def fig16_4():
results_dir = get_project_results_dir()
init_methods = [
'Rand-P', 'Rand-C', 'Maxmin', 'kmeans++', 'Bradley', 'Sorting',
'Projection', 'Luxburg', 'Split'
]
pu.figure_setup()
fig_size = pu.get_fig_size(15, 7)
fig = plt.figure(figsize=(fig_size))
ax = fig.add_subplot()
ax.title.set_text(f'Dataset G2 (CI final)')
ax.set_prop_cycle(color=plt.cm.Set1.colors)
dimensions_sizes = [
'1', '2', '4', '8', '16', '32', '64', '128', '256', '512', '1024'
]
std_sizes = list(range(10, 101, 10))
for init_method in init_methods:
final_percentages_means = []
for dimensions_size in dimensions_sizes:
final_percentages = []
for std_size in std_sizes:
df = pd.read_csv(results_dir /
f"g2-{dimensions_size}-{std_size}" /
f"{init_method}.csv")
n_rows = df.shape[0]
final_zeros = df['ci_final'].value_counts().get(0, 0)
final_percentage = (final_zeros / n_rows) * 100
final_percentages.append(final_percentage)
final_percentages_means.append(np.mean(final_percentages))
label = "KMeansPP" if init_method == "kmeans++" else init_method
ax.plot(dimensions_sizes, final_percentages_means, label=label)
ax.set_xticks(dimensions_sizes)
ax.set_xlabel('Dimensões')
ax.set_yticks([99, 99.20, 99.40, 99.60, 99.80, 100])
ax.set_yticklabels(
['99\%', '99,2\%', '99,4\%', '99,6\%', '99,8\%', '100\%'])
ax.set_ylabel('Taxa de sucesso (\%)')
plt.legend(bbox_to_anchor=(1, 1), loc="upper left")
plt.tight_layout()
filename = get_project_results_dir().joinpath('fig16_4.eps')
return fig, str(filename)
def fig16_1():
results_dir = get_project_results_dir()
init_methods = [
'Rand-P', 'Rand-C', 'Maxmin', 'kmeans++', 'Bradley', 'Sorting',
'Projection', 'Luxburg', 'Split'
]
pu.figure_setup()
fig_size = pu.get_fig_size(15, 7)
fig = plt.figure(figsize=(fig_size))
ax = fig.add_subplot()
ax.title.set_text(f'Dataset DIM (CI inicial)')
ax.set_prop_cycle(color=plt.cm.Set1.colors)
dimensions_sizes = ['32', '64', '128', '256', '512', '1024']
for init_method in init_methods:
initial_percentages = []
for dimensions_size in dimensions_sizes:
df = pd.read_csv(results_dir / f"dim{dimensions_size}" /
f"{init_method}.csv")
n_rows = df.shape[0]
initial_zeros = df['ci_initial'].value_counts().get(0, 0)
initial_percentage = (initial_zeros / n_rows) * 100
initial_percentages.append(initial_percentage)
label = "KMeansPP" if init_method == "kmeans++" else init_method
ax.plot(dimensions_sizes, initial_percentages, label=label)
ax.set_xticks(dimensions_sizes)
ax.set_xlabel('Dimensões')
ax.set_yticks([0, 20, 40, 60, 80, 100])
ax.set_yticklabels(['0\%', '20\%', '40\%', '60\%', '80\%', '100\%'])
ax.set_ylabel('Taxa de sucesso (\%)')
plt.legend(bbox_to_anchor=(1, 1), loc="upper left")
plt.tight_layout()
filename = get_project_results_dir().joinpath('fig16_1.eps')
return fig, str(filename)
def plot_hq_mtx(parameters_dict):
X = parameters_dict['X']
y = parameters_dict['y']
k_lst = parameters_dict['k_lst']
p_lst = parameters_dict['p_lst']
measures_lst = parameters_dict['measures_lst']
dataset_name = parameters_dict['dataset_name']
target_names = parameters_dict['target_names']
pu.figure_setup()
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1, shuffle=True, stratify=y)
fig_size = pu.get_fig_size(15, 4.4)
fig = plt.figure(figsize=(fig_size))
fig.suptitle(f'Dataset: {dataset_name.upper()}')
for i in range(3):
ax = fig.add_subplot(1,3,i+1)
ax.set_axisbelow(True)
curr_name = measures_lst[i]
pg = LVQ(prototypes_number=35, version=curr_name)
