def main():
parser = argparse.ArgumentParser(description='Semantic Segmentation')
parser.add_argument('--train_cfg',
type=str,
default='./configs/train_config.yaml',
help='train config path')
args = parser.parse_args()
config_folder = Path(args.train_cfg.strip("/"))
config = load_yaml(config_folder)
init_seed(config['SEED'])
df, train_ids, valid_ids = split_dataset(config['DATA_TRAIN'])
train_dataset = getattribute(config=config,
name_package='TRAIN_DATASET',
df=df,
img_ids=train_ids)
valid_dataset = getattribute(config=config,
name_package='VALID_DATASET',
df=df,
img_ids=valid_ids)
train_dataloader = getattribute(config=config,
name_package='TRAIN_DATALOADER',
dataset=train_dataset)
valid_dataloader = getattribute(config=config,
name_package='VALID_DATALOADER',
dataset=valid_dataset)
model = getattribute(config=config, name_package='MODEL')
criterion = getattribute(config=config, name_package='CRITERION')
optimizer = getattribute(config=config,
name_package='OPTIMIZER',
params=model.parameters())
scheduler = getattribute(config=config,
name_package='SCHEDULER',
optimizer=optimizer)
device = config['DEVICE']
metric_ftns = [accuracy_dice_score]
num_epoch = config['NUM_EPOCH']
gradient_clipping = config['GRADIENT_CLIPPING']
gradient_accumulation_steps = config['GRADIENT_ACCUMULATION_STEPS']
early_stopping = config['EARLY_STOPPING']
validation_frequency = config['VALIDATION_FREQUENCY']
saved_period = config['SAVED_PERIOD']
checkpoint_dir = Path(config['CHECKPOINT_DIR'], type(model).__name__)
checkpoint_dir.mkdir(exist_ok=True, parents=True)
resume_path = config['RESUME_PATH']
learning = Learning(model=model,
optimizer=optimizer,
criterion=criterion,
device=device,
metric_ftns=metric_ftns,
num_epoch=num_epoch,
scheduler=scheduler,
grad_clipping=gradient_clipping,
grad_accumulation_steps=gradient_accumulation_steps,
early_stopping=early_stopping,
validation_frequency=validation_frequency,
save_period=saved_period,
checkpoint_dir=checkpoint_dir,
resume_path=resume_path)
learning.train(tqdm(train_dataloader), tqdm(valid_dataloader))
def __init__(self, total_episodes, time_training_step, time_testing_step):
self.timestamp = datetime.now().strftime('%Y%m%d_%H_%M_%S')
# Ínitialize a track instance
self.Track = Track(self.timestamp)
# Return the shape of the track
self.way, self.dims = self.Track.retrieve_way(
), self.Track.retrieve_way().shape
# Initiate the Learning module using the dimensions of the track
self.Learning = Learning(self.timestamp, self.dims)
self.total_episodes = total_episodes
# setup
self.setup()
def __init__(self, pomdpfile='program.pomdp'):
self.time = ['morning', 'afternoon', 'evening']
self.location = ['classroom', 'library']
self.identity = ['student', 'professor', 'visitor']
self.intention = ['interested', 'not_interested']
self.reason = Reason('reason0.plog')
self.model = Model(filename='program.pomdp', parsing_print_flag=False)
self.policy = Policy(5, 4, output='program.policy')
self.instance = []
self.results = {}
self.learning = Learning('./', 'interposx.csv', 'interposy.csv')
self.trajectory_label = 0
def main():
codecool_msc = CodecoolClass.create_local()
late_students = MorningRoutine.random_things(codecool_msc.students)
JusticeDepartment.random_punishment(late_students)
print(codecool_msc.avg_energy())
print(codecool_msc.avg_knowledge())
Learning.morning_dojo(codecool_msc.students)
Breaks.random_breaks(codecool_msc.students)
coder_pairs = Pair.create_pairs(codecool_msc)
Learning.pair_programming(coder_pairs)
print(codecool_msc.avg_knowledge())
Breaks.mass_lunchbreak(codecool_msc)
NaturalDiasaster.random_disasters(codecool_msc.students)
print("The day is over and everyone went home, hopefully with more knowledge than at the start of the day.")
