alpha = 0.3
plt.rcParams['axes.prop_cycle'] = cycler('color', ACTIVE_COLORS)
plt.rcParams['text.color'] = LINE_COLOR
plt.rcParams['patch.edgecolor'] = LINE_COLOR
plt.rcParams['patch.facecolor'] = FILL_COLOR
plt.rcParams['axes.facecolor'] = my_palette['white']
plt.rcParams['axes.edgecolor'] = my_palette['grey']
plt.rcParams['axes.labelcolor'] = my_palette['grey']
plt.rcParams['xtick.color'] = my_palette['grey']
plt.rcParams['ytick.color'] = my_palette['grey']
plt.rcParams['grid.color'] = my_palette['light grey']
plt.rcParams['boxplot.boxprops.color'] = FILL_COLOR
plt.rcParams['boxplot.capprops.color'] = LINE_COLOR
plt.rcParams['boxplot.flierprops.color'] = my_palette['pink']
plt.rcParams['boxplot.flierprops.markeredgecolor'] = FILL_COLOR
plt.rcParams['boxplot.flierprops.markerfacecolor'] = FILL_COLOR
plt.rcParams['boxplot.whiskerprops.color'] = LINE_COLOR
plt.rcParams['boxplot.meanprops.color'] = my_palette['purple']
plt.rcParams['boxplot.meanprops.markeredgecolor'] = my_palette['purple']
plt.rcParams['boxplot.meanprops.markerfacecolor'] = my_palette['purple']
plt.rcParams['boxplot.medianprops.color'] = my_palette['green']
plt.rcParams['axes.prop_cycle'] = cycler('color', ACTIVE_COLORS)
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], accuracies, ylabel='Accuracy')
plt.suptitle('HFCR Accuracy Comparison')
plt.savefig(graphsDir + 'HFCR Accuracy Comparison')
graphsDir = './Results/Pattern Mining/'
if not os.path.exists(graphsDir):
os.makedirs(graphsDir)
bin_strategies = ['Uniform', 'Quantile', 'Kmeans']
n_bins = [3, 5, 10]
values = {
'with 3 bins': [18200, 20500, 20000],
'with 5 bins': [18200, 12000, 18000],
'with 10 bins': [12500, 5400, 10500],
}
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(bin_strategies, values, ylabel='Number of Patterns')
plt.suptitle('Number of Patterns for a support of 0.01')
plt.savefig(graphsDir + 'HFCR Number of Patterns')
values = {
'with 3 bins': [30, 59, 29],
'with 5 bins': [45, 81, 40],
'with 10 bins': [52, 50, 40],
}
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(bin_strategies, values, ylabel='Lift Score')
plt.suptitle('Lift of the top 10% for a support of 0.01')
plt.savefig(graphsDir + 'HFCR Lift Score')
bin_strategies = ['0.25', '0.57']
min_class = target_count.idxmin()
ind_min_class = target_count.index.get_loc(min_class)
print('Minority class:', target_count[ind_min_class])
print('Majority class:', target_count[1-ind_min_class])
print('Proportion:', round(target_count[ind_min_class] / target_count[1-ind_min_class], 2), ': 1')
RANDOM_STATE = 42
values = {'Original': [target_count.values[ind_min_class], target_count.values[1-ind_min_class]]}
df_class_min = unbal[unbal['DEATH_EVENT'] == min_class]
df_class_max = unbal[unbal['DEATH_EVENT'] != min_class]
df_under = df_class_max.sample(len(df_class_min))
values['UnderSample'] = [target_count.values[ind_min_class], len(df_under)]
df_over = df_class_min.sample(len(df_class_max), replace=True)
values['OverSample'] = [len(df_over), target_count.values[1-ind_min_class]]
smote = SMOTE(sampling_strategy='minority', random_state=RANDOM_STATE)
y = unbal.pop('DEATH_EVENT').values
X = unbal.values
smote_X, smote_y = smote.fit_sample(X, y)
smote_target_count = pd.Series(smote_y).value_counts()
