def sort_skeletons(task_skeleton_data):
ids = map(lambda line: line[0][2], task_skeleton_data)
print 'skeleton id: ', Counter(ids).most_common()
skeletons = []
for counter_ids in Counter(ids).most_common():
skeleton_id = task_skeleton_data
main_id = counter_ids[0]
new_joints_points = []
for i_point, points in enumerate(skeleton_id):
if points[0][2] == main_id:
if len(new_joints_points) == 0: print points[0]
new_joints_points.append(points)
skeleton_id = new_joints_points
skeletons.append(skeleton_id)
skeleton_data_sorted = skeletons[1]
data_organizer.save_matrix_pickle(
skeleton_data_sorted,
'C:/Users/dario.dotti/Documents/pecs_data_review/skeletons_confusion_behavior_08082017_test.txt'
)
return skeleton_data_sorted
def create_cluster_labels_participant_task(raw_features, AE_weights_level_2):
cluster_model = data_organizer.load_matrix_pickle(
'C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/head_joint_id1/10_cluster_model_layer2_new.txt'
)
final_labels = []
for participant in raw_features:
participant_label = []
for task in participant:
task_label = []
for f_vector in task:
hd = np.dot(f_vector, AE_weights_level_2[0][0])
act = sigmoid_function(hd + AE_weights_level_2[1])
label = cluster_model.predict(act.reshape((1, -1)))
task_label.append(label[0])
participant_label.append(task_label)
final_labels.append(participant_label)
data_organizer.save_matrix_pickle(
final_labels,
'C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/head_joint_id1/10clusters_labels_l2_participants_tasks_new.txt'
)
def get_ambient_sensor_features():
#read and parse file with recorded data
with open('C:/Users/dario.dotti/Documents/file_to_analyze_AS_5_labels.txt','r') as f:
files = f.read().split('\n')
print 'number of recorded files: '+str(len(files))
matrix_allData_as = []
for file in files:
print file
activation_matrix = ambient_sensor_analysis.feature_extraction_as(file)
print np.array(activation_matrix).shape
if len(matrix_allData_as)>0:
matrix_allData_as = np.vstack((matrix_allData_as,activation_matrix))
else:
matrix_allData_as = activation_matrix
my_data_org.save_matrix_pickle(matrix_allData_as,'C:/Users/dario.dotti/Documents/AS_activation_5_labels_transformed.txt')
def main_pecs_data():
##get raw data for displaying
task_skeleton_data = data_organizer.load_matrix_pickle(
'C:/Users/dario.dotti/Documents/pecs_data_review/skeletons_repetitive_behavior_08082017.txt'
)
##if data contains multiple skeletons here I sort them cronologically
sort_skeletons(task_skeleton_data)
my_room = np.zeros((424, 512, 3), dtype=np.uint8)
my_room += 255
list_poly = img_processing.divide_image(my_room)
#draw_joints_and_tracks(task_skeleton_data,my_room)
#return 0
skeleton_data_in_time_slices = org_data_in_timeIntervals(
task_skeleton_data, [0, 0, 2])
HOT_data, patient_ID = histograms_of_oriented_trajectories(
list_poly, skeleton_data_in_time_slices)
data_organizer.save_matrix_pickle(
HOT_data,
'C:/Users/dario.dotti/Documents/pecs_data_review/HOT_repetitive_behavior_08082017.txt'
)
def video_clustering_fit(HOT_matrix, filename):
global my_ms
my_ms.fit(HOT_matrix)
if len(filename) > 2:
data_organizer.save_matrix_pickle(my_ms, filename)
filename = os.path.basename(file)
video_traj.set_subject(filename.split('_')[0])
traj_features = video_traj.feature_extraction_video_traj(file)
matrix_allData_HOT.append(traj_features[1])
# if len(matrix_allData_HOT)>0:
# matrix_allData_HOT = np.vstack((matrix_allData_HOT,traj_features[1]))
# else:
# matrix_allData_HOT = np.array(traj_features[1])
print len(matrix_allData_HOT)
#scipy.io.savemat('C:/Users/dario.dotti/Documents/hot_spatial_grid_4x4.mat',mdict={'spatial_grid_4x4': matrix_allData_HOT})
<<<<<<< HEAD
my_data_org.save_matrix_pickle(matrix_allData_HOT,'C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/skeleton_data_in_tasks_time_slices_30fps_ordered_1sec.txt')##C:/Users/dario.dotti/Desktop/data_recordings_master/master_skeleton_data_in_tasks_time_slices_30fps_1sec.txt
=======
#my_data_org.save_matrix_pickle(matrix_allData_HOT,'C:/Users/dario.dotti/Desktop/data_recordings_master/master_skeleton_data_in_tasks_time_slices_30fps.txt') #C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/skeleton_data_in_tasks_time_slices_30fps_ordered
>>>>>>> 9348384985d2847c272133ff77ce6181ca1fa082
def get_ambient_sensor_features():
#read and parse file with recorded data