def __init__(self, scope, q, alpha, reward, discount, quotes, bankroll,
log=None):
self.logger = log
self.actions = ACTIONS # BUY, SELL, DO_NOTHING
Indicators.__init__(self, log)
Learning.__init__(self, q, alpha, reward, discount, self.state, \
self.actions)
Order.__init__(self, scope, bankroll, log)
self.num_trades = 0
self.performance = 1
self.volume = max(self.performance, 1)
self.logger = log
self.status = {'status':'','action':''}
self.quotes = quotes
self.states = None
def __init__(self,
scope,
q,
alpha,
reward,
discount,
quotes,
bankroll,
log=None):
self.logger = log
self.scope = scope
self.actions = ACTIONS
Indicators.__init__(self, log)
Order.__init__(self, scope, bankroll, log)
Learning.__init__(self, q, alpha, reward, discount, self.state, \
self.actions)
self.num_trades = 0
self.performance = 1
self.volume = max(self.performance, 1)
self.logger = log
self.status = {'status': 'idle', 'action': ''}
self.quotes = quotes
self.states = None
def __init__(self,args):
self.checkpoint_dir = '/data/tf/ckpts'
self.log_dir = '/data/tf/logs'
self.result_dir = 'results'
self._attrs = ['generator','rule','rule_apply', 'order', 'subsample', 'ratio'] #'gan', ,'parse'
Learning.__init__(self,
model=args.model, dataset=args.dataset,
model_path=args.model_path,data_path=args.data_path,
generator=args.generator, rule=args.rule,
gan=args.gan, parse=args.parse,
rule_apply=args.rule_apply,order=args.order,
subsample=args.subsample, ratio=args.ratio,
reload=args.reload, beam=args.beam)
tf.reset_default_graph()
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth=True
tf_config.intra_op_parallelism_threads=15
tf_config.inter_op_parallelism_threads=15
self.sess = tf.InteractiveSession(config=tf_config)
class Game:
""" Loop over multiple episodes to
train the reinforcement model
"""
def __init__(self, total_episodes, time_training_step, time_testing_step):
self.timestamp = datetime.now().strftime('%Y%m%d_%H_%M_%S')
# Ínitialize a track instance
self.Track = Track(self.timestamp)
# Return the shape of the track
self.way, self.dims = self.Track.retrieve_way(
), self.Track.retrieve_way().shape
# Initiate the Learning module using the dimensions of the track
self.Learning = Learning(self.timestamp, self.dims)
self.total_episodes = total_episodes
# setup
self.setup()
def setup(self):
""" Setup the logs """
self.all_episodes_moves = []
# setup the folders for the logs
if not os.path.isdir(os.path.join(os.getcwd(), "logs")):
os.path.mkdir(os.path.join(os.getcwd(), "logs"))
# Prepare the logging
log_num_name = "{timestamp}_log_numerics.csv".format(
timestamp=self.timestamp)
self.log_num_path = os.path.join(os.getcwd(), "logs", log_num_name)
plot_name = "{timestamp}_plot.png".format(timestamp=self.timestamp)
self.plot_path = os.path.join(os.getcwd(), "logs", plot_name)
moves_name = "all_moves.obj".format(timestamp=self.timestamp)
self.moves_path = os.path.join(os.getcwd(), "logs", moves_name)
topology_name = "track_topology.npy".format(timestamp=self.timestamp)
self.topology_path = os.path.join(os.getcwd(), "logs", topology_name)
# Save the track condition
np.save(self.topology_path, np.asarray(self.way))
def train_model(self):
""" Run the model a multiple number of times
to fill the Q matrix
"""
for episode in range(self.total_episodes):
Episode = Game_Episode(self.Learning, self.Track, episode,
self.timestamp, time_training_step)
episode_moves = Episode.run_train_episode()
self.all_episodes_moves.append(episode_moves)
def test_prediction(self):
""" Using the filled Q matrix, test the learned parameters """
for episode in range(self.total_episodes, self.total_episodes + 1):
Episode = Game_Episode(self.Learning, self.Track, episode,
self.timestamp, time_testing_step)
episode_moves = Episode.run_test_episode()
self.all_episodes_moves.append(episode_moves)
def clean_up(self):
""" Visualize the learning parameters.
Save the parameters in a separate file
"""
# Save the moves for every episode to a file
with open(self.moves_path, 'wb') as log:
pickle.dump(self.all_episodes_moves, log)
# Retrieve the learning parameters from each step performed
parameters = np.round(np.array(self.Learning.retrieve_l_parameters()),
2)
plt.plot(parameters[:, 0],
parameters[:, 2],
color='lightblue',
linewidth=3)
plt.title("Q Learning process")
plt.xlabel("Episode")
plt.ylabel("Sum of squared Q matrix elements")
plt.savefig(self.plot_path)
plt.show()
# Save the learning parameters explicitly to a file
with open(self.log_num_path, 'w') as log:
log.write(str(parameters) + "\n")
audio = 5
else:
audio = 1
video = 0
camera = 0
checkagain = 1
counter_silence = 0
##############################################################
elif info == 1:
data = ast.literal_eval(data.decode('utf-8'))
if menuvariable == 1:
machinelearningtext = data
conn.sendall(b"FinishLearning.endmes")
else:
learnpepper = Learning(data)
pas, law, saving, swerve = learnpepper.learn()
check = checklearning(pas, law, saving, swerve)
if check != 'ok':
mystring = "LearnMore.endmes" + check
string = mystring.encode('utf-8')
conn.sendall(string)
learn = 2
if interactionvariable == 1:
audio = 5
else:
audio = 1
info = 0
else:
import numpy as np
from environment_explauto.environment import TestEnvironment
from learning import Learning
if __name__ == "__main__":
print "Create environment"
environment = TestEnvironment()
print "Create agent"
learning = Learning(dict(m_mins=environment.conf.m_mins,
m_maxs=environment.conf.m_maxs,
s_mins=environment.conf.s_mins,
s_maxs=environment.conf.s_maxs),
condition="AMB")
learning.start()
print
print "Do 100 autonomous steps:"
for i in range(100):
context = environment.get_current_context()
m = learning.produce(context)
s = environment.update(m)
learning.perceive(s)
print "Do 1 arm demonstration"
m_demo_traj = np.zeros((25, 4)) + 0.001
m_demo = environment.torsodemo2m(m_demo_traj)
print "m_demo", m_demo
s = environment.update(m_demo)
learning.perceive(s, m_demo=m_demo)
spec = prep.pre_transportation(spec)
spec = prep.pre_land(spec)
spec = prep.pre_area(spec)
spec = prep.pre_structure(spec)
spec = prep.pre_age(spec)
spec = prep.pre_floor(spec)
spec = prep.pre_direction(spec)
spec = prep.pre_mcost(spec)
spec = prep.pre_rcost(spec)
spec = prep.encode_cat_to_label(spec)
spec = prep.pre_outlier(spec)
spec = prep.del_bid(spec)
prep.save_spec(spec)
#learning
learning = Learning(data_version)
gs = learning.xgb_learning(spec)
learning.show_results(gs)
learning.save_model(gs)
# spec_all,spec_bad=learning.check_prediction(spec)
# learning.save_prediction(spec_all)
elif mode == 'inference':
data_version = datetime.now().strftime("%Y%m%d")
model_version = input('which model do you use? ex.original...')