s_set = pg.generate(X_train,y_train)
classifier = Knn(n_neighbors=k_lst[-1]).fit(s_set[0], s_set[1])
plot_confusion_matrix(classifier, X_test, y_test,
display_labels=target_names,
ax=ax,
cmap=plt.cm.Blues,
normalize=None
)
ax.set_title(curr_name)
plt.tight_layout()
filename = get_project_results_dir().joinpath(dataset_name + '_cf_mtx.eps')
pu.save_fig(fig, str(filename))
def fig14():
results_dir = get_project_results_dir()
init_methods = [
'Rand-P', 'Rand-C', 'Maxmin', 'kmeans++', 'Bradley', 'Sorting',
'Projection', 'Luxburg', 'Split'
]
pu.figure_setup()
fig_size = pu.get_fig_size(15, 7)
fig = plt.figure(figsize=(fig_size))
ax = fig.add_subplot()
ax.set_prop_cycle(color=plt.cm.Set1.colors)
cluster_sizes = list(range(10, 101, 10))
for init_method in init_methods:
final_percentages = []
for cluster_size in cluster_sizes:
df = pd.read_csv(results_dir / f"b2-sub-{cluster_size}" /
f"{init_method}.csv")
n_rows = df.shape[0]
final_zeros = df['ci_final'].value_counts().get(0, 0)
final_percentage = (final_zeros / n_rows) * 100
final_percentages.append(final_percentage)
label = "KMeansPP" if init_method == "kmeans++" else init_method
ax.plot(cluster_sizes, final_percentages, label=label)
ax.set_xticks(cluster_sizes)
ax.set_xlabel('Clusters (k)')
ax.set_yticks([0, 20, 40, 60, 80, 100])
ax.set_yticklabels(['0\%', '20\%', '40\%', '60\%', '80\%', '100\%'])
ax.set_ylabel('Taxa de sucesso (\%)')
plt.legend(bbox_to_anchor=(1, 1), loc="upper left")
plt.tight_layout()
filename = get_project_results_dir().joinpath('fig14.eps')
return fig, str(filename)
def fig15():
results_dir = get_project_results_dir()
init_methods = [
'Rand-P', 'Rand-C', 'Maxmin', 'kmeans++', 'Bradley', 'Sorting',
'Projection', 'Luxburg', 'Split'
]
pu.figure_setup()
fig_size = pu.get_fig_size(15, 7)
fig = plt.figure(figsize=(fig_size))
ax = fig.add_subplot()
ax.set_prop_cycle(color=plt.cm.Set1.colors)
cluster_sizes = list(range(10, 101, 10))
for init_method in init_methods:
relative_cis = []
for cluster_size in cluster_sizes:
df = pd.read_csv(results_dir / f"b2-sub-{cluster_size}" /
f"{init_method}.csv")
n_rows = df.shape[0]
ci_mean = df['ci_final'].mean()
relative_cis.append(ci_mean / cluster_size)
label = "KMeansPP" if init_method == "kmeans++" else init_method
ax.plot(cluster_sizes, relative_cis, label=label)
ax.set_xticks(cluster_sizes)
ax.set_xlabel('Clusters (k)')
ax.set_ylabel('CI relativo (CI/k)')
plt.legend(bbox_to_anchor=(1, 1), loc="upper left")
plt.tight_layout()
filename = get_project_results_dir().joinpath('fig15.eps')
return fig, str(filename)
def plot_hq_summary_p(parameters_dict):
k_lst = parameters_dict['k_lst']
p_lst = parameters_dict['p_lst']
measures_lst = parameters_dict['measures_lst']
dataset_name = parameters_dict['dataset_name']
fmt = ['ro--','g^--','bs--']
pu.figure_setup()
fig_size = pu.get_fig_size(15, 6)
fig = plt.figure(figsize=(fig_size))
fig.suptitle(f'Dataset: {dataset_name.upper()}')
ax = fig.add_subplot(1,2,1)
ax.set_xlabel('Protótipos')
ax.set_ylabel('Tempo de Processamento (s)')
ax.set_axisbelow(True)
for i in range(len(fmt)):
curr_measure = f'{measures_lst[i]}-p'
curr_name = measures_lst[i]
ax.plot(
p_lst,
parameters_dict[curr_measure][0],
fmt[i],
markersize=1.5,
linewidth=0.5,
label=curr_name)
ax.set_xticks(p_lst)
plt.legend()
plt.tight_layout()
ax = fig.add_subplot(1,2,2)
ax.set_xlabel('Protótipos')
ax.set_ylabel('Acurácia')
ax.set_axisbelow(True)
for i in range(len(fmt)):
curr_measure = f'{measures_lst[i]}-p'
