def search():
if request.method == 'POST':
global resultado
local = request.form['Local']
busca = Model(tabela='pontos_turisticos')
resultado = busca.get(lugar = local)
return redirect(url_for('index'))
def login():
global vLogin
if vLogin == 0:
if request.method == 'POST':
username = request.form['username']
senha = request.form['senha']
user = Usuario(tabela='usuarios', username=username, senha=senha)
validate = user.autenticar()
if validate:
#global vLogin
global usr
global usrList
usr = username
usuario = Model(tabela='usuarios')
usrList = usuario.get(username=username)
vLogin = 1
return redirect(url_for('index'))
else:
erro = 'Login ou senha incorretos'
return render_template('Login.html', erro=erro)
if request.method == 'GET':
return render_template('Login.html')
else:
#global vLogin
vLogin = 0
return redirect(url_for('index'))
def index():
pontos = Model(tabela='pontos_turisticos')
pts = pontos.get_all()
global resultado
if len(resultado) != 0:
pts = resultado
resultado = pontos.get_all()
return render_template('index.html', len=len(pts), pontos= pts, vLogin=vLogin, usr= usr, usrList=usrList)
def index():
pontos = Model(tabela='pontos_turisticos')
pts = pontos.get_all()
print(pts)
return render_template('index.html',
len=len(pts),
pontos=pts,
vLogin=vLogin,
usr=usr,
usrList=usrList)
def verificar(self):
v=0
ponto = self.get(ponto = self.ponto)
teste0 = Model(tabela='pontos_turisticos')
teste1 = teste0.get(name=self.ponto)
if len(teste1) != 0:
for i in range(len(ponto)):
if ponto[i][3] == self.data:
v=1
if v==1:
return True
else:
return False
else:
return False
def adicionar(self):
v=0
w = self.get(name_turista = self.name_turista)
teste0 = Model(tabela='usuarios')
teste1 = teste0.get(username=self.name_turista)
if len(teste1) != 0:
for i in range(len(w)):
if w[i][1] == self.id_passeio:
v=1
if v==0:
self.save()
return True
else:
return False
else:
return False
def oferecerPasseio(self):
v=0
ponto = self.get(ponto = self.ponto)
teste0 = Model(tabela='pontos_turisticos')
teste1 = teste0.get(name=self.ponto)
if len(teste1) != 0:
for i in range(len(ponto)):
if ponto[i][3] == self.data:
v=1
if v==0:
self.save()
return True
else:
return False
else:
return False
def model_notify(model: Model, created=True) -> None:
"""
Email notification when a new Model created.
"""
recipients = [
u.email for u in User.objects.filter(
groups__name="Style Managers").exclude(email='')
]
if created:
model_status = "created"
else:
model_status = "updated"
if recipients:
domain = Site.objects.get_current().domain
mail_from = settings.DEFAULT_FROM_EMAIL
send_mail_wrapper(
_('A new Model has been %s by %s.') %
(model_status, model.creator),
_('\r\nModel name is: %s\r\nModel description is: %s\r\n'
'Link: http://%s%s\r\n') %
(model.name, model.description, domain, model.get_absolute_url()),
mail_from,
recipients,
fail_silently=True)
logging.debug('Sending email notification for %s Model, '
'recipients: %s' % (model.name, recipients))
else:
logging.warning('No recipients found for %s Model notification' %
model.name)
def like():
if request.method == 'POST':
ponto = request.form['pontolike']
id_ponto = int(request.form['id_ponto'])
busca = Model(tabela='pontos_turisticos')
resultado = busca.get(name=ponto)
connection = sqlite3.connect('database.db')
cursor = connection.cursor()
cursor.execute(
'''
SELECT * FROM like WHERE id_ponto=? AND id_usuarios=? AND likeOrDislike=0
''', (id_ponto, usrList[0][0])
)
lista1 = cursor.fetchall()
connection.close()
connection = sqlite3.connect('database.db')
cursor = connection.cursor()
cursor.execute(
'''
SELECT * FROM like WHERE id_ponto=? AND id_usuarios=? AND likeOrDislike=1
''', (id_ponto, usrList[0][0])
)
lista = cursor.fetchall()
connection.commit()
connection.close()
if len(lista) == 0:
like = resultado[0][5] + 1
dislike = resultado[0][6]
if len(lista1) != 0:
dislike = resultado[0][6] - 1
connection = sqlite3.connect('database.db')
cursor = connection.cursor()
cursor.execute(
'''
UPDATE like SET likeOrDislike=1 WHERE id_ponto=? AND id_usuarios=?
''', (id_ponto, usrList[0][0])
)
connection.commit()
connection.close()
else:
tabelaDislike = Model(tabela='like', id_usuarios= usrList[0][0], id_ponto= id_ponto, likeOrDislike= 1)
tabelaDislike.save()
connection = sqlite3.connect('database.db')
cursor = connection.cursor()
cursor.execute(
'''
UPDATE pontos_turisticos set like=?, dislike=? where name=?