#spec取得
get_spec = GetSpec(data_version)
page_num = get_spec.get_page_num()
urls = get_spec.get_urls(page_num)
pages = get_spec.get_pages(urls)
get_spec.save_pages(pages)
def main():
parser = argparse.ArgumentParser(description='Semantic Segmentation')
parser.add_argument('--train_cfg',
type=str,
default='./configs/train.yaml',
help='train config path')
args = parser.parse_args()
config_folder = Path(args.train_cfg.strip("/"))
config = load_yaml(config_folder)
init_seed(config['SEED'])
image_datasets = {
x: vinDataset(root_dir=config['ROOT_DIR'],
file_name=config['FILE_NAME'],
num_triplet=config['NUM_TRIPLET'],
phase=x)
for x in ['train', 'valid']
}
dataloaders = {
x: torch.utils.data.DataLoader(image_datasets[x],
batch_size=config['BATCH_SIZE'],
shuffle=True,
num_workers=4,
pin_memory=True)
for x in ['train', 'valid']
}
model = getattribute(config=config, name_package='MODEL')
criterion = getattribute(config=config, name_package='CRITERION')
metric_ftns = [accuracy_score]
optimizer = getattribute(config=config,
name_package='OPTIMIZER',
params=model.parameters())
scheduler = getattribute(config=config,
name_package='SCHEDULER',
optimizer=optimizer)
device = config['DEVICE']
num_epoch = config['NUM_EPOCH']
gradient_clipping = config['GRADIENT_CLIPPING']
gradient_accumulation_steps = config['GRADIENT_ACCUMULATION_STEPS']
early_stopping = config['EARLY_STOPPING']
validation_frequency = config['VALIDATION_FREQUENCY']
saved_period = config['SAVED_PERIOD']
checkpoint_dir = Path(config['CHECKPOINT_DIR'], type(model).__name__)
checkpoint_dir.mkdir(exist_ok=True, parents=True)
resume_path = config['RESUME_PATH']
learning = Learning(model=model,
criterion=criterion,
metric_ftns=metric_ftns,
optimizer=optimizer,
device=device,
num_epoch=num_epoch,
scheduler=scheduler,
grad_clipping=gradient_clipping,
grad_accumulation_steps=gradient_accumulation_steps,
early_stopping=early_stopping,
validation_frequency=validation_frequency,
save_period=saved_period,
checkpoint_dir=checkpoint_dir,
resume_path=resume_path)
learning.train(tqdm(dataloaders['train']), tqdm(dataloaders['valid']))
ITERATIVE_LEARNING = 1 # Evaluate as iterative learning?
IMAGE_NAME_COLUMN = 256
SESSION_COLUMN = 257
CATEGORY_COLUMN = 258
INSTANCE_COLUMN = 259
TRAIN_SESSIONS = [1, 2, 4, 5, 6, 8, 9, 11]
TEST_SESSIONS = [3, 7, 10]
DATA_DIMENSION = 256
FACTOR_FRAMES = 4 # Every Nth frame will be selected. Only 2 and 4 are reasonable values. Original number of
