep=" ")
if clf.solution.number_of_effective_clusters == k:
idx_success_simulation[name_classifier[idx_alg]][
k] = idx_success_simulation[
name_classifier[idx_alg]][k] + 1
else:
idx_fail_simulation[
name_classifier[idx_alg]][k] = idx_fail_simulation[
name_classifier[idx_alg]][k] + 1
idx_trial[name_classifier[idx_alg]][k] = idx_trial[
name_classifier[idx_alg]][k] + 1
if (idx_success_simulation[name_classifier[idx_alg]][k] >=
num_simulations
or idx_trial[name_classifier[idx_alg]][k] >=
num_trials):
check_sim_its_over[name_classifier[idx_alg]][k] = True
for d in check_sim_its_over.values():
if not all(d.values()):
break
else:
finished = True
print(
"Round(" + str(Round) + ") ........................ " +
"(SUCCESS = " + str(idx_success_simulation[
name_classifier[idx_alg]][k]) + ", "
"FAIL = " +
str(idx_fail_simulation[name_classifier[idx_alg]][k]) +
", " + "TRAIL = " +
str(idx_trial[name_classifier[idx_alg]][k]) + ", " +
"CLF = " + name_classifier[idx_alg] + ", K = " +
str(k) + ")" + "\n")
for m in range(len(criteria)):
idx_k = 0
for ik in KList:
if len(metrics_list_sim_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]][ik]) > 0:
mean_metric_by_algorithm_and_k[criteria[m]][name_classifier[idx_alg]][idx_k] = \
np.mean(metrics_list_sim_by_algorithm_and_k[criteria[m]][name_classifier[idx_alg]][ik])
std_metric_by_algorithm_and_k[criteria[m]][name_classifier[idx_alg]][idx_k] = \
np.std(metrics_list_sim_by_algorithm_and_k[criteria[m]][name_classifier[idx_alg]][ik])
list_results_by_metric[criteria[m]].append([
name_classifier[idx_alg], ik,
np.mean(metrics_list_sim_by_algorithm_and_k[
criteria[m]][name_classifier[idx_alg]][ik]),
np.std(metrics_list_sim_by_algorithm_and_k[
criteria[m]][name_classifier[idx_alg]][ik])
])
idx_k = idx_k + 1
for m in range(len(criteria)):
min_value_mean[criteria[m]] = np.inf
max_value_mean[criteria[m]] = -np.inf
max_value_std[criteria[m]] = -np.inf
for m in range(len(criteria)):
min_value_mean[criteria[m]] = np.amin(
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]])
max_value_mean[criteria[m]] = np.amax(
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]])
max_value_std[criteria[m]] = np.amax(
std_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]])
for m in range(len(criteria)):
#print mean_metric_by_algorithm_and_k
plt.figure()
plt.title(
str(name_classifier[idx_alg]) + ' - ' +
table_criteria[criteria[m]][0])
# plt.errorbar(KList, mean_metric_by_algorithm_and_k[criteria[m]], yerr=std_metric_by_algorithm_and_k[criteria[m]], linewidth=0.5, elinewidth=0.5, color='b', capthick=2, barsabove=True)
plt.errorbar(KList,
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]],
yerr=(std_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]]),
linewidth=0.5,
elinewidth=0.5,
color='b')
plt.plot(KList,
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]],
color='b',
marker='o',
linewidth=0.5,
markersize=5)
plt.xlabel('K')
plt.ylabel(table_criteria[criteria[m]][0])
ymin = min_value_mean[criteria[m]] - max_value_std[criteria[m]]
ymax = max_value_mean[criteria[m]] + max_value_std[criteria[m]]
delta = ymax - ymin
plt.ylim([ymin - 0.01 * delta, ymax + 0.01 * delta])
plt.tight_layout()
plt.savefig(pathname_output + '/data/' +
name_classifier[idx_alg] + "/" +
name_classifier[idx_alg] + "-" +
table_criteria[criteria[m]][1] + '.pdf')
plt.close("all")
for m in range(len(criteria)):
df_by_metric = pd.DataFrame(list_results_by_metric[criteria[m]])
df_by_metric.columns = ['ALGORITHM', 'CLUSTERS', 'MEAN', 'STD']
df_by_metric.to_csv(pathname_output + '/data/' +
table_criteria[criteria[m]][1] + '.csv',
index=False)
# mean_metric_by_algorithm[name_classifier[idx_alg]] = mean_metric_by_algorithm_and_k
# std_metric_by_algorithm[name_classifier[idx_alg]] = std_metric_by_algorithm_and_k
#
for m in range(len(criteria)):