values['SMOTE'] = [smote_target_count.values[ind_min_class], smote_target_count.values[1-ind_min_class]]
fig = plt.figure()
ds.multiple_bar_chart([target_count.index[ind_min_class], target_count.index[1-ind_min_class]], values,
title='Target', xlabel='frequency', ylabel='Class balance')
plt.savefig(graphsDir + 'HFCR Balancing - Class Balanced')
y: np.ndarray = data.pop('DEATH_EVENT').values
X: np.ndarray = data.values
labels: np.ndarray = pd.unique(y)
skf = StratifiedKFold(n_splits=n_splits, shuffle=True)
splitIterator = iter(skf.split(X, y))
splitCounter = 1
for model in splitIterator:
trnX = X[model[0]]
trnY = y[model[0]]
tstX = X[model[1]]
tstY = y[model[1]]
values['Train Split ' + str(splitCounter)] = [
len(np.delete(trnY, np.argwhere(trnY == negative))),
len(np.delete(trnY, np.argwhere(trnY == positive)))
]
values['Test Split ' + str(splitCounter)] = [
len(np.delete(tstY, np.argwhere(tstY == negative))),
len(np.delete(tstY, np.argwhere(tstY == positive)))
]
splitCounter += 1
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart([positive, negative],
values,
title='Data distribution per dataset')
plt.suptitle('HFCR Training Strategies')
plt.savefig(graphsDir + 'HFCR Training Strategies')
centers = ds.compute_centroids(data, labels)
mse.append(ds.compute_mse(data.values, labels, centers))
mae.append(ds.compute_mae(data.values, labels, centers))
sc.append(silhouette_score(data, labels))
db.append(davies_bouldin_score(data, labels))
ds.plot_clusters(data, eixo_x, eixo_y, labels, centers, k,
f'Hierarchical k={k} metric={m} link={link}', ax=axs[i,j])
values_mse[m] = mse
values_mae[m] = mae
values_sc[m] = sc
values_db[m] = db
plt.suptitle('QOT Clustering - Metric (Hierarchical) after PCA')
plt.savefig(subDir + 'QOT Clustering - Metric (Hierarchical) after PCA')
print('QOT Clustering - Metric (Hierarchical) MSE vs MAE vs SC vs DB after PCA')
_, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.multiple_bar_chart(LINKS, values_mse, title=f'Hierarchical MSE', xlabel='metric', ylabel='MSE', ax=ax[0, 0])
ds.multiple_bar_chart(LINKS, values_mae, title=f'Hierarchical MAE', xlabel='metric', ylabel='MAE', ax=ax[0, 1])
ds.multiple_bar_chart(LINKS, values_sc, title=f'Hierarchical SC', xlabel='metric', ylabel='SC', ax=ax[0, 2], percentage=True)
ds.multiple_bar_chart(LINKS, values_db, title=f'Hierarchical DB', xlabel='metric', ylabel='DB', ax=ax[0, 3])
plt.suptitle('QOT Clustering - Metric (Hierarchical) MSE vs MAE vs SC vs DB after PCA')
plt.savefig(subDir + 'QOT Clustering - Metric (Hierarchical) MSE vs MAE vs SC vs DB after PCA')
count += 1
plt.close("all")
plt.clf()
features_file.close()
ax=axs[0, k],
title='Overfitting for dist = %s' % (d),
xlabel='K Neighbours',
ylabel='accuracy',
percentage=True)
plt.suptitle('QOT Overfitting - KNN')
plt.savefig(subDir + 'QOT Overfitting - KNN')
clf = knn = KNeighborsClassifier(n_neighbors=best[0], metric=best[1])
clf.fit(trnX, trnY)
prd_trn = clf.predict(trnX)
prd_tst = clf.predict(tstX)
ds.plot_evaluation_results(["negative", "positive"], trnY, prd_trn,
tstY, prd_tst)
plt.suptitle('QOT KNN - ' + key +
'- Performance & Confusion matrix - %d neighbors and %s' %
(best[0], best[1]))
plt.savefig(subDir + 'QOT KNN - ' + key +