# frames is 8. In this case it will be reduced to 4 and 2, respectively.
# ------------------------------------ Initialization --------------------------------------------------------------
rgwr = GammaGWR()
utils = Utilities()
learning = Learning()
args = utils.parse_arguments()
# Get data.
original_data = utils.load_data(args.dataset).values
original_data_normalized = utils.normalize_data(original_data,
DATA_DIMENSION)
train_data = original_data_normalized[np.in1d(
original_data_normalized[:, SESSION_COLUMN], TRAIN_SESSIONS)]
train_data = utils.reduce_number_of_frames(train_data, FACTOR_FRAMES)
test_data = original_data_normalized[np.in1d(
original_data_normalized[:, SESSION_COLUMN], TEST_SESSIONS)]
# ------------------------------------ Batch learning---------------------------------------------------------------
import math
import time
import torch
import torch.nn as nn
import torchvision.transforms as t
import torch.nn.functional as F
import torch.optim as optim
import random
import numpy as np
import matplotlib.pyplot as plt
from mario_q import MarioManager
from helpers import Transition
from helpers import ReplayMemory
from helpers import DQN
from learning import Learning
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
em = MarioManager(device)
memory = ReplayMemory(1000000)
policy = DQN(em.screen_height(), em.screen_width()).to(device)
target = DQN(em.screen_height(), em.screen_width()).to(device)
optimizer = optim.Adam(params=policy.parameters(), lr=0.001)
target.load_state_dict(policy.state_dict())
target.eval()
learning_agent = Learning(policy, target, em, memory, optimizer)
learning_agent.learn()
learning_agent.plot_on_figure()
def train_fold(train_config, experiment_folder, pipeline_name, log_dir,
fold_id, train_dataloader, val_dataloader, binarizer_fn,
eval_fn):
fold_logger = helpers.init_logger(log_dir, f'train_log_{fold_id}.log')
best_checkpoint_folder = Path(experiment_folder,
train_config['CHECKPOINTS']['BEST_FOLDER'])
best_checkpoint_folder.mkdir(parents=True, exist_ok=True)
checkpoints_history_folder = Path(
experiment_folder, train_config['CHECKPOINTS']['FULL_FOLDER'],
f'fold_{fold_id}')
checkpoints_history_folder.mkdir(parents=True, exist_ok=True)
checkpoints_topk = train_config['CHECKPOINTS']['TOPK']
calculation_name = f'{pipeline_name}_fold_{fold_id}'
device = train_config['DEVICE']
module = importlib.import_module(train_config['MODEL']['PY'])
model_class = getattr(module, train_config['MODEL']['CLASS'])
model = model_class(**train_config['MODEL']['ARGS'])
pretrained_model_config = train_config['MODEL'].get('PRETRAINED', False)
if pretrained_model_config:
loaded_pipeline_name = pretrained_model_config['PIPELINE_NAME']
pretrained_model_path = Path(
pretrained_model_config['PIPELINE_PATH'],
pretrained_model_config['CHECKPOINTS_FOLDER'],
f'{loaded_pipeline_name}_fold_{fold_id}.pth')
if pretrained_model_path.is_file():
model.load_state_dict(torch.load(pretrained_model_path))
fold_logger.info(f'Load model from {pretrained_model_path}')
if len(train_config['DEVICE_LIST']) > 1:
model = torch.nn.DataParallel(model)
module = importlib.import_module(train_config['CRITERION']['PY'])
loss_class = getattr(module, train_config['CRITERION']['CLASS'])
loss_fn = loss_class(**train_config['CRITERION']['ARGS'])
optimizer_class = getattr(torch.optim, train_config['OPTIMIZER']['CLASS'])
optimizer = optimizer_class(model.parameters(),
**train_config['OPTIMIZER']['ARGS'])
scheduler_class = getattr(torch.optim.lr_scheduler,
train_config['SCHEDULER']['CLASS'])
scheduler = scheduler_class(optimizer, **train_config['SCHEDULER']['ARGS'])
n_epochs = train_config['EPOCHS']
grad_clip = train_config['GRADIENT_CLIPPING']
grad_accum = train_config['GRADIENT_ACCUMULATION_STEPS']
early_stopping = train_config['EARLY_STOPPING']
validation_frequency = train_config.get('VALIDATION_FREQUENCY', 1)
freeze_model = train_config['MODEL']['FREEZE']
Learning(optimizer, binarizer_fn, loss_fn, eval_fn, device, n_epochs,
scheduler, freeze_model, grad_clip, grad_accum, early_stopping,
validation_frequency, calculation_name, best_checkpoint_folder,
checkpoints_history_folder, checkpoints_topk,
fold_logger).run_train(model, train_dataloader, val_dataloader)
class Simulator:
def __init__(self, pomdpfile='program.pomdp'):
self.time = ['morning', 'afternoon', 'evening']
self.location = ['classroom', 'library']
self.identity = ['student', 'professor', 'visitor']
self.intention = ['interested', 'not_interested']
self.reason = Reason('reason0.plog')
self.model = Model(filename='program.pomdp', parsing_print_flag=False)
self.policy = Policy(5, 4, output='program.policy')
self.instance = []
self.results = {}