min_value_mean[criteria[m]] = np.inf
max_value_mean[criteria[m]] = -np.inf
max_value_std[criteria[m]] = -np.inf
for m in range(len(criteria)):
for idx_alg in range(len(name_classifier)):
curr_min_mean = np.amin(mean_metric_by_algorithm_and_k[criteria[m]]
[name_classifier[idx_alg]])
curr_max_mean = np.amax(mean_metric_by_algorithm_and_k[criteria[m]]
[name_classifier[idx_alg]])
min_value_mean[criteria[m]] = np.minimum(
min_value_mean[criteria[m]], curr_min_mean)
max_value_mean[criteria[m]] = np.maximum(
max_value_mean[criteria[m]], curr_max_mean)
curr_max_std = np.amax(std_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]])
max_value_std[criteria[m]] = np.maximum(max_value_std[criteria[m]],
curr_max_std)
for m in range(len(criteria)):
for idx_alg in range(len(name_classifier)):
plt.figure()
plt.title(
str(name_classifier[idx_alg]) + ' - ' +
table_criteria[criteria[m]][0])
# plt.errorbar(KList, mean_metric_by_algorithm_and_k[criteria[m]], yerr=std_metric_by_algorithm_and_k[criteria[m]], linewidth=0.5, elinewidth=0.5, color='b', capthick=2, barsabove=True)
plt.errorbar(KList,
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]],
yerr=std_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]],
linewidth=0.5,
elinewidth=0.5,
color='b')
plt.plot(KList,
mean_metric_by_algorithm_and_k[criteria[m]][
name_classifier[idx_alg]],
color='b',
marker='o',
linewidth=0.5,
markersize=5)
plt.xlabel('K')
plt.ylabel(table_criteria[criteria[m]][0])
ymin = min_value_mean[criteria[m]] - max_value_std[criteria[m]]
ymax = max_value_mean[criteria[m]] + max_value_std[criteria[m]]
delta = ymax - ymin
plt.ylim([ymin - 0.01 * delta, ymax + 0.01 * delta])
plt.tight_layout()
plt.savefig(pathname_output + '/data/' + name_classifier[idx_alg] +
"/" + name_classifier[idx_alg] + "-" +
table_criteria[criteria[m]][1] + '_final.pdf')
plt.close("all")
def main():
num_exec = 30
swarm_size = 30
num_iter = 1000
# names = ['KMeans', 'FCMeans', 'PSOC', 'PSC', 'KMPSOC', 'PSOCKM']
names = ['PSC', 'PSOC', 'KMPSOC', 'PSOCKM']
# names = ['PSOC', 'PSC', 'KMPSOC', 'PSOCKM']
print("Loading dataset")
os.chdir('../../..')
df = pd.read_csv('data/booking_website/booking_website_without_empty_values.csv')
df = df.drop(['id'], axis=1)
# df = df.drop(['idade'], axis=1)
df = df.drop(['sexo'], axis=1)
x = df[df.apply(lambda x: sum([x_ == '?' for x_ in x]) == 0, axis=1)]
x = x.iloc[:, :].values.astype(float)
print("Nomalizing dataset so that all dimenions are in the same scale")
std = MinMaxScaler()
x = std.fit_transform(x)
x = x[np.random.permutation(len(x))]
for i in range(len(names)):
metrics = []
rng = range(2, 11)
for metricNumber in ["intraClusterStatistic", "quantizationError", "sumInterClusterDistance"]:
# for metricNumber in ["gap"]:
print("Algorithm: " + names[i])
mean = []
std = []
dff = []
for k in rng:
# print(" Number of Clusters = " + str(k))
for j in tqdm(range(num_exec)):
if names[i] == 'KPSO':
clf = KMPSOC(n_clusters=k, swarm_size=swarm_size, n_iter=num_iter, w=0.72, c1=1.49, c2=1.49)
elif names[i] == 'PSOC':
clf = PSOC(n_clusters=k, swarm_size=swarm_size, n_iter=num_iter, w=0.72, c1=1.49, c2=1.49)
elif names[i] == 'PSC':
clf = PSC(swarm_size=k, n_iter=num_iter, w=0.95, c1=2.05, c2=2.05, c3=1.0, c4=1.0, v_max=0.001)
elif names[i] == 'PSOCKM':
clf = PSOCKM(n_clusters=k, swarm_size=swarm_size, n_iter=num_iter, w=0.72, c1=1.49, c2=1.49)
elif names[i] == 'KMeans':
clf = KMeans(n_clusters=k, n_iter=num_iter, shuffle=True, tolerance=0.00001)
elif names[i] == 'FCMeans':
clf = FCMeans(n_clusters=k, n_iter=num_iter, fuzzy_c=2, tolerance=0.001)
clf.fit(x)
out_dir = "results/booking/algorithm_{}/metric_{}/".format(names[i], metricNumber)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
