def train_markov(training_data, training_labels, training_files, eval_data,
eval_labels):
global hmm_models
global hidden_states
print('Init training markov')
cls = 0
while True:
list_data = list()
list_files = list()
for index in range(len(training_data)):
if training_labels[index] == cls:
seq_list = training_data[index]
file = training_files[index]
list_data.append(seq_list)
list_files.append(file)
if len(list_data) == 0:
break
else:
model_path = os.path.join(ClassHMM.model_hmm_folder_action,
'model{0}.pkl'.format(cls))
hmm_model = ClassHMM(model_path)
print('Training model {0}'.format(cls))
hmm_model.train(list_data, hidden_states)
hmm_models.append(hmm_model)
cls += 1
print('Models trained')
eval_markov(eval_data, eval_labels)
def main():
print('Initializing main function')
# Withdrawing tkinter
# Loading model dirs
list_folder_data = [
('/home/mauricio/CNN/Classes/Door', 0.05),
('/home/mauricio/CNN/Classes/Tires', 0.05),
('/home/mauricio/CNN/Classes/Walk', 0.05),
]
list_hmm = []
for folder_data in list_folder_data:
label_name = get_label_from_folder(folder_data[0])
full_model_dir = os.path.join(hnn_model_folder,
'{0}.pkl'.format(label_name))
list_hmm.append(ClassHMM(full_model_dir))
# Initializing instances
instance_pose = ClassOpenPose()
instance_nn = ClassNN.load_from_params(nn_model_dir)
option = input('Select 1 to train, 2 to eval hmm, 3 to preprocess: ')
if option == '1':
print('Train hmm selection')
train_hmm(list_folder_data, list_hmm, instance_nn, instance_pose)
elif option == '2':
eval_hmm(list_folder_data, list_hmm, instance_nn, instance_pose)
elif option == '3':
recalculate = False
pre_process_images(instance_pose, list_folder_data, recalculate)
else:
print('Invalid argument: {0}'.format(option))
def classify_markov():
global hmm_models
print('Init classification using markov')
training_data = list()
training_labels = list()
training_files = list()
eval_data = list()
eval_labels = list()
eval_files = list()
# Loading classes
for index, item in enumerate(list_folder_data):
folder = item['folderPath']
label = item['label']
num_file = 0
list_paths = list()
for root, _, files in os.walk(folder):
for file in files:
full_path = os.path.join(root, file)
ext = ClassUtils.get_filename_extension(full_path)
if ext == '.json':
list_paths.append(full_path)
total_samples = len(list_paths)
total_train = int(total_samples * 80 / 100)
# Shuffle samples
random.Random(seed).shuffle(list_paths)
for full_path in list_paths:
list_key_poses = list()
with open(full_path, 'r') as f:
json_txt = f.read()
json_data = json.loads(json_txt)
list_poses = json_data['listPoses']
for pose in list_poses:
list_key_poses.append(pose['keyPose'])
if num_file < total_train:
training_data.append(list_key_poses)
training_labels.append(label)
training_files.append(full_path)
else:
eval_data.append(list_key_poses)
eval_labels.append(label)
eval_files.append(full_path)
num_file += 1
print('Total training: {0}'.format(len(training_data)))
print('Total eval: {0}'.format(len(eval_data)))
res = input('Press 1 to train. Press 2 to eval - Press 3 to train iter: ')
if res == '1':
train_markov(training_data, training_labels, training_files, eval_data,
eval_labels)
elif res == '2':
# Pre-loading models
for i in range(len(list_folder_data)):
model_path = os.path.join(ClassHMM.model_hmm_folder_action,
'model{0}.pkl'.format(i))
hmm_model = ClassHMM(model_path)
hmm_models.append(hmm_model)
eval_markov(eval_data, eval_labels)
elif res == '3':
train_markov_iter(training_data, training_labels, training_files,
eval_data, eval_labels)
else:
raise Exception('Option not implemented: {0}'.format(res))
def train_markov_iter(training_data, training_labels, training_files,
eval_data, eval_labels):
global hmm_models
global hidden_states
print('Init training markov iteration')
selected_models = list()
max_precision = 0
for _ in range(iterations):
# Reset iterations
hmm_models.clear()
cls = 0
while True:
list_data = list()
list_files = list()
for index in range(len(training_data)):
if training_labels[index] == cls:
seq_list = training_data[index]
file = training_files[index]
list_data.append(seq_list)
list_files.append(file)
if len(list_data) == 0:
break
else:
model_path = os.path.join(ClassHMM.model_hmm_folder_action,
'model{0}.pkl'.format(cls))
hmm_model = ClassHMM(model_path)
print('Training model {0}'.format(cls))
hmm_model.train(list_data, hidden_states)
hmm_models.append(hmm_model)
cls += 1
print('Models trained')
precision = eval_markov(eval_data, eval_labels)
if precision > max_precision:
selected_models.clear()
# Copy list no reference
for model in hmm_models:
selected_models.append(model)
max_precision = precision
# Saving selected models
hmm_models = selected_models
for model in selected_models:
model.save_model()
print('Evaluating model for final precision: {0}'.format(max_precision))
eval_markov(eval_data, eval_labels)
print('Evaluating model second chance')
eval_markov(eval_data, eval_labels)
print('Done!')