self.learning = Learning('./', 'interposx.csv', 'interposy.csv')
self.trajectory_label = 0
def sample(self, alist, distribution):
return np.random.choice(alist, p=distribution)
def create_instance(self, i):
random.seed(i)
#person = random.choice(self.identity)
person = 'visitor'
#print ('\nIdentity (uniform sampling): [student, visitor, professor] : '), person
# print ('identity is:'), person
if person == 'student':
#place ='classroom'
place = self.sample(self.location, [0.7, 0.3])
#time ='evening'
#time =self.sample(self.time,[0.15,0.15,0.7])
#intention =self.sample(self.intention,[0.3,0.7])
#intention = 'not_interested'
elif person == 'professor':
place = self.sample(self.location, [0.9, 0.1])
time = self.sample(self.time, [0.8, 0.1, 0.1])
intention = self.sample(self.intention, [0.1, 0.9])
else:
#place = self.sample(self.location,[0.2,0.8])
place = 'classroom'
#time =self.sample(self.time,[0.1,0.7,0.2])
time = 'afternoon'
intention = self.sample(self.intention, [0.8, 0.2])
intention = 'interested'
self.trajectory_label = self.learning.get_traj(intention)
print('Sampling time, location and intention for the identity: ' +
person)
self.instance.append(person)
self.instance.append(time) #1
self.instance.append(place) #2
self.instance.append(intention)
self.instance.append('trajectory with label ' +
str(self.trajectory_label))
print('Instance: ')
print(self.instance[0], self.instance[1], self.instance[2],
self.instance[3], self.instance[4])
return self.instance
def observe_fact(self, i):
random.seed(i)
#print '\nObservations:'
time = self.instance[1]
location = self.instance[2]
#print ('Observed time: '),time
#print ('Observed location: '),location
return time, location
def init_belief(self, int_prob):
l = len(self.model.states)
b = np.zeros(l)
# initialize the beliefs of the states with index=0 evenly
int_prob = float(int_prob)
init_belief = [0, 0, 1.0 - int_prob, int_prob, 0]
b = np.zeros(len(self.model.states))
for i in range(len(self.model.states)):
b[i] = init_belief[i] / sum(init_belief)
print 'The normalized initial belief would be: '
print b
return b
def get_state_index(self, state):
return self.model.states.index(state)
def init_state(self):
state = random.choice(
['not_turned_not_interested', 'not_turned_interested'])
#print '\nRandomly selected state from [not_forward_not_interested,not_forward_interested] =',state
s_idx = self.get_state_index(state)
#print s_idx
return s_idx, state
def get_obs_index(self, obs):
return self.model.observations.index(obs)
def observe(self, a_idx, intention, pln_obs_acc):
p = pln_obs_acc
if self.model.actions[
a_idx] == 'move_forward' and intention == 'interested':
#obs='physical'
obs = self.sample(['pos', 'neg'], [p, 1 - p])
elif self.model.actions[
a_idx] == 'move_forward' and intention == 'not_interested':
obs = obs = self.sample(['pos', 'neg'], [1 - p, p])
elif self.model.actions[a_idx] == 'greet' and intention == 'interested':
#obs = 'verbal'
obs = self.sample(['pos', 'neg'], [p, 1 - p])
elif self.model.actions[
a_idx] == 'greet' and intention == 'not_interested':
obs = self.sample(['pos', 'neg'], [1 - p, p])
elif self.model.actions[a_idx] == 'turn' and intention == 'interested':
#obs = 'verbal'
obs = self.sample(['pos', 'neg'], [p, 1 - p])
elif self.model.actions[
a_idx] == 'turn' and intention == 'not_interested':
obs = self.sample(['pos', 'neg'], [1 - p, p])
else:
obs = 'na'
#l=len(self.model.observations)-1
#o_idx=randint(0,l)
o_idx = self.get_obs_index(obs)
print('random observation is: ', self.model.observations[o_idx])
return o_idx
def update(self, a_idx, o_idx, b):
b = np.dot(b, self.model.trans_mat[a_idx, :])
b = [
b[i] * self.model.obs_mat[a_idx, i, o_idx]
for i in range(len(self.model.states))
]
b = b / sum(b)
return b
def run(self, strategy, time, location, r_thresh, l_thresh, pln_obs_acc):
a_cnt = 0
success = 0
tp = 0
tn = 0
fp = 0
fn = 0
cost = 0
R = 0
if strategy == 'corpp':
#prob = self.reason.query_nolstm(time, location,'reason_nolstm.plog')
prob = self.reason.query_nolstm(time, location, 'reason0.plog')
print colored('\nSTRATEGY: ', 'red'), colored(strategy, 'red')
print '\nOur POMDP Model states are: '
print self.model.states
s_idx, temp = self.init_state()
b = self.init_belief(prob)
#print ( 'b shape is,', b.shape )
#print b
while True:
a_idx = self.policy.select_action(b)
a = self.model.actions[a_idx]
a_cnt = a_cnt + 1
if a_cnt > 20:
print('POLICY IS NOT REPORTING UNDER 20 ACTIONS')
sys.exit()
print('action selected', a)
o_idx = self.observe(a_idx, self.instance[3], pln_obs_acc)
#print ('transition matrix shape is', self.model.trans_mat.shape)
#print self.model.trans_mat[a_idx,:,:]
#print ('observation matrix shape is', self.model.obs_mat.shape)
#print self.model.trans_mat[a_idx,:,:]