file_name = out_dir + "{}_k_{}_exec_{}.csv".format('centroids', k, j)
save_centroids = pd.DataFrame(clf.centroids)
save_centroids = save_centroids.transpose()
save_centroids.to_csv(file_name)
clusters = {}
for c in clf.centroids:
clusters[c] = []
for xi in x:
dist = [np.linalg.norm(xi - clf.centroids[c]) for c in clf.centroids]
class_ = dist.index(min(dist))
clusters[class_].append(xi)
clusters_file = open(out_dir + "{}_k_{}_exec_{}.csv".format('clusters', k, j), 'w')
clusters_file.write(str(len(clf.centroids)) + '\n')
for c in range(len(clf.centroids)):
clusters_file.write(str(len(clusters[c])) + '\n')
for xi in range(len(clusters[c])):
clusters_file.write(str(clusters[c][xi][0]))
for xij in range(1, len(clusters[c][xi])):
clusters_file.write(' ' + str(clusters[c][xi][xij]))
clusters_file.write('\n')
clusters_file.close()
c = clf.centroids[0][0]
metrics.append(
Metrics.clustering_evaluation("{}".format(metricNumber), centroids=clf.centroids, data=x,
clf=clf, m=2.0, number_neighbors=2, centroidUnique=c))
mean.append(np.mean(metrics))
std.append(np.std(metrics))
dff.append([names[i], k, np.mean(metrics), np.std(metrics)])
metrics = []
# plt.subplot(130 + (i + 1))
plt.figure()
figure_name = "results/booking/algorithm_{}/metric_{}/plot.png".format(names[i], metricNumber)
plt.title(str(names[i]) + ' - Metric ' + metricNumber)
plt.errorbar(rng, mean, yerr=std, marker='o', ecolor='b', capthick=2, barsabove=True)
plt.xlabel('Clusters')
plt.ylabel('Metric')
plt.tight_layout()
plt.savefig(figure_name)
save_name = "results/booking/algorithm_{}/metric_{}/output.csv".format(names[i], metricNumber)
dff = pd.DataFrame(dff)
dff.columns = ['ALGORITHM', 'CLUSTERS', 'MEAN', 'STD']
dff.to_csv(save_name)
def main():
print("Loading dataset")
eachdataset = 072017
df = pd.read_excel(io='data/Saidafinal4periodo.xlsx', sheetname='Plan1')
df.drop(['ID', 'Nome', 'E-mail'], 1, inplace=True)
x = df.iloc[:, :].values.astype(float)
print("Nomalizing dataset so that all dimenions are in the same scale")
std_metric_by_algorithm_and_k = MinMaxScaler()
x = std_metric_by_algorithm_and_k.fit_transform(x)
indices_attributes = np.array([1,7,8,10,12,13,16])
indices_attributes = indices_attributes -1
x = x[:,indices_attributes]
SIMULATIONS = 30
# name = ['KMeans', 'FCMeans']
name = ['PSOC']
# name = ['KPSO', 'CPSO', 'PSC']
for i in range(len(name)):
metrics = []
mean = []
std = []
rng = range(2, 10)
for metricNumber in range(1, 26):
for k in rng:
print("Number of Clusters = " + str(k) + "\n")
for j in range(SIMULATIONS):
print("Run ====> " + str(j))
# if (name[i] == 'KPSO'):
# clf = KPSO(n_clusters=k, swarm_size=15, n_iter=500, w=0.72, lb_w=0.4, c1=1.49, c2=1.49)
# elif (name[i] == 'CPSO'):
# clf = CPSO(n_clusters=k, swarm_size=15, n_iter=500, w=0.72, lb_w=0.4, c1=1.49, c2=1.49)
# elif (name[i] == 'PSC'):
# clf = PSC(minf=0, maxf=1, swarm_size=k, n_iter=500, w=0.95, v_max=0.01)
if (name[i] == 'KMeans'):
clf = KMeans(n_clusters=k)
elif (name[i] == 'FCMeans'):
clf = FCMeans(n_clusters=k)
elif (name[i] == 'PSOC'):
clf = PSOC(n_clusters=k, swarm_size=30, n_iter=1000, w=0.72, c1=1.49, c2=1.49)
clf.fit(x)
if not os.path.isdir("results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i]))+"/"):
os.makedirs("results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i]))+"/")
sn = "results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i])+"/")
sn = sn + "dataset_{0}".format(str(eachdataset))
sn = sn + "_metric_{0}".format(str(metricNumber))
sn = sn + "_algorithm_{0}".format(str(name[i]))
sn = sn + "_k_{0}".format(str(k))
sn = sn + "_simulation_{0}".format(str(j))
savecentroids = pd.DataFrame(clf.centroids)
savecentroids = savecentroids.transpose()
savecentroids.to_csv(sn+"_centroids.csv")