with open('C:/Users/dario.dotti/Documents/file_to_analyze_AS_5_labels.txt','r') as f:
files = f.read().split('\n')
print 'number of recorded files: '+str(len(files))
matrix_allData_as = []
for file in files:
print file
final_matrix = []
final_orig_points = []
final_matrix_realcoord =[]
for participant in skeleton_data_in_tasks_and_time_slices:
#extract_traj_word_spatio_temporal_grid(participant, n_layer=1)
<<<<<<< HEAD
feature_participant,orig_point_participant = extract_traj_word_temporal_window(participant, n_layer=1)
=======
feature_participant,orig_point_participant,matrix_real_coord = extract_traj_word_temporal_window(participant, n_layer=2)
final_matrix_realcoord.append(matrix_real_coord)
>>>>>>> 9348384985d2847c272133ff77ce6181ca1fa082
final_matrix.append(feature_participant)
final_orig_points.append(orig_point_participant)
print len(final_matrix),len(final_orig_points)
final_matrix=final_matrix+final_matrix_realcoord
data_organizer.save_matrix_pickle(final_matrix, 'C:/Users/dario.dotti/Desktop/Hier_AE_deliverable/head_joint_id1/feature_matrix_participant_task_l2_new_realCoordinates.txt')
#data_organizer.save_matrix_pickle(final_orig_points,
# 'C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/head_joint_id1/orig_points_participant_task_l2_new.txt')
if __name__ == '__main__':
main_realtime_traj_dict()
def visualize_activations(matrix_activation):
### Training 625 matrix activation on shallow ae on only arms
# participant_length = [0, 2197, 2082, 1873, 1595, 1779, 1991, 2148, 1702, 2484, 1744, 2902, 1947, 1860, 1743, 1645,
# 2398, 2287, 1998, 1573]
# s = []
# dim = 30
# for l in xrange(1, len(participant_length)):
# slide = matrix_activation[participant_length[l - 1]:(participant_length[l - 1] + participant_length[l])]
#
# for m in xrange(0, len(slide) - dim, dim):
# if len(s) > 0:
# s = np.vstack((s, matrix_activation[m:m + dim].reshape((1, -1))))
# else:
# s = matrix_activation[m:m + dim].reshape((1, -1))
### trained deep AE on upperBody
participant_length = [
0, 2876, 2394, 2256, 1998, 1887, 2597, 2703, 2105, 3137, 2190, 4072,
2226, 2282, 2480, 2120, 2536, 2507, 2511, 1675
]
s = []
dim = 50
for l in xrange(1, len(participant_length)):
slide = matrix_activation[participant_length[l - 1]:(
participant_length[l - 1] + participant_length[l])]
for m in xrange(0, len(slide) - dim, dim):
if len(s) > 0:
s = np.vstack((s, matrix_activation[m:m + dim].reshape(
(1, -1))))
else:
s = matrix_activation[m:m + dim].reshape((1, -1))
print s.shape
#s = np.array(random.sample(matrix_activation, 30000))
# kernel_bandwith = 5.1
# X = img_processing.my_mean_shift(s, iterations=5, kernel_bandwith=kernel_bandwith)
# print datetime.now().time()
# my_kmean = KMeans(n_clusters=3, n_jobs=-1, algorithm='full')
# X = my_kmean.fit(s)
# means = np.mean(X,axis=1)
pca = decomposition.PCA(
n_components=100
) # 2-dimensional PCA whiten=True, svd_solver='randomized'
s_t = pca.fit(s)
data_organizer.save_matrix_pickle(
s_t,
'C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/posture_data/upperBody/100pca_deep900225AE_5sec_data.txt'
)
s_t = pca.transform(s)
#print s_t.shape
print np.sum(pca.explained_variance_ratio_)
# plt.bar(range(100), pca.explained_variance_ratio_)
# plt.show()
## testing clustering
#m_s = KMeans(n_clusters=10, n_jobs=-1)
#m_s = MeanShift(n_jobs=-1,bandwidth=0.9)
#m_s.fit(s_t)
#s_t = s
m_s = AgglomerativeClustering(n_clusters=15,
affinity='cosine',
linkage='average')
m_s.fit(s_t)
y_tr = m_s.fit_predict(s_t)
print Counter(y_tr)
##since agglomerative clustering doesnt have predict I use svm with the cluster labels for classification
clf = svm.LinearSVC().fit(s_t, y_tr)
data_organizer.save_matrix_pickle(
clf,
'C:/Users/dario.dotti/Documents/data_for_personality_exp/after_data_cleaning/posture_data/linearSVM_agglomerative15c_5sec_100pca.txt'
)
#print 'file saved'
colors = np.array(np.random.randint(0, 255, size=(20, 3))) / 255.0
color_labels = [colors[p] for p in y_tr]
## 2D
plt.scatter(s_t[:, 0], s_t[:, 1], c=color_labels)
#plt.scatter(m_s[:, 0], m_s[:, 1], marker='^', c='r')
plt.show()
##3D
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(s_t[:, 0], s_t[:, 1], s_t[:, 2],
c=color_labels) # s_t[:, 1], s_t[:, 2],s_t[:,0]
#ax.scatter(m_s[:, 0], m_s.means_[:, 1], m_s.means_[:, 2], marker='^', c='r')
plt.show()
return s, s_t, m_s, y_tr