#print s_idx
R = R + self.model.reward_mat[a_idx, s_idx]
print 'Reward is : ', cost
#print ('Total reward is,' , cost)
b = self.update(a_idx, o_idx, b)
print b
if 'report' in a:
if 'not_interested' in a and 'not_interested' == self.instance[
3]:
success = 1
tn = 1
print 'Trial was successfull'
elif 'report_interested' in a and 'interested' == self.instance[
3]:
success = 1
tp = 1
print 'Trial was successful'
elif 'report_interested' in a and 'not_interested' == self.instance[
3]:
fp = 1
print 'Trial was unsuccessful'
elif 'not_interested' in a and 'interested' == self.instance[
3]:
fn = 1
print('Finished\n ')
break
cost = cost + self.model.reward_mat[a_idx, s_idx]
print 'cost is : ', cost
if strategy == 'planning':
#prob = self.reason.query_nolstm(time, location,'reason_nolstm.plog')
print colored('\nSTRATEGY', 'green'), colored(strategy, 'green')
print '\nOur POMDP Model states are: '
print self.model.states
s_idx, temp = self.init_state()
#init_belief = [0.25, 0.25, 0.25, 0.25, 0]
#b = np.zeros(len(init_belief))
b = np.ones(len(self.model.states))
for i in range(len(self.model.states)):
b[i] = b[i] / len(self.model.states)
print 'initial belief', b
while True:
a_idx = self.policy.select_action(b)
a = self.model.actions[a_idx]
print('action selected', a)
o_idx = self.observe(a_idx, self.instance[3], pln_obs_acc)
R = R + self.model.reward_mat[a_idx, s_idx]
print 'R is : ', cost
#print ('Total reward is,' , cost)
b = self.update(a_idx, o_idx, b)
print b
if 'report' in a:
if 'not_interested' in a and 'not_interested' == self.instance[
3]:
success = 1
tn = 1
print 'Trial was successfull'
elif 'report_interested' in a and 'interested' == self.instance[
3]:
success = 1
tp = 1
print 'Trial was successful'
elif 'report_interested' in a and 'not_interested' == self.instance[
3]:
fp = 1
print 'Trial was unsuccessful'
elif 'not_interested' in a and 'interested' == self.instance[
3]:
fn = 1
print('Finished\n ')
break
cost = cost + self.model.reward_mat[a_idx, s_idx]
print 'cost is : ', cost
if strategy == 'reasoning':
print colored('\nSTRATEGY is: ',
'yellow'), colored(strategy, 'yellow')
#prob = self.reason.query_nolstm(time, location,'reason_nolstm.plog') prob = self.reason.query_nolstm(time, location,'reason_nolstm.plog')
prob = self.reason.query_nolstm(time, location, 'reason0.plog')
print('P(ineterested)| observed time and location: '), prob
prob = float(prob)
if prob >= r_thresh and 'interested' == self.instance[3]:
success = 1
print('Greater than threshold =' + str(r_thresh) +
', -> human IS interested')
print 'Trial was successful'
tp = 1
elif prob >= r_thresh and 'not_interested' == self.instance[3]:
success = 0
fp = 1
print('Greater than threshold =' + str(r_thresh) +
', -> human IS interested')
print 'Trial was unsuccessful'
elif prob < r_thresh and 'interested' == self.instance[3]:
success = 0
print('Less than threshold =' + str(r_thresh) +
', -> human IS NOT interested')
fn = 1
print 'Trial was unsuccessful'
else:
success = 1
print('Less than threshold =' + str(r_thresh) +
', -> human IS NOT interested')
print 'Trial was successful'
tn = 1
if strategy == 'learning':
print colored('\nStrategy is: ', 'blue'), colored(strategy, 'blue')
res = self.learning.predict()
if res > l_thresh and self.trajectory_label == 1.0:
print('CASE I the trajectory shows person is interested')
success = 1
tp = 1
elif res < l_thresh and self.trajectory_label == 0:
print('CASE II the person is not interested')
success = 1
tn = 1
elif res > l_thresh and self.trajectory_label == 0:
sucess = 0
fp = 1
print('CASE III the trajectory shows person is interested')
elif res < l_thresh and self.trajectory_label == 1.0:
fn = 1
success = 0
('CASE IV the person is not interested')
if strategy == 'lstm-corpp':
print colored('\nSTRATEGY is: ',
'magenta'), colored(strategy, 'magenta')
res = self.learning.predict()
#do not change 0.2 below
if res > l_thresh:
#prob = self.reason.query(time, location,'one','reason.plog')
prob = self.reason.query(time, location, 'one', 'reason0.plog')
else:
prob = self.reason.query(time, location, 'zero',
'reason0.plog')
print '\nOur POMDP Model states are: '
print self.model.states
s_idx, temp = self.init_state()
b = self.init_belief(prob)
while True:
a_idx = self.policy.select_action(b)
a = self.model.actions[a_idx]
print('action selected', a)
o_idx = self.observe(a_idx, self.instance[3], pln_obs_acc)
R = R + self.model.reward_mat[a_idx, s_idx]
print 'R is : ', cost
#print ('Total reward is,' , cost)
b = self.update(a_idx, o_idx, b)
print b
if 'report' in a:
if 'not_interested' in a and 'not_interested' == self.instance[
3]:
success = 1
tn = 1
print 'Trial was successfull'
elif 'report_interested' in a and 'interested' == self.instance[