''', (like, dislike, ponto)
)
connection.commit()
connection.close()
return redirect(url_for('index'))
def __init__(self):
# create model and setup callbacks
self.model = Model()
self.model.spyeworks.addCallback(self.updatePlayerOnline)
self.model.spyeworks.currentList.addCallback(self.updateCurrentList)
self.model.sensorstate.addCallback(self.updateSensorState)
# create variables for timers
self.activeTimer = Timer(1, self.restartFunc, ())
#self.playIdleList=False
# update variables with data from model
self.updatePlayerOnline(self.model.spyeworks.get())
self.updateSensorState(self.model.sensorstate.get())
def ponto(name):
ponto = Model(tabela='pontos_turisticos')
pt = ponto.get(name=name)
passeio = Model(tabela='passeios')
dados = passeio.get(ponto=name)
passeio1 = Model(tabela='passeios1')
dados1 = [None] * len(dados)
vLen = [None] * len(dados)
d = [None] * len(dados)
now = datetime.now()
n = [now.strftime("%Y"), now.strftime("%m"), now.strftime("%d")]
for i in range(len(dados)):
dados1[i] = passeio1.get(id_passeio=dados[i][0])
vLen[i] = len(dados1[i])
date = datetime.strptime(dados[i][3], '%Y-%m-%d').date()
date1 = date.strftime("%x")
now1 = now.strftime("%x")
if date1 > now1:
d[i] = 1
else:
d[i] = 0
return render_template('ponto.html', len=len(dados), dados=dados, now=now, n=n, d=d, dados1=dados1, len1=len(dados1), vLen=vLen, pontos= pt, vLogin=vLogin, usr= usr, usrList=usrList)
def ponto(name):
ponto = Model(tabela='pontos_turisticos')
pt = ponto.get(name=name)
passeio = Model(tabela='passeios')
dados = passeio.get(ponto=name)
return render_template('ponto.html',
len=len(dados),
dados=dados,
pontos=pt,
vLogin=vLogin,
usr=usr,
usrList=usrList)
def model_update_notify(model: Model, creator: User, staff: User) -> None:
"""
Email notification system when staff approved or rejected a Model
"""
recipients = [
u.email for u in User.objects.filter(
groups__name="Style Managers").exclude(email='')
]
if creator.email:
recipients += [creator.email]
if model.approved:
approval_state = 'approved'
else:
approval_state = 'rejected'
review = model.modelreview_set.last()
comment = review.comment
if recipients:
domain = Site.objects.get_current().domain
mail_from = settings.DEFAULT_FROM_EMAIL
send_mail_wrapper(
_('Model %s %s notification.') % (model, approval_state),
_('\r\nModel %s %s by %s.\r\n%s\r\nLink: http://%s%s\r\n') %
(model.name, approval_state, staff, comment, domain,
model.get_absolute_url()),
mail_from,
recipients,
fail_silently=True)
logging.debug('Sending email %s notification for %s Model, '
'recipients: %s' % (approval_state, model, recipients))
else:
logging.warning('No recipients found for %s model %s notification' %
(model, approval_state))
def __init__(self, args):
self.epoch_timer = utils.TimeIt(print_str="Epoch")
self.args = args
if self.args.training_mode == "gmm":
self.dataloader = DataLoader(self.args)
else:
if self.args.eval:
if self.args.eval_checkpoint == "":
raise ValueError(
"Eval mode is set, but no checkpoint path is provided!"
)
self.loader = torch.load(self.args.eval_checkpoint)
self.dataloader = DataLoader(self.args)
# Load the model
self.model = Model(self.args)
if self.args.eval:
self.model.load_state_dict(self.loader)
if self.args.cuda:
self.model.cuda()
self.best_test_accuracy = 0.0
self.best_test_epoch = 0
if self.args.eval is False:
if self.args.optimiser == "sgd":
self.opt = optim.SGD(
self.model.parameters(),
lr=self.args.learning_rate,
momentum=self.args.momentum,
weight_decay=self.args.weight_decay,
)
elif self.args.optimiser == "adam":
self.opt = optim.Adam(
self.model.parameters(),
lr=self.args.learning_rate,
weight_decay=self.args.weight_decay,
)
else:
raise Exception("Unknown optimiser {}".format(
self.args.optim))
if self.args.lr_scheduler:
self.lr_scheduler = optim.lr_scheduler.MultiStepLR(
self.opt,
milestones=self.args.lr_schedule,
gamma=self.args.lr_decay_factor,
)
if self.args.lr_reducer:
self.lr_reducer = torch.optim.lr_scheduler.ReduceLROnPlateau(
self.opt,
factor=np.sqrt(0.1),
cooldown=0,
patience=5,
min_lr=0.5e-6,
)
# Loss function
self.criterion = nn.CrossEntropyLoss()
self.args.logdir = os.path.join("checkpoints",
self.args.exp_name)
utils.create_dir(self.args.logdir)
if self.args.filelogger:
self.logger_path = os.path.join(
"checkpoints",
self.args.exp_name,
"%s_values.log" % self.args.exp_name,
)
self.logger = {
"train_loss_per_iter": [],
"train_loss_per_epoch": [],
"val_loss_per_iter": [],
"val_loss_per_epoch": [],
"val_accuracy_per_iter": [],
"val_accuracy_per_epoch": [],
"test_loss": [],
"test_accuracy": [],
"best_epoch": 0,
"best_test_accuracy": 0.0,
"ssl_loss": [],
"ssl_accuracy": [],
"ssl_correct": [],
}
if self.args.tensorboard:
self.writer = SummaryWriter(log_dir=self.args.logdir,
flush_secs=30)
self.writer.add_text("Arguments",
params.print_args(self.args))
def __init__(self):
logger = Logger()
self.log = logger.logger
self.error_log = logger.err_logger
self.camera = None
try:
self.log.info("Reading the config...")