def train_hmm(training_list_cls, training_cls_labels, validate_list_cls,
validate_cls_labels, eval_list_poses, eval_labels, option,
base_data):
print('Train HMM')
# Iterating approach
iterations = 10
selected_models = list()
max_precision = 0
for _ in range(iterations):
cls = 0
hmm_models = list()
while True:
list_data = list()
for index in range(len(training_list_cls)):
if training_cls_labels[index] == cls:
list_poses = training_list_cls[index]
list_cls = list()
for pose in list_poses:
list_cls.append(pose['class'])
list_data.append(list_cls)
if len(list_data) == 0:
break
else:
model_path = os.path.join(ClassHMM.model_hmm_folder_action,
'model_{0}.pkl'.format(cls))
hmm_model = ClassHMM(model_path)
print('Training model {0}'.format(cls))
hmm_model.train(list_data, hidden_states)
hmm_models.append(hmm_model)
cls += 1
# Eval Markov
precision = eval_markov(validate_list_cls, validate_cls_labels,
hmm_models)
if precision > max_precision:
selected_models.clear()
for model in hmm_models:
selected_models.append(model)
max_precision = precision
# Saving selected models
hmm_models = selected_models
for model in selected_models:
model.save_model()
precision = eval_markov(validate_list_cls, validate_cls_labels, hmm_models)
real_precision = eval_markov(eval_list_poses, eval_labels, hmm_models)
print('Precision: {0}'.format(precision))
print('Max precision: {0}'.format(max_precision))
print('Real Precision: {0}'.format(real_precision))
# Evaluating current actions
for folder_info in list_classes:
folder = folder_info['folderPath']
for root, _, files in os.walk(folder):
for file in files:
full_path = os.path.join(root, file)
if '_{0}_partialdata'.format(base_data) in full_path:
process_file_action(full_path,
option,
hmm_models=hmm_models,
accuracy=real_precision)
def train_hmm(training_list_cls, training_cls_labels, validate_list_cls,
validate_cls_labels, eval_list_actions, eval_labels,
option: Option, base_data_1, base_data_2):
print('Initializing training HMM')
hmm_models = list()
hidden_states = 6
# Training using iterations
iterations = 5
selected_models = list()
max_precision = 0
for _ in range(iterations):
hmm_models.clear()
for index in range(len(list_classes)):
# Creating model for each class
model_path = os.path.join(ClassHMM.model_hmm_folder_activities,
'model{0}.pkl'.format(index))
hmm_model = ClassHMM(model_path)
# Get sequences for class:
list_data = list()
for idx_label, label in enumerate(training_cls_labels):
if label == index:
list_actions = training_list_cls[idx_label]
seq = get_seq_hmm(list_actions)
list_data.append(seq)
print('Training model {0}'.format(index))
hmm_model.train(list_data, hidden_states)
hmm_models.append(hmm_model)
res = eval_markov(validate_list_cls, validate_cls_labels, hmm_models)
print('Precision for model. {0}'.format(res['precision']))
if res['precision'] > max_precision:
selected_models.clear()
for model in hmm_models:
selected_models.append(model)
max_precision = res['precision']
# Saving selected models
hmm_models = selected_models
for model in selected_models:
model.save_model()
precision = eval_markov(validate_list_cls, validate_cls_labels,
hmm_models)['precision']
real_precision = eval_markov(eval_list_actions, eval_labels,
hmm_models)['precision']
print('Precision: {0}'.format(precision))
print('Max precision: {0}'.format(max_precision))
print('Real Precision: {0}'.format(real_precision))
apply_classifier(option,
base_data_1,
base_data_2,
hmm_models=hmm_models,
accuracy=precision,
real_accuracy=real_precision)
def class_method():
print('Class Method')
seq1 = [[3, 2, 3, 2, 3], [1, 2], [0, 1, 0, 1, 0]]
seq2 = [[3, 3, 3, 2, 2], [1, 1], [2, 2, 2, 3, 3]]
model_1_path = '/home/mauricio/models/hmm/train1.pkl'
model_2_path = '/home/mauricio/models/hmm/train2.pkl'
model_1 = ClassHMM(model_1_path)
model_2 = ClassHMM(model_2_path)
hidden_states = 6
model_1.train(seq1, hidden_states)
model_2.train(seq2, hidden_states)
seq_test = [3, 2, 3, 2, 3]
score1 = model_1.get_score(seq_test)
score2 = model_2.get_score(seq_test)
print('Score 1: {0} - Score 2: {1}'.format(score1, score2))
seq_test_2 = [4, 2, 3, 2, 3]
score1 = model_1.get_score(seq_test_2)
score2 = model_2.get_score(seq_test_2)
print('Score 1: {0} - Score 2: {1}'.format(score1, score2))
seq_test3 = [3, 3, 3, 3, 3]
score1 = model_1.get_score(seq_test3)
score2 = model_2.get_score(seq_test3)
print('Score 1: {0} - Score 2: {1}'.format(score1, score2))
print('Done!')