3]:
success = 1
tp = 1
print 'Trial was successful'
elif 'report_interested' in a and 'not_interested' == self.instance[
3]:
fp = 1
print 'Trial was unsuccessful'
elif 'not_interested' in a and 'interested' == self.instance[
3]:
fn = 1
print('Finished\n ')
break
cost = cost + self.model.reward_mat[a_idx, s_idx]
print 'cost is : ', cost
if strategy == 'lreasoning':
print colored('\nStrategy is: learning + reasoning ', 'cyan')
res = self.learning.predict()
if res > l_thresh:
prob = self.reason.query(time, location, 'one', 'reason0.plog')
else:
prob = self.reason.query(time, location, 'zero',
'reason0.plog')
lr_thresh = r_thresh
prob = float(prob)
if prob >= r_thresh and 'interested' == self.instance[3]:
success = 1
print('Greater than threshold =' + str(r_thresh) +
', -> human IS interested')
print 'Trial was successful'
tp = 1
elif prob >= r_thresh and 'not_interested' == self.instance[3]:
success = 0
fp = 1
print('Greater than threshold =' + str(r_thresh) +
', -> human IS interested')
print 'Trial was unsuccessful'
elif prob < r_thresh and 'interested' == self.instance[3]:
success = 0
print('Less than threshold =' + str(r_thresh) +
', -> human IS NOT interested')
fn = 1
print 'Trial was unsuccessful'
else:
success = 1
print('Less than threshold =' + str(r_thresh) +
', -> human IS NOT interested')
print 'Trial was successful'
tn = 1
return cost, success, tp, tn, fp, fn, R
def trial_num(self, num, strategylist, r_thresh, l_thresh, pln_obs_acc):
df = pd.DataFrame()
total_success = {}
total_cost = {}
total_tp = {}
total_tn = {}
total_fp = {}
total_fn = {}
prec = {}
recall = {}
SDcost = {}
reward = {}
SDreward = {}
for strategy in strategylist:
total_success[strategy] = 0
total_cost[strategy] = 0
total_tp[strategy] = 0
total_tn[strategy] = 0
total_fp[strategy] = 0
total_fn[strategy] = 0
prec[strategy] = 0
recall[strategy] = 0
reward[strategy] = 0
SDreward[strategy] = []
SDcost[strategy] = []
for i in range(num):
print colored('######TRIAL:',
'blue'), colored(i,
'red'), colored('#######', 'blue')
del self.instance[:]
self.create_instance(i)
time, location = self.observe_fact(i)
for strategy in strategylist:
c, s, tp, tn, fp, fn, R = self.run(strategy, time, location,
r_thresh, l_thresh,
pln_obs_acc)
reward[strategy] += R
total_cost[strategy] += c
total_success[strategy] += s
total_tp[strategy] += tp
total_tn[strategy] += tn
total_fp[strategy] += fp
total_fn[strategy] += fn
SDcost[strategy].append(c)
SDreward[strategy].append(R)
# print ('total_tp:'), total_tp
# print ('total_tn:'), total_tn
# print ('total_fp:'), total_fp
# print ('total_fn:'), total_fn
try:
df.at[strategy, 'Reward'] = float(reward[strategy]) / num
df.at[strategy,
'SDCost'] = statistics.stdev(SDcost[strategy])
df.at[strategy,
'SDreward'] = statistics.stdev(SDreward[strategy])
df.at[strategy, 'Cost'] = float(total_cost[strategy]) / num
df.at[strategy,
'Success'] = float(total_success[strategy]) / num
prec[strategy] = round(
float(total_tp[strategy]) /
(total_tp[strategy] + total_fp[strategy]), 2)
recall[strategy] = round(
float(total_tp[strategy]) /
(total_tp[strategy] + total_fn[strategy]), 2)
df.at[strategy, 'Precision'] = prec[strategy]
df.at[strategy, 'Recall'] = recall[strategy]
df.at[strategy, 'F1 Score'] = round(
2 * prec[strategy] * recall[strategy] /
(prec[strategy] + recall[strategy]), 2)
except:
print 'Can not divide by zero'
df.at[strategy, 'Precision'] = 0
df.at[strategy, 'Recall'] = 0
df.at[strategy, 'F1 Score'] = 0
#print 'fp',total_fp['learning']
#print 'fn',total_fn['learning']
#self.instance =[]
return df
def print_results(self, df):
print '\nWRAP UP OF RESULTS:'
print df
filename = "supporter.cfg"
config = configparser.SafeConfigParser()
config.read(filename)
jid = config.get("xmpp", "jid")
password = config.get("xmpp", "password")
room = config.get("xmpp", "room")
nick = config.get("xmpp", "nick")
logging.basicConfig(level=logging.INFO,
format='%(levelname)-8s %(message)s')
words = config.get("brain", "words")
synonyms = config.get("brain", "synonyms")
thoughts = config.get("brain", "thoughts")
messages = config.get("brain", "messages")
state = config.get("brain", "state")
brain = Learning(words, synonyms, thoughts, messages, state)
xmpp = Client(jid, password, room, nick)
xmpp.register_plugin('xep_0030') # Service Discovery
xmpp.register_plugin('xep_0045') # Multi-User Chat
xmpp.register_plugin('xep_0199') # XMPP Ping
def do_brain(nick, msg, **keywords):
brain(msg, nick, xmpp.muc_send)
xmpp.add_message_listener(do_brain)
if xmpp.connect():
xmpp.process(block=True)
else:
print("Unable to connect")
NUMBER_FRAMES = 75 # The number of frames per session. 4 sessions make 300 (75 * 4) session for an object.