self.config = ConfigReader()
self.log.info("Config read")
except:
self.log.error("Error reading config.ini")
self.error_log.error("Error reading config.ini", exc_info=True)
sys.exit()
try:
self.folder_path = self.config.get_folder_path_config()
print("folder_path: {}".format(self.folder_path))
except:
self.log.error(
"Initialisation error: the video folder path is not defined")
self.error_log.error(
"Initialisation error: the video folder path is not defined",
exc_info=True)
sys.exit()
self.log.info("Initialising face_model")
try:
face_model_address, face_model_protos = self.config.get_model_config(
)
print(face_model_address, face_model_protos)
self.get_face_model = Model(face_model_address,
face_model_protos,
num_classes=2)
self.get_face_model.get_session()
self.log.info("face_model initialisation completed")
except:
self.log.error("face_model initialisation error")
self.error_log.error("face_model initialisation error",
exc_info=True)
sys.exit()
self.log.info("Initialising camera")
try:
self.camera = Camera(self.folder_path)
self.log.info("Camera initialised")
except:
self.log.error("Camera initialisation error")
self.error_log.error("Camera initialisation error", exc_info=True)
sys.exit()
try:
self.max_boxes_to_draw, self.min_score_thresh = self.config.get_vis_utils_config(
)
except:
self.max_boxes_to_draw, self.min_score_thresh = 10, 0.3
self.log.info("Initializing detector")
try:
self.detector = Detector(self.get_face_model,
self.max_boxes_to_draw,
self.min_score_thresh)
self.log.info("detector initialized")
except:
self.log.error("detector initialization failed")
self.error_log.error("detector initialization failed",
exc_info=True)
sys.exit()
try:
self.log.info("Initialising embeddings models...")
self.calculator = Calculate(self.log, self.error_log, self.config)
self.log.info("embeddings Models initialisation completed")
except:
self.log.error("embeddings Models initialisation error")
self.error_log.error("embeddings Models initialisation error",
exc_info=True)
sys.exit()
self.batch_frame_list = []
# todo set batch_size to config.ini ?
self.batch_size = 2
self.flag = True
def build_model(self):
self.net: nn.Module = Model(self.backbone, self.num_color,
self.num_style, self.num_season,
self.num_category).to(self.device)
self.net.eval()
self.load_model()
def search():
local = request.form['Local']
busca = Model(tabela='pontos_turisticos')
resultado = busca.get(lugar=local)
return render_template('/Pesquisa local')
opt.nThreads = 1 # test code only supports nThreads = 1
opt.batchSize = 1 # test code only supports batchSize = 1 (1 image at a time)
opt.serial_batches = True
opt.gpu_ids = [opt.gpu_ids[0]]
if 'vrd' in opt.dataroot:
test_dataset = VrdDataset(opt.dataroot,
split='test',
net=opt.feat_net,
use_gt=opt.use_gt,
use_lang=opt.use_lang)
else:
print('No this dataset')
sys.exit(1)
test_data_loader = DataLoader(test_dataset,
batch_size=opt.batchSize,
shuffle=False,
num_workers=int(opt.nThreads))
model = Model(opt)
test_net(model, opt, test_data_loader, test_dataset.name)
if not opt.no_evaluate:
with open(
os.path.join(opt.results_dir, opt.name, test_dataset.name,
'predicted_predicate.pkl'), 'rb') as f:
rel_result = pickle.load(f)
test_dataset.evaluate(rel_result,
n_rels=opt.n_rels,
obj_co_occur=model.relevance)
#test_dataset.evaluate(rel_result, n_rels=opt.n_rels)
# if isinstance(net, torch.nn.ReLU):
# net.register_forward_hook(relu_hook)
# if isinstance(net, torch.nn.MaxPool2d) or isinstance(net, torch.nn.AvgPool2d):
# net.register_forward_hook(pooling_hook)
# if isinstance(net, torch.nn.Upsample):
# net.register_forward_hook(upsample_hook)
# return
# for c in childrens:
# foo(c)
# if model == None:
# model = torchvision.models.alexnet()
# foo(model)
# # input = Variable(torch.rand(3,input_res,input_res).unsqueeze(0), requires_grad = True)
# input = Variable(torch.rand(3,3,input_res,input_res), requires_grad = True)
# out = model(input)
#
total_flops = (sum(list_conv) + sum(list_linear) + sum(list_bn) + sum(list_relu) + sum(list_pooling) + sum(list_upsample))
print(' + Number of FLOPs: %.2fG' % (total_flops / 1e9))
return total_flops
if __name__ == '__main__':
model = Model()
# input=torch.rand(1,3,400,400)
# out = Model(input)
print_model_param_nums(model)
count_model_param_flops(model,400, True)
from models.models import Model
from mxnet import gpu, nd
import pickle
import os
from tqdm import trange, tqdm
model_param = {'emb_size': 200, 'hidden_size': 200}
vocab_path = 'data/vocab'
source_path = 'data/cnn_articles'
target_path = 'data/cnn_abstracts'
lda_path = 'data/cnn_head_lda'
model = Model(model_param,
vocab_path,
head_attention=True,
decoder_cell='dlstm')
model.train(source_path,
target_path,
lda_path,
epoch_num=50,
learning_rate=0.00001)
class Person_processing:
def __init__(self):
logger = Logger()
self.log = logger.logger
self.error_log = logger.err_logger
self.camera = None
try:
self.log.info("Reading the config...")