' - Performance & Confusion matrix')
plt.close("all")
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], best_accuracies, ylabel='Accuracy')
plt.suptitle('QOT Sampling & Feature Selection')
plt.savefig(graphsDir + 'QOT Sampling & Feature Selection')
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], recalls, ylabel='Recall')
plt.suptitle('QOT Recall Comparison')
plt.savefig(graphsDir + 'QOT Recall Comparison')
break
if (last_accuracy > best_accuracy and best_accuracy != -1):
best_accuracy = last_accuracy
last_accuracy = -1
count += offset
offset -= 1
elif (best_accuracy == -1):
best_accuracy = last_accuracy
count += 1
else:
count += 1
offset -= 1
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'],
values_by_criteria,
ylabel='Accuracy')
plt.suptitle('HFCR Gradient Boosting Criteria')
plt.savefig(subDir + 'HFCR Gradient Boosting Criteria')
plt.style.use('dslabs.mplstyle')
my_palette = {
'yellow': '#ECD474',
'pale orange': '#E9AE4E',
'salmon': '#E2A36B',
'orange': '#F79522',
'dark orange': '#D7725E',
'pale acqua': '#92C4AF',
'acqua': '#64B29E',
'marine': '#3D9EA9',
plt.rcParams['boxplot.boxprops.color'] = FILL_COLOR
plt.rcParams['boxplot.capprops.color'] = LINE_COLOR
plt.rcParams['boxplot.flierprops.color'] = my_palette['pink']
plt.rcParams['boxplot.flierprops.markeredgecolor'] = FILL_COLOR
plt.rcParams['boxplot.flierprops.markerfacecolor'] = FILL_COLOR
plt.rcParams['boxplot.whiskerprops.color'] = LINE_COLOR
plt.rcParams['boxplot.meanprops.color'] = my_palette['purple']
plt.rcParams['boxplot.meanprops.markeredgecolor'] = my_palette['purple']
plt.rcParams['boxplot.meanprops.markerfacecolor'] = my_palette['purple']
plt.rcParams['boxplot.medianprops.color'] = my_palette['green']
plt.rcParams['axes.prop_cycle'] = cycler('color', ACTIVE_COLORS)
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], accuracies, ylabel='Accuracy')
plt.suptitle('HFCR Accuracy Comparison')
plt.savefig(graphsDir + 'HFCR Accuracy Comparison')
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], recalls, ylabel='Recall')
plt.suptitle('HFCR Recall Comparison')
plt.savefig(graphsDir + 'HFCR Recall Comparison')
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], specificities, ylabel='Specificity')
plt.suptitle('HFCR Specificity Comparison')
plt.savefig(graphsDir + 'HFCR Specificity Comparison')
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], precisions, ylabel='Precision')
def pca_function(data, subDir):
data.pop('DEATH_EVENT')
variables = data.columns.values
eixo_x = 0
eixo_y = 4
eixo_z = 7
plt.figure()
plt.xlabel(variables[eixo_y])
plt.ylabel(variables[eixo_z])
plt.scatter(data.iloc[:, eixo_y], data.iloc[:, eixo_z])
print('HFCR Feature Extraction - PCA')
mean = (data.mean(axis=0)).tolist()
centered_data = data - mean
cov_mtx = centered_data.cov()
eigvals, eigvecs = np.linalg.eig(cov_mtx)
pca = PCA()
pca.fit(centered_data)
PC = pca.components_
var = pca.explained_variance_
# PLOT EXPLAINED VARIANCE RATIO
fig = plt.figure(figsize=(4, 4))
plt.title('Explained variance ratio')
plt.xlabel('PC')
plt.ylabel('ratio')
x_values = [str(i) for i in range(1, len(pca.components_) + 1)]
bwidth = 0.5
ax = plt.gca()
ax.set_xticklabels(x_values)
ax.set_ylim(0.0, 1.0)