CATEGORY_COLUMN = 256
INSTANCE_COLUMN = 257
SESSION_COLUMN = 258
DAY_COLUMN = 259
CAMERA_COLUMN = 260
IMAGE_NAME_COLUMN = 261
DATA_DIMENSION = 256
FACTOR_FRAMES = 2 # Every Nth frame will be selected. Only 2 and 4 are reasonable values. Original number of
# frames is 8. In this case it will be reduced to 4 and 2, respectively.
# ------------------------------------ Initialization --------------------------------------------------------------
rgwr = GammaGWR()
utils = Utilities()
learning = Learning()
args = utils.parse_arguments()
# Get data.
original_data = utils.load_data(args.dataset).values
original_data_normalized = utils.normalize_data(original_data,
DATA_DIMENSION)
original_data_day_one = original_data_normalized[np.in1d(
original_data_normalized[:, DAY_COLUMN], ONE_DAY)]
original_data_left_camera = original_data_day_one[np.in1d(
original_data_day_one[:, CAMERA_COLUMN], CAMERA)]
selected_data = original_data_left_camera[np.in1d(
original_data_left_camera[:, CATEGORY_COLUMN], CATEGORIES)]
# Comment if categorization instead of identification to use. For the rest of the evaluation the CATEGORY column
def main():
parser = argparse.ArgumentParser(description='Pytorch parser')
parser.add_argument('--train_cfg',
type=str,
default='./configs/efficientdet-d0.yaml',
help='train config path')
parser.add_argument('-d',
'--device',
default=None,
type=str,
help='indices of GPUs to enable (default: all)')
parser.add_argument('-r',
'--resume',
default=None,
type=str,
help='path to latest checkpoint (default: None)')
CustomArgs = collections.namedtuple('CustomArgs', 'flags type target')
options = [
CustomArgs(['-lr', '--learning_rate'],
type=float,
target='OPTIMIZER,ARGS,lr'),
CustomArgs(
['-bs', '--batch_size'],
type=int,
target=
'TRAIN_DATALOADER,ARGS,batch_size;VALID_DATALOADER,ARGS,batch_size'
)
]
config = config_parser(parser, options)
init_seed(config['SEED'])
train_dataset = VOCDetection(root=VOC_ROOT,
transform=SSDAugmentation(
voc['min_dim'], MEANS))
train_dataloader = getattribute(config=config,
name_package='TRAIN_DATALOADER',
dataset=train_dataset,
collate_fn=detection_collate)
# valid_dataloader = getattribute(config = config, name_package = 'VALID_DATALOADER', dataset = valid_dataset)
model = getattribute(config=config, name_package='MODEL')
criterion = getattribute(config=config, name_package='CRITERION')
optimizer = getattribute(config=config,
name_package='OPTIMIZER',
params=model.parameters())
scheduler = getattribute(config=config,
name_package='SCHEDULER',
optimizer=optimizer)
device = config['DEVICE']
metric_ftns = []
num_epoch = config['NUM_EPOCH']
gradient_clipping = config['GRADIENT_CLIPPING']
gradient_accumulation_steps = config['GRADIENT_ACCUMULATION_STEPS']
early_stopping = config['EARLY_STOPPING']
validation_frequency = config['VALIDATION_FREQUENCY']
tensorboard = config['TENSORBOARD']
checkpoint_dir = Path(config['CHECKPOINT_DIR'], type(model).__name__)
checkpoint_dir.mkdir(exist_ok=True, parents=True)
resume_path = config['RESUME_PATH']
learning = Learning(model=model,
criterion=criterion,
optimizer=optimizer,
scheduler=scheduler,
metric_ftns=metric_ftns,
device=device,
num_epoch=num_epoch,
grad_clipping=gradient_clipping,
grad_accumulation_steps=gradient_accumulation_steps,
early_stopping=early_stopping,
validation_frequency=validation_frequency,
tensorboard=tensorboard,
checkpoint_dir=checkpoint_dir,
resume_path=resume_path)
learning.train(tqdm(train_dataloader))