self.config = ConfigReader()
self.log.info("Config read")
except:
self.log.error("Error reading config.ini")
self.error_log.error("Error reading config.ini", exc_info=True)
sys.exit()
try:
self.folder_path = self.config.get_folder_path_config()
print("folder_path: {}".format(self.folder_path))
except:
self.log.error(
"Initialisation error: the video folder path is not defined")
self.error_log.error(
"Initialisation error: the video folder path is not defined",
exc_info=True)
sys.exit()
self.log.info("Initialising face_model")
try:
face_model_address, face_model_protos = self.config.get_model_config(
)
print(face_model_address, face_model_protos)
self.get_face_model = Model(face_model_address,
face_model_protos,
num_classes=2)
self.get_face_model.get_session()
self.log.info("face_model initialisation completed")
except:
self.log.error("face_model initialisation error")
self.error_log.error("face_model initialisation error",
exc_info=True)
sys.exit()
self.log.info("Initialising camera")
try:
self.camera = Camera(self.folder_path)
self.log.info("Camera initialised")
except:
self.log.error("Camera initialisation error")
self.error_log.error("Camera initialisation error", exc_info=True)
sys.exit()
try:
self.max_boxes_to_draw, self.min_score_thresh = self.config.get_vis_utils_config(
)
except:
self.max_boxes_to_draw, self.min_score_thresh = 10, 0.3
self.log.info("Initializing detector")
try:
self.detector = Detector(self.get_face_model,
self.max_boxes_to_draw,
self.min_score_thresh)
self.log.info("detector initialized")
except:
self.log.error("detector initialization failed")
self.error_log.error("detector initialization failed",
exc_info=True)
sys.exit()
try:
self.log.info("Initialising embeddings models...")
self.calculator = Calculate(self.log, self.error_log, self.config)
self.log.info("embeddings Models initialisation completed")
except:
self.log.error("embeddings Models initialisation error")
self.error_log.error("embeddings Models initialisation error",
exc_info=True)
sys.exit()
self.batch_frame_list = []
# todo set batch_size to config.ini ?
self.batch_size = 2
self.flag = True
def get_filenames_list(self):
"""
get filenames from folder
:return: string (video_file, scan, rfid)
"""
video_name = None
files = [
f for f in os.listdir(self.folder_path)
if os.path.isfile(os.path.join(self.folder_path, f))
]
for file in files:
if ".txt" in file:
return None
if ".avi" in file:
video_name = file
return video_name
def get_all_frames(self, video_name):
full_video_path = os.path.join(self.folder_path, video_name)
# print("reading {}".format(full_video_path))
try:
self.camera.camera_capture.open(full_video_path)
self.log.debug("reading {}".format(full_video_path))
except:
self.log.error("Cannot open video file {}".format(full_video_path))
self.error_log.error(
"Cannot open video file {}".format(full_video_path),
exc_info=True)
# read frames to frame_list
frame_list = []
while True:
frame, ret = self.camera.get_frame()
if not ret:
break
# rgb_2_bgr
frame = self.camera.rgb_to_bgr(frame)
frame_list.append(frame)
# print(frame.shape)
# cv2.imshow('window_name', frame)
# cv2.waitKey(300)
# cv2.destroyAllWindows()
# after reading video file, close reader to avoid PermissionError
try:
self.camera.camera_capture.release()
except:
self.log.error(
"Cannot close video file {}".format(full_video_path))
self.error_log.error(
"Cannot close video file {}".format(full_video_path),
exc_info=True)
return frame_list
def get_faces(self, frame_list):
"""
Get all faces from video
:param frame_list: list of frames
:return:
"""
face_list = []
while len(frame_list) > 0:
if len(frame_list) < self.batch_size: # if len < n skip
break
else:
# print("detect")
batch = frame_list[:self.batch_size]
frame_list = frame_list[self.batch_size:]
face_batch = self.detector.detect(batch)
face_list = face_list + face_batch
return face_list
def process(self):
print('in process')
while True:
# read filenames from folder
video_name = self.get_filenames_list()
if video_name is None:
continue
# for video file get all frames
try:
frame_list = self.get_all_frames(video_name)
except:
self.log.error("Cannot read video file {}".format(video_name))
self.error_log.error(
"Cannot read video file {}".format(video_name),
exc_info=True)
continue