curr_name = measures_lst[i]
ax.plot(
p_lst,
parameters_dict[curr_measure][1],
fmt[i],
markersize=1.5,
linewidth=0.5,
label=curr_name)
ax.set_xticks(p_lst)
plt.legend()
plt.tight_layout()
filename = get_project_results_dir().joinpath(dataset_name + '_summary_p.eps')
pu.save_fig(fig, str(filename))
def fig13():
results_dir = get_project_results_dir()
init_methods = [
'Rand-P', 'Rand-C', 'Maxmin', 'kmeans++', 'Bradley', 'Sorting',
'Projection', 'Luxburg', 'Split'
]
init_methods_bar = [
'Rand-P', 'Rand-C', 'Maxmin', 'KMeansPP', 'Bradley', 'Sorting',
'Projection', 'Luxburg', 'Split'
]
high_overlap_datasets = [
'g2-2-40',
'g2-2-50',
'g2-2-60',
'g2-2-70',
'g2-2-80',
'g2-2-90',
'g2-2-100',
'g2-4-50',
'g2-4-60',
'g2-4-70',
'g2-4-80',
'g2-4-90',
'g2-4-100',
'g2-8-70',
'g2-8-80',
'g2-8-90',
'g2-8-100',
'g2-16-90',
'g2-16-100',
]
pu.figure_setup()
fig_size = pu.get_fig_size(15, 7)
fig = plt.figure(figsize=(fig_size))
axs = fig.subplots(ncols=2)
axs[0].title.set_text(f'Baixa sobreposição')
initial = []
final = []
for init_method in init_methods:
initial_percentages = []
final_percentages = []
for dataset in results_dir.glob('g2*'):
if dataset.stem in high_overlap_datasets:
continue
df = pd.read_csv(dataset / f"{init_method}.csv")
n_rows = df.shape[0]
initial_zeros = df['ci_initial'].value_counts().get(0, 0)
initial_percentage = (initial_zeros / n_rows) * 100
initial_percentages.append(initial_percentage)
final_zeros = df['ci_final'].value_counts().get(0, 0)
final_percentage = (final_zeros / n_rows) * 100
final_percentages.append(final_percentage)
initial.append(np.mean(initial_percentages))
final.append(np.mean(final_percentages) - np.mean(initial_percentages))
# axs[0].bar(init_methods_bar, initial, label='Inicial')
axs[0].bar(init_methods_bar,
initial,
label='Inicial',
color='gray',
edgecolor='black')
axs[0].bar(init_methods_bar,
final,
bottom=initial,
label='Final',
color='white',
edgecolor='black')
axs[0].set_ylabel('Taxa de sucesso (\%)')
axs[0].tick_params('x', labelrotation=70)
axs[0].set_ylim([0, 110])
axs[0].set_yticks([0, 20, 40, 60, 80, 100])
axs[0].set_yticklabels(['0\%', '20\%', '40\%', '60\%', '80\%', '100\%'])
axs[0].grid(b=False, axis='x')
axs[1].title.set_text(f'Alta sobreposição')
initial = []
final = []
for init_method in init_methods:
initial_percentages = []
final_percentages = []
for dataset in high_overlap_datasets:
df = pd.read_csv(results_dir / dataset / f"{init_method}.csv")
n_rows = df.shape[0]
initial_zeros = df['ci_initial'].value_counts().get(0, 0)
initial_percentage = (initial_zeros / n_rows) * 100
initial_percentages.append(initial_percentage)
final_zeros = df['ci_final'].value_counts().get(0, 0)
final_percentage = (final_zeros / n_rows) * 100
final_percentages.append(final_percentage)
initial.append(np.mean(initial_percentages))
final.append(np.mean(final_percentages) - np.mean(initial_percentages))
init_methods[3] = 'KMeansPP'
axs[1].bar(init_methods_bar,
initial,
label='Inicial',
color='gray',
edgecolor='black')
axs[1].bar(init_methods_bar,
final,
bottom=initial,
label='Final',
color='white',
edgecolor='black')
axs[1].tick_params('x', labelrotation=70)
axs[1].set_ylim([0, 110])
axs[1].set_yticks([0, 20, 40, 60, 80, 100])
axs[1].set_yticklabels(['0\%', '20\%', '40\%', '60\%', '80\%', '100\%'])
axs[1].grid(b=False, axis='x')
plt.legend()
plt.tight_layout()
filename = get_project_results_dir().joinpath('fig13.eps')
return fig, str(filename)