ax.bar(x_values, pca.explained_variance_ratio_, width=bwidth)
ax.plot(pca.explained_variance_ratio_)
for i, v in enumerate(pca.explained_variance_ratio_):
ax.text(i, v + 0.05, f'{v*100:.1f}', ha='center', fontweight='bold')
plt.suptitle('HFCR Feature Extraction - PCA')
plt.savefig(subDir + 'HFCR Feature Extraction - PCA')
print('HFCR Feature Extraction - PCA 2')
transf = pca.transform(data)
_, axs = plt.subplots(1, 2, figsize=(2 * 5, 1 * 5), squeeze=False)
axs[0, 0].set_xlabel(variables[eixo_y])
axs[0, 0].set_ylabel(variables[eixo_z])
axs[0, 0].scatter(data.iloc[:, eixo_y], data.iloc[:, eixo_z])
axs[0, 1].set_xlabel('PC1')
axs[0, 1].set_ylabel('PC2')
axs[0, 1].scatter(transf[:, 0], transf[:, 1])
plt.suptitle('HFCR Feature Extraction - PCA')
plt.savefig(subDir + 'HFCR Feature Extraction - PCA')
print('Clustering after PCA')
data = pd.DataFrame(transf[:, :2], columns=['PC1', 'PC2'])
eixo_x = 0
eixo_y = 1
N_CLUSTERS = [2, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29]
rows, cols = ds.choose_grid(len(N_CLUSTERS))
print('HFCR Clustering - K-Means after PCA')
mse: list = []
mae: list = []
sc: list = []
db: list = []
_, axs = plt.subplots(rows,
cols,
figsize=(cols * 5, rows * 5),
squeeze=False)
i, j = 0, 0
for n in range(len(N_CLUSTERS)):
k = N_CLUSTERS[n]
estimator = KMeans(n_clusters=k)
estimator.fit(data)
mse.append(estimator.inertia_)
mae.append(
ds.compute_mae(data.values, estimator.labels_,
estimator.cluster_centers_))
sc.append(silhouette_score(data, estimator.labels_))
db.append(davies_bouldin_score(data, estimator.labels_))
ds.plot_clusters(data,
eixo_x,
eixo_y,
estimator.labels_.astype(float),
estimator.cluster_centers_,
k,
f'KMeans k={k}',
ax=axs[i, j])
i, j = (i + 1, 0) if (n + 1) % cols == 0 else (i, j + 1)
plt.suptitle('HFCR Clustering - K-Means after PCA')
plt.savefig(subDir + 'HFCR Clustering - K-Means after PCA')
print(
'HFCR Clustering - K-Means after PCA MSE vs MAE vs SC vs DB after PCA')
fig, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.plot_line(N_CLUSTERS,
mse,
title='KMeans MSE',
xlabel='k',
ylabel='MSE',
ax=ax[0, 0])
ds.plot_line(N_CLUSTERS,
mae,
title='KMeans MAE',
xlabel='k',
ylabel='MAE',
ax=ax[0, 1])
ds.plot_line(N_CLUSTERS,
sc,
title='KMeans SC',
xlabel='k',
ylabel='SC',
ax=ax[0, 2],
percentage=True)
ds.plot_line(N_CLUSTERS,
db,
title='KMeans DB',
xlabel='k',
ylabel='DB',
ax=ax[0, 3])
plt.suptitle(
'HFCR Clustering - K-Means after PCA MSE vs MAE vs SC vs DB after PCA')
plt.savefig(
subDir +
'HFCR Clustering - K-Means after PCA MSE vs MAE vs SC vs DB after PCA')
print('HFCR Clustering - Expectation-Maximization after PCA')
mse: list = []
mae: list = []
sc: list = []
db: list = []
_, axs = plt.subplots(rows,
cols,
figsize=(cols * 5, rows * 5),
squeeze=False)
i, j = 0, 0
for n in range(len(N_CLUSTERS)):
k = N_CLUSTERS[n]
estimator = GaussianMixture(n_components=k)
estimator.fit(data)
labels = estimator.predict(data)
mse.append(ds.compute_mse(data.values, labels, estimator.means_))
mae.append(ds.compute_mae(data.values, labels, estimator.means_))
sc.append(silhouette_score(data, labels))
db.append(davies_bouldin_score(data, labels))
ds.plot_clusters(data,
eixo_x,
eixo_y,
labels.astype(float),
estimator.means_,
k,
f'EM k={k}',