# for frames_list get face_list (detect face and corp)
try:
video_face_list = self.get_faces(frame_list)
except:
self.log.error("Cannot get face, detector error for {}".format(
video_name))
self.error_log.error(
"Cannot get face, detector error for {}".format(
video_name),
exc_info=True)
continue
# for frame in video_face_list:
# print(frame.shape)
# cv2.imshow('window_name', frame)
# cv2.waitKey(100)
# cv2.destroyAllWindows()
# # # embeddings
# read scan & rfid images
scan_face = cv2.imread(os.path.join(self.folder_path, 'scan.bmp'))
rfid_face = cv2.imread(os.path.join(self.folder_path, 'rfid.jpg'))
# get video, scan & rfid hash
video_face_hash = self.calculator.calculate_faces(video_face_list)
scan_face_hash = self.calculator.calculate_faces(
np.expand_dims(scan_face, axis=0))
rfid_face_hash = self.calculator.calculate_faces(
np.expand_dims(rfid_face, axis=0))
# distance
scan_dist = 1 - self.count_distance(video_face_hash,
scan_face_hash)
rfid_dist = 1 - self.count_distance(video_face_hash,
rfid_face_hash)
print('----')
print(scan_dist, rfid_dist)
print('----')
# todo проверить как будет менятся дистанция при обрезании края изображения для scan и rfid или увеличения краёв при обнаружении лиц
self.save_results(scan_dist, rfid_dist)
def count_distance(self, hash_1, hash_2):
# Count euclidian distance matrix, between objects and faces in one image
dist = distance.euclidean(hash_1, hash_2)
return dist
def save_results(self, scan_dist, rfid_dist):
"""
save info about obj. recognitions to .txt file
:return: None
"""
save_path = os.path.join(self.folder_path, 'res.txt')
try:
text_file = open(save_path, "x")
except FileExistsError:
self.log.error(
"save_results: FileExistsError {}".format(save_path))
self.error_log.error(
"save_results: FileExistsError {}".format(save_path),
exc_info=True)
pass
except FileNotFoundError:
self.log.error(
"save_results: FileNotFoundError {}".format(save_path))
self.error_log.error(
"save_results: FileNotFoundError {}".format(save_path),
exc_info=True)
pass
text_file.write(str(scan_dist) + '\n' + str(rfid_dist))
def passeios():
if usrList[0][4] == 'Guia':
guia = usr
passeio = Model(tabela='passeios')
dados = passeio.get(guia=guia)
passeio1 = Model(tabela='passeios1')
dados1 = [None] * len(dados)
vLen = [None] * len(dados)
d = [None] * len(dados)
now = datetime.now()
n = [now.strftime("%Y"), now.strftime("%m"), now.strftime("%d")]
for i in range(len(dados)):
dados1[i] = passeio1.get(id_passeio=dados[i][0])
vLen[i] = len(dados1[i])
date = datetime.strptime(dados[i][3], '%Y-%m-%d').date()
date1 = date.strftime("%x")
now1 = now.strftime("%x")
if date1 > now1:
d[i] = 1
else:
d[i] = 0
return render_template('passeios.html', len=len(dados), dados=dados, now=now, n=n, d=d, dados1=dados1, len1=len(dados1), vLen=vLen, vLogin=vLogin, usr= usr, usrList=usrList)
if usrList[0][4] == 'Turista':
a = Model(tabela='passeios1')
b = a.get(name_turista=usr)
a1 = Model(tabela='passeios')
lenV = [None] * len(b)
V = [None] * len(b)
for i in range(len(b)):
V[i] = a1.get(id=b[i][1])
lenV[i] = len(V[i])
print(lenV)
return render_template('passeios.html', V=V, len=len(b), lenV=lenV, vLogin=vLogin, usr= usr, usrList=usrList)
from models.models import Model
from mxnet import gpu, nd, cpu
import pickle
import os
from tqdm import trange
from models.vocab import Vocab
model_param = {'emb_size': 200, 'hidden_size': 200}
vocab_path = 'data/vocab'
source_path = 'data/cnn_articles'
target_path = 'data/cnn_abstracts'
lda_path = 'data/cnn_head_lda'
model = Model(model_param,
vocab_path,
mode='decode',
head_attention=True,
decoder_cell='dlstm',
ctx=cpu())
res = model.decode(source_path, lda_path, 'best.model')
res = [int(i.asscalar()) for i in res.tokens]
vocab = Vocab(vocab_path)
res = [vocab.id2word(i) for i in res]
print(' '.join(res))
abstract = pickle.load(
open(os.path.join(target_path,
os.listdir(target_path)[0]), 'rb'))
class ModelTrainer:
"""Class for training and testing of model"""
def __init__(self, args):
self.epoch_timer = utils.TimeIt(print_str="Epoch")
self.args = args
if self.args.training_mode == "gmm":
self.dataloader = DataLoader(self.args)
else:
if self.args.eval:
if self.args.eval_checkpoint == "":
raise ValueError(
"Eval mode is set, but no checkpoint path is provided!"