# clust = Metrics.get_clusters(x, clf.centroids)
# # print np.array(clust[0])
# c = []
# for ii in range(len(clust)):
# c.append(np.array(clust[ii]))
# sc = pd.DataFrame(c)
# # precisa inverter?
# sc.to_csv(sn+"_clusters.csv")
file = open(sn+"_clusters.csv", 'w')
file.write(str(len(clf.centroids)) + '\n')
file.write(str(clf.solution.number_of_effective_clusters)+ '\n')
for c in range(len(clf.centroids)):
file.write(str(len(clf.solution.clusters[c])) + '\n')
for xi in range(len(clf.solution.clusters[c])):
file.write(str(clf.solution.clusters[c][xi][0]))
for xij in range(1,len(clf.solution.clusters[c][xi])):
file.write(' ' + str(clf.solution.clusters[c][xi][xij]))
file.write('\n')
file.close()
# if not os.path.exists('gap/' + name[i] + '/' + str(k) + '-centroids'):
# os.makedirs('gap/' + name[i] + '/' + str(k) + '-centroids')
# centroids.to_csv('gap/' + name[i] + '/' + str(k) + '-centroids/centroid-simulation' + str(j) + '.csv', index=False)
# metrics.append(Metrics.Metrics.inter_cluster_statistic(clf))
# metrics.append(Metrics.Metrics.cluster_separation_crisp(data=x, clf=clf))
# metrics.append(Metrics.Metrics.cluster_separation_fuzzy(data=x, clf=clf, m=2.0))
# metrics.append(Metrics.Metrics.abgss(data=x, clf=clf))
# metrics.append(Metrics.Metrics.edge_index(data=x, clf=clf, number_neighbors=4))
# metrics.append(Metrics.Metrics.cluster_connectedness(data=x, clf=clf, number_neighbors=4))
# metrics.append(Metrics.Metrics.intra_cluster_statistic(clf))
# metrics.append(Metrics.Metrics.ball_hall(data=x, clf=clf))
# metrics.append( Metrics.Metrics.j_index(data=x, clf=clf, m=2.0) )
# metrics.append( Metrics.Metrics.total_within_cluster_variance(data=x, clf=clf) )
# metrics.append(Metrics.Metrics.classification_entropy(data=x, clf=clf, m=2.0))
# metrics.append(Metrics.Metrics.intra_cluster_entropy(data=x, clf=clf))
# metrics.append(Metrics.Metrics.variance_based_ch(data=x, clf=clf))
# metrics.append(Metrics.Metrics.hartigan(data=x, clf=clf))
# metrics.append(Metrics.Metrics.xu(data=x, clf=clf))
# metrics.append(Metrics.Metrics.rl(data=x, clf=clf))
# metrics.append(Metrics.Metrics.wb(data=x, clf=clf))
# metrics.append(Metrics.Metrics.xie_beni(data=x, clf=clf, m=2.0))
c = clf.centroids[0][0]
# metrics.append(Metrics.Metrics.i_index(data=x, clf=clf, centroidUnique=c))
# metrics.append(Metrics.Metrics.dunn_index(data=x, clf=clf))
# metrics.append(Metrics.Metrics.davies_bouldin(data=x, clf=clf))
# metrics.append(Metrics.Metrics.cs_index(data=x, clf=clf))
# metrics.append(Metrics.Metrics.silhouette(data=x, clf=clf))
# metrics.append(Metrics.Metrics.min_max_cut(data=x, clf=clf))
# metrics.append(Metrics.Metrics.gap_statistic(data=x, clf=clf))
metrics.append(Metrics.clustering_evaluation("{0}".format(str(metricNumber)), centroids=clf.centroids, data=x, clf=clf, m=2.0, number_neighbors=2, centroidUnique=c))
mean.append(np.mean(metrics))
std.append(np.std(metrics))
df.append([name[i], k, np.mean(metrics), np.std(metrics)])
metrics = []
# plt.subplot(130 + (i + 1))
plt.clf()
plt.title(str(name[i]) + ' - Metric')
plt.errorbar(rng, mean, yerr=std, marker='o', ecolor='b', capthick=2, barsabove=True)
plt.xlabel('Clusters')
plt.ylabel('Metric')
saveName = "results/metrics/"+"dataset_{0}".format(str(eachdataset))+"/metric_{0}".format(str(metricNumber))+"/algorithm_{0}".format(str(name[i])+"/")
saveName = saveName + "dataset_{0}".format(str(eachdataset))
saveName = saveName + "_metric_{0}".format(str(metricNumber))
saveName = saveName + "_algorithm_{0}".format(str(name[i]))
plt.savefig(saveName+".pdf")
df = pd.DataFrame(df)
df.columns = ['ALGORITHM', 'CLUSTERS', 'MEAN', 'STD']
df.to_csv(saveName+".csv")
mean = []
std = []
plt.tight_layout()