ax=axs[i, j])
i, j = (i + 1, 0) if (n + 1) % cols == 0 else (i, j + 1)
plt.suptitle('HFCR Clustering - Expectation-Maximization after PCA')
plt.savefig(subDir +
'HFCR Clustering - Expectation-Maximization after PCA')
print(
'HFCR Clustering - Expectation-Maximization MSE vs MAE vs SC vs DB after PCA'
)
fig, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.plot_line(N_CLUSTERS,
mse,
title='EM MSE',
xlabel='k',
ylabel='MSE',
ax=ax[0, 0])
ds.plot_line(N_CLUSTERS,
mae,
title='EM MAE',
xlabel='k',
ylabel='MAE',
ax=ax[0, 1])
ds.plot_line(N_CLUSTERS,
sc,
title='EM SC',
xlabel='k',
ylabel='SC',
ax=ax[0, 2],
percentage=True)
ds.plot_line(N_CLUSTERS,
db,
title='EM DB',
xlabel='k',
ylabel='DB',
ax=ax[0, 3])
plt.suptitle(
'HFCR Clustering - Expectation-Maximization MSE vs MAE vs SC vs DB after PCA'
)
plt.savefig(
subDir +
'HFCR Clustering - Expectation-Maximization MSE vs MAE vs SC vs DB after PCA'
)
print('HFCR Clustering - EPS (Density-based) after PCA')
EPS = [2.5, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
mse: list = []
mae: list = []
sc: list = []
db: list = []
rows, cols = ds.choose_grid(len(EPS))
_, axs = plt.subplots(rows,
cols,
figsize=(cols * 5, rows * 5),
squeeze=False)
i, j = 0, 0
for n in range(len(EPS)):
estimator = DBSCAN(eps=EPS[n], min_samples=2)
estimator.fit(data)
labels = estimator.labels_
k = len(set(labels)) - (1 if -1 in labels else 0)
if k > 1:
centers = ds.compute_centroids(data, labels)
mse.append(ds.compute_mse(data.values, labels, centers))
mae.append(ds.compute_mae(data.values, labels, centers))
sc.append(silhouette_score(data, labels))
db.append(davies_bouldin_score(data, labels))
ds.plot_clusters(data,
eixo_x,
eixo_y,
labels.astype(float),
estimator.components_,
k,
f'DBSCAN eps={EPS[n]} k={k}',
ax=axs[i, j])
i, j = (i + 1, 0) if (n + 1) % cols == 0 else (i, j + 1)
else:
mse.append(0)
mae.append(0)
sc.append(0)
db.append(0)
plt.suptitle('HFCR Clustering - EPS (Density-based) after PCA')
plt.savefig(subDir + 'HFCR Clustering - EPS (Density-based) after PCA')
print(
'HFCR Clustering - EPS (Density-based) MSE vs MAE vs SC vs DB after PCA'
)
fig, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.plot_line(EPS,
mse,
title='DBSCAN MSE',
xlabel='eps',
ylabel='MSE',
ax=ax[0, 0])
ds.plot_line(EPS,
mae,
title='DBSCAN MAE',
xlabel='eps',
ylabel='MAE',
ax=ax[0, 1])
ds.plot_line(EPS,
sc,
title='DBSCAN SC',
xlabel='eps',
ylabel='SC',
ax=ax[0, 2],
percentage=True)
ds.plot_line(EPS,
db,
title='DBSCAN DB',
xlabel='eps',
ylabel='DB',
ax=ax[0, 3])
plt.suptitle(
'HFCR Clustering - EPS (Density-based) MSE vs MAE vs SC vs DB after PCA'
)
plt.savefig(
subDir +
'HFCR Clustering - EPS (Density-based) MSE vs MAE vs SC vs DB after PCA'
)
print('HFCR Clustering - Metric (Density-based) after PCA')
METRICS = ['euclidean', 'cityblock', 'chebyshev', 'cosine', 'jaccard']
distances = []
for m in METRICS:
dist = np.mean(np.mean(squareform(pdist(data.values, metric=m))))
distances.append(dist)
print('AVG distances among records', distances)
distances[0] = 80
distances[1] = 50
distances[2] = 80
distances[3] = 0.0005
distances[4] = 0.0009
print('CHOSEN EPS', distances)
mse: list = []
mae: list = []
sc: list = []
db: list = []
rows, cols = ds.choose_grid(len(METRICS))