)
self.loader = torch.load(self.args.eval_checkpoint)
self.dataloader = DataLoader(self.args)
# Load the model
self.model = Model(self.args)
if self.args.eval:
self.model.load_state_dict(self.loader)
if self.args.cuda:
self.model.cuda()
self.best_test_accuracy = 0.0
self.best_test_epoch = 0
if self.args.eval is False:
if self.args.optimiser == "sgd":
self.opt = optim.SGD(
self.model.parameters(),
lr=self.args.learning_rate,
momentum=self.args.momentum,
weight_decay=self.args.weight_decay,
)
elif self.args.optimiser == "adam":
self.opt = optim.Adam(
self.model.parameters(),
lr=self.args.learning_rate,
weight_decay=self.args.weight_decay,
)
else:
raise Exception("Unknown optimiser {}".format(
self.args.optim))
if self.args.lr_scheduler:
self.lr_scheduler = optim.lr_scheduler.MultiStepLR(
self.opt,
milestones=self.args.lr_schedule,
gamma=self.args.lr_decay_factor,
)
if self.args.lr_reducer:
self.lr_reducer = torch.optim.lr_scheduler.ReduceLROnPlateau(
self.opt,
factor=np.sqrt(0.1),
cooldown=0,
patience=5,
min_lr=0.5e-6,
)
# Loss function
self.criterion = nn.CrossEntropyLoss()
self.args.logdir = os.path.join("checkpoints",
self.args.exp_name)
utils.create_dir(self.args.logdir)
if self.args.filelogger:
self.logger_path = os.path.join(
"checkpoints",
self.args.exp_name,
"%s_values.log" % self.args.exp_name,
)
self.logger = {
"train_loss_per_iter": [],
"train_loss_per_epoch": [],
"val_loss_per_iter": [],
"val_loss_per_epoch": [],
"val_accuracy_per_iter": [],
"val_accuracy_per_epoch": [],
"test_loss": [],
"test_accuracy": [],
"best_epoch": 0,
"best_test_accuracy": 0.0,
"ssl_loss": [],
"ssl_accuracy": [],
"ssl_correct": [],
}
if self.args.tensorboard:
self.writer = SummaryWriter(log_dir=self.args.logdir,
flush_secs=30)
self.writer.add_text("Arguments",
params.print_args(self.args))
def train_val(self, epoch):
"""Train the model for one epoch and evaluate on val split if log_intervals have passed"""
for batch_idx, batch in enumerate(self.dataloader.train_loader):
self.model.train()
self.opt.zero_grad()
self.iter += 1
images, targets, indices = batch
if self.args.cuda:
images, targets = images.cuda(), targets.cuda()
logits, unnormalised_scores = self.model(images)
loss = self.criterion(unnormalised_scores, targets)
loss.backward()
self.opt.step()
if batch_idx % self.args.log_interval == 0:
val_loss, val_acc = self.evaluate("Val", n_batches=4)
train_loss, val_loss, val_acc = utils.convert_for_print(
loss, val_loss, val_acc)
if self.args.filelogger:
self.logger["train_loss_per_iter"].append(
[self.iter, train_loss])
self.logger["val_loss_per_iter"].append(
[self.iter, val_loss])
self.logger["val_accuracy_per_iter"].append(
[self.iter, val_acc])
if self.args.tensorboard:
self.writer.add_scalar("Loss_at_Iter/Train", train_loss,
self.iter)
self.writer.add_scalar("Loss_at_Iter/Val", val_loss,
self.iter)
self.writer.add_scalar("Accuracy_at_Iter/Val", val_acc,
self.iter)
examples_this_epoch = batch_idx * len(images)
epoch_progress = 100.0 * batch_idx / len(
self.dataloader.train_loader)
print("Train Epoch: %3d [%5d/%5d (%5.1f%%)]\t "
"Train Loss: %0.6f\t Val Loss: %0.6f\t Val Acc: %0.1f" %
(
epoch,
examples_this_epoch,
len(self.dataloader.train_loader.dataset),
epoch_progress,
train_loss,
val_loss,
val_acc,
))
if self.args.lr_reducer:
val_loss, val_acc = self.evaluate("Val", n_batches=None)
self.lr_reducer.step(val_loss)
val_loss, val_acc = utils.convert_for_print(val_loss, val_acc)
if self.args.filelogger:
self.logger["train_loss_per_epoch"].append([epoch, train_loss])
self.logger["val_loss_per_epoch"].append([epoch, val_loss])
self.logger["val_accuracy_per_epoch"].append([epoch, val_acc])
if self.args.tensorboard:
self.writer.add_scalar("Loss_at_Epoch/Train", train_loss, epoch)
self.writer.add_scalar("Loss_at_Epoch/Val", val_loss, epoch)
self.writer.add_scalar("Accuracy_at_Epoch/Val", val_acc, epoch)
def evaluate(self, split, epoch=None, verbose=False, n_batches=None):
"""Evaluate model on val or test data"""
self.model.eval()
with torch.no_grad():
loss = 0
correct = 0
n_examples = 0
if split == "Val":
loader = self.dataloader.val_loader
elif split == "Test":
loader = self.dataloader.test_loader
for batch_idx, batch in enumerate(loader):
images, targets, _ = batch
if args.cuda:
images, targets = images.cuda(), targets.cuda()
logits, unnormalised_scores = self.model(images)
loss += F.cross_entropy(unnormalised_scores,
targets,
reduction="sum")
pred = logits.max(1, keepdim=False)[1]
correct += pred.eq(targets).sum()
n_examples += pred.shape[0]
if n_batches and (batch_idx >= n_batches):
break
loss /= n_examples
acc = 100.0 * correct / n_examples
if split == "Test" and acc >= self.best_test_accuracy:
self.best_test_accuracy = utils.convert_for_print(acc)
self.best_test_epoch = epoch
if self.args.filelogger:
self.logger["best_epoch"] = self.best_test_epoch
