def run_SVC():
print('# ======================= SVC ========================= #')
svc = SVC(kernel='linear', C=1)
svc.fit(train_x, train_y)
pred_y = svc.predict(test_x)
score, confusion_matrix, report = report_results(test_y, np.array(pred_y), cmat_flag=True)
print()
def run_kNN():
print('# ======================= K-Nearest Neighbors ========================= #')
acc_list = []
for k in K_LIST:
knn = KNeighborsClassifier(n_neighbors=k)
knn.fit(train_x, train_y)
knn_pred_y = knn.predict(test_x)
print('k-neighbors: %d' % k)
acc, cmat, rpt = report_results(test_y, knn_pred_y)
acc_list.append(acc)
print()
def run_RandomForest():
print('# ======================= Random Forest ========================= #')
acc_list_rf = []
for n in [25]:
rand_forest = RandomForestClassifier(n_estimators=n)
rand_forest.fit(train_x, train_y)
rf_pred_y = rand_forest.predict(test_x)
print('Tree count: %d' % n)
acc, cm, rpt = report_results(test_y, rf_pred_y, cmat_flag=True)
acc_list_rf.append(acc)
task5(cm, 'Random Forest (25 trees)')
print()
# for img, lbl in zip(test_faces, test_labels):
# key_points, descriptors = extract_sift(img)
# pred_list = rand_forest.predict(descriptors)
# counts = np.bincount(pred_list) # bin counting for upcoming argmax
# pred = np.argmax(counts)
# pred_y.append(pred)
# test_y.append(lbl)
# score, confusion_matrix, report = report_results_face(test_y, pred_y)
# acc_list_rf.append(score)
# print('Tree count: %d \t Accuracy: %.3f' % (n, score))
# print()
print('# ======================= SVC ========================= #')
c = list(zip(train_x, train_y))
random.shuffle(c)
train_x, train_y = zip(*c)
svc = SVC(kernel='rbf', C=10, gamma=0.00001)
svc.fit(train_x, train_y)
test_y, pred_y = [], []
for img, lbl in zip(test_faces, test_labels):
key_points, descriptors = extract_sift(img)
pred_list = svc.predict(descriptors)
counts = np.bincount(pred_list) # bin counting for upcoming argmax
pred = np.argmax(counts)
pred_y.append(pred)
test_y.append(lbl)
score, confusion_matrix, report = report_results(test_y, np.array(pred_y), cmat_flag=True)
pred_gait = gait_model.predict_proba(gait_test_x)
pred_gait_avg = np.zeros((0, 10))
for i in range(10):
t2 = pred_gait[i * 8:(i + 1) * 8, :] # pick 8 rows
pred_gait_avg = np.vstack((pred_gait_avg, t2.mean(axis=0)))
alpha_list = [x / 10.0 for x in range(7, 8)]
acc_list = []
for alpha in alpha_list:
print('alpha=%.2f' % alpha)
acc_count = 0
pred_y = []
for i in range(
face_test_x.shape[1]): # loop through all test face images
true_y = face_test_y[i]
pf = pred_face[i, :] # face prediction vector of shape (10,)
gf = pred_gait_avg[
true_y, :] # gait prediction vector of shape (10,)
fused_pred_vector = alpha * pf + (1 - alpha) * gf
fused_pred = np.argmax(fused_pred_vector)
pred_y.append(fused_pred)
acc_count += 1 if fused_pred == true_y else 0
acc, cm, rpt = report_results(face_test_y, pred_y, cmat_flag=True)
task8(cm)
acc = acc_count / face_test_x.shape[1]
acc_list.append(acc)
print('\tAccuracy: %.2f' % acc)
# question5(alpha_list, acc_list)
# test_y, pred_y = [], []
# for img, lbl in zip(test_faces, test_labels):
# key_points, descriptors = extract_surf(img)
# pred_list = knn.predict(descriptors)
# counts = np.bincount(pred_list) # bin counting for upcoming argmax
# pred = np.argmax(counts)
# pred_y.append(pred)
# test_y.append(lbl)
# # knn_pred_y = knn.predict(test_x)
# print('k-neighbors: %d' % k)
# score, confusion_matrix, report = report_results_face(test_y, pred_y)
# acc_list.append(score)
# print()
print('# ======================= Random Forest ========================= #')
acc_list_rf = []
for n in N_ESTIMATORS:
rand_forest = RandomForestClassifier(n_estimators=n)
rand_forest.fit(train_x, train_y)
test_y, pred_y = [], []
for img, lbl in zip(test_faces, test_labels):
key_points, descriptors = extract_surf(img)
pred_list = rand_forest.predict(descriptors)
counts = np.bincount(pred_list) # bin counting for upcoming argmax
pred = np.argmax(counts)
pred_y.append(pred)
test_y.append(lbl)
score, confusion_matrix, report = report_results(test_y, pred_y)
acc_list_rf.append(score)
print('Tree count: %d \t Accuracy: %.3f' % (n, score))
print()
# # plt.savefig(os.path.join(RESULTS_HOME, 'question4_confusion_matrix_D_'+str(d)+'.png'), dpi=300)
# # # plt.show()
# question3(D_LIST, acc_list)
print(
'# ======================= K-Nearest Neighbors ========================= #'
)
acc_list = []
cmat_k5 = None
for k in K_LIST:
knn = KNeighborsClassifier(n_neighbors=k)
knn.fit(train_x.T, train_y)
knn_pred_y = knn.predict(test_x.T)
print('k-neighbors: %d' % k)
acc, cmat, rpt = report_results(test_y, knn_pred_y) if k != 5 \
else report_results(test_y, knn_pred_y, cmat_flag=True)
cmat_k5 = cmat if k == 5 else cmat_k5
acc_list.append(acc)
print()
print(
'# ======================= Random Forest ========================= #')
acc_list_rf = []
cmat_n100 = None
for n in N_ESTIMATORS:
rand_forest = RandomForestClassifier(n_estimators=n)
rand_forest.fit(train_x.T, train_y)
rf_pred_y = rand_forest.predict(test_x.T)
score = metrics.accuracy_score(test_y, rf_pred_y)
acc, cmat, rpt = report_results(test_y, rf_pred_y) if n != 100 \
clsfr_titles = ['RandomForest %d' % n for n in N_ESTIMATORS_LIST]
clsfr_titles.extend('kNearestNeighbors %d' % k for k in K_LIST)
clsfr_titles.append('DecisionTree default')
clsfr_titles.append('SupportVector rbf')
clsfr_titles.append('DecisionTree linear')
return clsfr, clsfr_titles
if __name__ == '__main__':
dataset = read_data(GAIT_DATA_HOME, ids, action, sensor)
# 1. GAIT visualizing
# task6(dataset[0])
# 2. Train classifiers
train_x, train_y, test_x, test_y = get_train_test_features(dataset)
print(train_x.shape, train_y.shape, test_x.shape, test_y.shape)
classifiers, classifiers_title = get_classifiers()
# Performing classification using each classifier
for i in range(len(classifiers)):
print('========================================================')
print('\tRunning:', classifiers_title[i])
print('========================================================')
classifiers[i].fit(train_x, train_y) # train
pred_y = classifiers[i].predict(test_x)
score, confusion_matrix, report = report_results(test_y, pred_y, cmat_flag=True)
np.save(os.path.join(RESULTS_HOME, 'gait', get_safe_string(classifiers_title[i]) + '_cmat.npy'),
confusion_matrix)
print(score)