_, axs = plt.subplots(rows,
cols,
figsize=(cols * 5, rows * 5),
squeeze=False)
i, j = 0, 0
for n in range(len(METRICS)):
estimator = DBSCAN(eps=distances[n], min_samples=2, metric=METRICS[n])
estimator.fit(data)
labels = estimator.labels_
k = len(set(labels)) - (1 if -1 in labels else 0)
if k > 1:
centers = ds.compute_centroids(data, labels)
mse.append(ds.compute_mse(data.values, labels, centers))
mae.append(ds.compute_mae(data.values, labels, centers))
sc.append(silhouette_score(data, labels))
db.append(davies_bouldin_score(data, labels))
ds.plot_clusters(
data,
eixo_x,
eixo_y,
labels.astype(float),
estimator.components_,
k,
f'DBSCAN metric={METRICS[n]} eps={distances[n]:.2f} k={k}',
ax=axs[i, j])
else:
print(k)
mse.append(0)
mae.append(0)
sc.append(0)
db.append(0)
i, j = (i + 1, 0) if (n + 1) % cols == 0 else (i, j + 1)
plt.suptitle('HFCR Clustering - Metric (Density-based) after PCA')
plt.savefig(subDir + 'HFCR Clustering - Metric (Density-based) after PCA')
print(
'HFCR Clustering - Metric (Density-based) MSE vs MAE vs SC vs DB after PCA'
)
fig, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.bar_chart(METRICS,
mse,
title='DBSCAN MSE',
xlabel='metric',
ylabel='MSE',
ax=ax[0, 0])
ds.bar_chart(METRICS,
mae,
title='DBSCAN MAE',
xlabel='metric',
ylabel='MAE',
ax=ax[0, 1])
ds.bar_chart(METRICS,
sc,
title='DBSCAN SC',
xlabel='metric',
ylabel='SC',
ax=ax[0, 2],
percentage=True)
ds.bar_chart(METRICS,
db,
title='DBSCAN DB',
xlabel='metric',
ylabel='DB',
ax=ax[0, 3])
plt.suptitle(
'HFCR Clustering - Metric (Density-based) MSE vs MAE vs SC vs DB after PCA'
)
plt.savefig(
subDir +
'HFCR Clustering - Metric (Density-based) MSE vs MAE vs SC vs DB after PCA'
)
print('HFCR Clustering - Hierarchical after PCA')
mse: list = []
mae: list = []
sc: list = []
db: list = []
rows, cols = ds.choose_grid(len(N_CLUSTERS))
_, axs = plt.subplots(rows,
cols,
figsize=(cols * 5, rows * 5),
squeeze=False)
i, j = 0, 0
for n in range(len(N_CLUSTERS)):
k = N_CLUSTERS[n]
estimator = AgglomerativeClustering(n_clusters=k)
estimator.fit(data)
labels = estimator.labels_
centers = ds.compute_centroids(data, labels)
mse.append(ds.compute_mse(data.values, labels, centers))
mae.append(ds.compute_mae(data.values, labels, centers))
sc.append(silhouette_score(data, labels))
db.append(davies_bouldin_score(data, labels))
ds.plot_clusters(data,
eixo_x,
eixo_y,
labels,
centers,
k,
f'Hierarchical k={k}',
ax=axs[i, j])
i, j = (i + 1, 0) if (n + 1) % cols == 0 else (i, j + 1)
plt.suptitle('HFCR Clustering - Hierarchical after PCA')
plt.savefig(subDir + 'HFCR Clustering - Hierarchical after PCA')
print('HFCR Clustering - Hierarchical MSE vs MAE vs SC vs DB after PCA')
fig, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.plot_line(N_CLUSTERS,
mse,
title='Hierarchical MSE',
xlabel='k',
ylabel='MSE',
ax=ax[0, 0])
ds.plot_line(N_CLUSTERS,
mae,
title='Hierarchical MAE',
xlabel='k',
ylabel='MAE',
ax=ax[0, 1])
ds.plot_line(N_CLUSTERS,
sc,
title='Hierarchical SC',
xlabel='k',
ylabel='SC',
ax=ax[0, 2],
percentage=True)
ds.plot_line(N_CLUSTERS,
db,
title='Hierarchical DB',
xlabel='k',
ylabel='DB',
ax=ax[0, 3])
plt.suptitle(
'HFCR Clustering - Hierarchical MSE vs MAE vs SC vs DB after PCA')
plt.savefig(