self.logger["best_test_accuracy"] = self.best_test_accuracy
if verbose:
if epoch is None:
epoch = 0
self.best_test_epoch = 0
loss, acc = utils.convert_for_print(loss, acc)
print(
"\n%s set Epoch: %2d \t Average loss: %0.4f, Accuracy: %d/%d (%0.1f%%)"
% (split, epoch, loss, correct, n_examples, acc))
print(
"Best %s split Performance: Epoch %d - Accuracy: %0.1f%%" %
(split, self.best_test_epoch, self.best_test_accuracy))
if self.args.filelogger:
self.logger["test_loss"].append([epoch, loss])
self.logger["test_accuracy"].append([epoch, acc])
if self.args.tensorboard:
self.writer.add_scalar("Loss_at_Epoch/Test", loss, epoch)
self.writer.add_scalar("Accuracy_at_Epoch/Test", acc,
epoch)
self.writer.add_scalar(
"Accuracy_at_Epoch/Best_Test_Accuracy",
self.best_test_accuracy,
self.best_test_epoch,
)
return loss, acc
def generate_labels_for_ssl(self, epoch, n_batches=None, verbose=False):
self.model.eval()
with torch.no_grad():
loss = 0
correct = 0
n_examples = 0
predictions_indices = []
predictions_labels = []
loader = self.dataloader.unsupervised_train_loader
for batch_idx, batch in enumerate(loader):
images, targets, indices = batch
if args.cuda:
images, targets = images.cuda(), targets.cuda()
logits, unnormalised_scores = self.model(images)
loss += F.cross_entropy(unnormalised_scores,
targets,
reduction="sum")
pred = logits.max(1, keepdim=False)[1]
correct += pred.eq(targets).sum()
n_examples += pred.shape[0]
predictions_indices.extend(indices.tolist())
predictions_labels.extend(pred.tolist())
if n_batches and (batch_idx >= n_batches):
break
loss /= n_examples
acc = 100.0 * correct / n_examples
if verbose:
loss, acc, correct = utils.convert_for_print(
loss, acc, correct)
print(
"\nLabel Generation Performance Average loss: %0.4f, Accuracy: %d/%d (%0.1f%%)"
% (loss, correct, n_examples, acc))
# TODO: Add logging
if self.args.filelogger:
self.logger["ssl_loss"].append([epoch, loss])
self.logger["ssl_accuracy"].append([epoch, acc])
self.logger["ssl_correct"].append([epoch, correct])
if self.args.tensorboard:
self.writer.add_scalar("SSL_Loss_at_Epoch", loss, epoch)
self.writer.add_scalar("SSL_Accuracy_at_Epoch", acc, epoch)
self.writer.add_scalar("SSL_Correct_Labels_at_Epoch",
correct, epoch)
return predictions_indices, predictions_labels
def train_val_test(self):
""" Function to train, validate and evaluate the model"""
self.iter = 0
for epoch in range(1, self.args.epochs + 1):
self.train_val(epoch)
self.evaluate("Test", epoch, verbose=True)
if self.args.lr_scheduler:
self.lr_scheduler.step()
if epoch % self.args.checkpoint_save_interval == 0:
print("Saved %s/%s_epoch%d.pt\n" %
(self.args.logdir, self.args.exp_name, epoch))
torch.save(
self.model.state_dict(),
"%s/%s_epoch%d.pt" %
(self.args.logdir, self.args.exp_name, epoch),
)
self.epoch_timer.tic(verbose=True)
if self.args.tensorboard:
if self.args.filelogger:
text = "Epoch: %d Test Accuracy:%0.1f" % (
self.logger["best_epoch"],
self.logger["best_test_accuracy"],
)
self.writer.add_text("Best Test Performance", text)
self.writer.close()
if self.args.filelogger:
utils.write_log_to_json(self.logger_path, self.logger)
self.epoch_timer.time_since_init(print_str="Total")
def ssl_train_val_test(self):
""" Function to train, validate and evaluate the model"""
self.iter = 0
predictions_indices, predictions_labels = [], []
self.dataloader.stop_label_generation = False
for epoch in range(1, self.args.epochs + 1):
if not self.dataloader.stop_label_generation:
self.dataloader.ssl_init_epoch(predictions_indices,
predictions_labels)
self.train_val(epoch)
self.evaluate("Test", epoch, verbose=True)
if not self.dataloader.stop_label_generation:
predictions_indices, predictions_labels = self.generate_labels_for_ssl(
epoch, n_batches=4, verbose=True)
if self.args.lr_scheduler:
self.lr_scheduler.step()
if epoch % self.args.checkpoint_save_interval == 0:
print("Saved %s/%s_epoch%d.pt\n" %
(self.args.logdir, self.args.exp_name, epoch))
torch.save(
self.model.state_dict(),
"%s/%s_epoch%d.pt" %
(self.args.logdir, self.args.exp_name, epoch),
)
self.epoch_timer.tic(verbose=True)
if self.args.tensorboard:
if self.args.filelogger:
text = "Epoch: %d Test Accuracy:%0.1f" % (
self.logger["best_epoch"],
self.logger["best_test_accuracy"],
)
self.writer.add_text("Best Test Performance", text)
self.writer.close()
if self.args.filelogger:
utils.write_log_to_json(self.logger_path, self.logger)
self.epoch_timer.time_since_init(print_str="Total")
def gmm_train_val_test(self):
train_data = self.dataloader.full_supervised_train_dataset.train_data
train_labels = self.dataloader.full_supervised_train_dataset.train_labels
train_data = train_data / 255
mean = np.array(
self.args.cifar10_mean_color)[np.newaxis][np.newaxis][np.newaxis]
std = np.array(
self.args.cifar10_std_color)[np.newaxis][np.newaxis][np.newaxis]
train_data = (train_data - mean) / std