subDir +
'HFCR Clustering - Hierarchical MSE vs MAE vs SC vs DB after PCA')
print('HFCR Clustering - Metric (Hierarchical) after PCA')
METRICS = ['euclidean', 'cityblock', 'chebyshev', 'cosine', 'jaccard']
LINKS = ['complete', 'average']
k = 3
values_mse = {}
values_mae = {}
values_sc = {}
values_db = {}
rows = len(METRICS)
cols = len(LINKS)
_, axs = plt.subplots(rows,
cols,
figsize=(cols * 5, rows * 5),
squeeze=False)
for i in range(len(METRICS)):
mse: list = []
mae: list = []
sc: list = []
db: list = []
m = METRICS[i]
for j in range(len(LINKS)):
link = LINKS[j]
estimator = AgglomerativeClustering(n_clusters=k,
linkage=link,
affinity=m)
estimator.fit(data)
labels = estimator.labels_
centers = ds.compute_centroids(data, labels)
mse.append(ds.compute_mse(data.values, labels, centers))
mae.append(ds.compute_mae(data.values, labels, centers))
sc.append(silhouette_score(data, labels))
db.append(davies_bouldin_score(data, labels))
ds.plot_clusters(data,
eixo_x,
eixo_y,
labels,
centers,
k,
f'Hierarchical k={k} metric={m} link={link}',
ax=axs[i, j])
values_mse[m] = mse
values_mae[m] = mae
values_sc[m] = sc
values_db[m] = db
plt.suptitle('HFCR Clustering - Metric (Hierarchical) after PCA')
plt.savefig(subDir + 'HFCR Clustering - Metric (Hierarchical) after PCA')
print(
'HFCR Clustering - Metric (Hierarchical) MSE vs MAE vs SC vs DB after PCA'
)
_, ax = plt.subplots(1, 4, figsize=(10, 3), squeeze=False)
ds.multiple_bar_chart(LINKS,
values_mse,
title=f'Hierarchical MSE',
xlabel='metric',
ylabel='MSE',
ax=ax[0, 0])
ds.multiple_bar_chart(LINKS,
values_mae,
title=f'Hierarchical MAE',
xlabel='metric',
ylabel='MAE',
ax=ax[0, 1])
ds.multiple_bar_chart(LINKS,
values_sc,
title=f'Hierarchical SC',
xlabel='metric',
ylabel='SC',
ax=ax[0, 2],
percentage=True)
ds.multiple_bar_chart(LINKS,
values_db,
title=f'Hierarchical DB',
xlabel='metric',
ylabel='DB',
ax=ax[0, 3])
plt.suptitle(
'HFCR Clustering - Metric (Hierarchical) MSE vs MAE vs SC vs DB after PCA'
)
plt.savefig(
subDir +
'HFCR Clustering - Metric (Hierarchical) MSE vs MAE vs SC vs DB after PCA'
)
plt.close("all")
plt.clf()
plt.rcParams['boxplot.medianprops.color'] = my_palette['green']
plt.rcParams['axes.prop_cycle'] = cycler('color', ACTIVE_COLORS)
graphsDir = './Results/Recalls/Gradient Boosting/'
if not os.path.exists(graphsDir):
os.makedirs(graphsDir)
recalls = {
'Original': [1, 0.7368],
' - No Outliers - Original': [1, 0.7368],
' - Scaling - Original': [1, 0.7368],
' - Scaling & Feature Selection - Original': [1, 0.7368],
'UnderSample': [1, 0.7368],
' - No Outliers - UnderSample': [1, 0.6842],
' - No Outliers & Scaling - UnderSample': [1, 0.7368],
' - No Outliers & Feature Selection - UnderSample': [1, 0.7368],
'OverSample': [1, 0.6842],
' - No Outliers - OverSample': [1, 0.6842],
' - Scaling - OverSample': [1, 0.7368],
' - Scaling & Feature Selection - OverSample': [1, 0.7368],
'SMOTE': [1, 0.7368],
' - No Outliers - SMOTE': [1, 0.7368],
' - No Outliers & Scaling - SMOTE': [1, 0.6842],
' - No Outliers & Feature Selection - SMOTE': [1, 0.75],
}
plt.figure(figsize=(7, 7))
ds.multiple_bar_chart(['Train', 'Test'], recalls, ylabel='Recall')
plt.suptitle('HFCR Recall Comparison')
plt.savefig(graphsDir + 'HFCR Recall Comparison')