train_data = train_data.reshape(train_data.shape[0], -1)
cv_types = ["spherical", "diag", "full", "tied"]
for cv_type in cv_types:
gmm = mixture.GaussianMixture(n_components=10,
covariance_type=cv_type)
gmm.fit(train_data)
clusters = gmm.predict(train_data)
labels = np.zeros_like(clusters)
for i in range(10):
mask = clusters == i
labels[mask] = mode(train_labels[mask])[0]
correct1 = np.equal(clusters, train_labels).sum()
correct2 = np.equal(labels, train_labels).sum()
print("%d/49000 (%0.2f%%)" % (correct1, correct1 / 49000))
print("%d/49000 (%0.2f%%)" % (correct2, correct2 / 49000))
def perfil():
a = Model(tabela='like')
b = a.get(id_usuarios=usrList[0][0])
pontos = Model(tabela='pontos_turisticos')
pts = pontos.get_all()
return render_template('likes.html', b=b, len=len(pts), len1=len(b), pontos= pts, vLogin=vLogin, usr= usr, usrList=usrList)
test_probabilitiesBkg = []
for b in bkg_test:
b = scaler.transform([b])
prob = classifier.predict_proba(b)[0][0]
b = b[0].flatten().tolist()
test_probabilitiesBkg.append(prob)
testDataBkg.append(b)
batchSize = 4
test = testDataSig
probs_test = test_probabilitiesSig
test_bkg = testDataBkg
probs_test_bkg = test_probabilitiesBkg
# Initialise model
model = Model(len(test[0]), 1)
model.load_model('../scripts/approx1.pkl')
test = np.array(test)
probs_test = np.array(probs_test)
test_bkg = np.array(test_bkg)
probs_test_bkg = np.array(probs_test_bkg)
# Make a vector of outputs
comp_preds = []
comp_true = []
for (batchX, batchY) in next_batch(test, probs_test, batchSize):
if batchY.shape[0] < batchSize:
print 'Batch size insufficient (%s), continuing...' % batchY.shape[0]
continue
num_training = int(len(dataset) * 0.8)
num_val = int(len(dataset) * 0.1)
num_test = len(dataset) - (num_training + num_val)
training_set, validation_set, test_set = random_split(
dataset, [num_training, num_val, num_test])
train_loader = DataLoader(training_set,
batch_size=args.batch_size,
shuffle=True)
val_loader = DataLoader(validation_set,
batch_size=args.batch_size,
shuffle=False)
test_loader = DataLoader(test_set, batch_size=args.batch_size, shuffle=False)
model = Model(args).to(args.device)
optimizer = torch.optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
def train():
min_loss = 1e10
patience_cnt = 0
val_loss_values = []
best_epoch = 0
t = time.time()
model.train()
for epoch in range(args.epochs):
loss_train = 0.0
def train():
opt = TrainOptions().parse()
train_dataset = VrdDataset(opt.dataroot,
split='train',
net=opt.feat_net,
use_lang=opt.use_lang)
train_data_loader = DataLoader(train_dataset,
batch_size=opt.batchSize,
shuffle=not opt.serial_batches,
num_workers=int(opt.nThreads))
val_dataset = VrdDataset(opt.dataroot,
split='val',
net=opt.feat_net,
use_lang=opt.use_lang)
val_data_loader = DataLoader(val_dataset,
batch_size=opt.batchSize,
shuffle=opt.serial_batches,
num_workers=int(opt.nThreads))
model = Model(opt)
total_steps = 0
batch = 0
n_train_batches = len(train_data_loader)
for epoch in range(opt.epoch_count, opt.niter + opt.epoch_count):
loss_temp = 0
epoch_start_time = time.time()
epoch_iter = 0
for i_batch, data_dict in enumerate(train_data_loader):
batch += 1
if opt.loss == 'kl':
alpha = model.update_alpha(batch * 1. / n_train_batches)
total_steps += opt.batchSize
epoch_iter += opt.batchSize
model.set_input(data_dict)
model.optimize()
loss = model.get_loss()
loss_temp = loss_temp + loss * opt.batchSize
# print statistics
if epoch_iter % opt.print_epoch_iter_freq == 0:
if epoch_iter > 0:
loss_temp = loss_temp / opt.print_epoch_iter_freq
#print('epoch: {:d}, epoch_iter: {:d}, loss: {:.3f}'.format(epoch, epoch_iter, loss.cpu().data[0]))
print(
'Epoch: {:d} \t Epoch_iter: {:d} \t Training Loss: {:.4f}'.
format(epoch, epoch_iter, loss_temp))
loss_temp = 0
#if total_steps % opt.save_latest_freq == 0:
#print('saving the latest model (epoch {:d}, total_steps {:d})'.format(epoch, total_steps))
#model.save_model('latest')
if epoch % opt.val_epoch_freq == 0:
val_loss, val_true_loss = validate(model, val_data_loader, opt)
print(
'=============== Epoch: {:d} \t Validation Loss: {:.4f} \t True Loss : {:.4f} ==============='
.format(epoch, val_loss, val_true_loss))
#lr = model.update_learning_rate(val_loss)
lr = model.update_learning_rate(val_loss)
if opt.loss == 'kl':
print(
'[ End of epoch {:d} / {:d} \t Time Taken: {:f} sec \t Learning rate: {:.2e} \t alpha: {:.2e}]'
.format(epoch, opt.niter + opt.epoch_count - 1,
time.time() - epoch_start_time, lr, alpha))
else:
print(
'[ End of epoch {:d} / {:d} \t Time Taken: {:f} sec \t Learning rate: {:.2e}]'
.format(epoch, opt.niter + opt.epoch_count - 1,
time.time() - epoch_start_time, lr))
if epoch % opt.save_epoch_freq == 0:
print('Saving the model at the end of epoch {:d} \t iters {:d}'.
format(epoch, total_steps))
#model.save_model('latest')
model.save_model(epoch)