def scoreUpdate(self, **kw): print "-------------" print kw id = kw['id'] value = kw['value'] score = Score.get(int(id)) crit = score.criterion assessment = score.assessment category = crit.category score.value = value user = identity.current.user eh = EditHistory.selectBy(assessmentID=assessment.id, category=category) if eh.count() > 0: eh[0].editor = user.firstlast() eh[0].program = user.program.name else: EditHistory(assessmentID=assessment.id, category=category, editor=user.firstlast(), program=user.program.name) if crit.prefix in ['Y','Z']: st_crit = Criterion.selectBy(level=assessment.level, category=category, prefix="Z")[0] na_crit = Criterion.selectBy(level=assessment.level, category=category, prefix="Y")[0] possible = util.scorePossible(category, assessment.level) subtotal = Score.selectBy(criterionID=st_crit.id, assessmentID=assessment.id)[0].value na = Score.selectBy(criterionID=na_crit.id, assessmentID=assessment.id)[0].value return str(util.percent(subtotal,possible,na)) return dict()
def __init__(self, parser_params, device):
"""
:param parser_params: parser args
"""
num_hidden = [int(x) for x in parser_params.num_hidden.split(',')]
self.seq_length = parser_params.seq_length
self.batch_size = parser_params.batch_size
self.patch_size = parser_params.patch_size
self.num_layers = len(num_hidden)
networks_map = {'BiLSTM': BiLSTM.RNN, 'Bi-LSTM3': BiLSTM3.RNN}
if parser_params.model_name in networks_map:
if parser_params.LSTM_pretrained:
print("Loading LSTM pretrained model...")
Network = networks_map[parser_params.model_name]
self.network = Network(
self.num_layers, num_hidden, parser_params.seq_length,
parser_params.patch_size, parser_params.batch_size,
parser_params.img_size, parser_params.img_channel,
parser_params.filter_size, parser_params.stride)
self.network.load_state_dict(
torch.load(parser_params.LSTM_pretrained))
self.network = DataParallel(self.network, device_ids=[0, 1, 2])
self.network.to(device)
### freeze weight
for param in self.network.parameters():
param.requires_grad = False
else:
Network = networks_map[parser_params.model_name]
self.network = Network(
self.num_layers, num_hidden, parser_params.seq_length,
parser_params.patch_size, parser_params.batch_size,
parser_params.img_size, parser_params.img_channel,
parser_params.filter_size, parser_params.stride)
self.network = DataParallel(self.network, device_ids=[0, 1, 2])
self.network.to(device)
# Fine tune network
Network2 = networks_map[parser_params.model_name]
self.network2 = Network2(
self.num_layers, num_hidden, parser_params.seq_length,
parser_params.patch_size, parser_params.batch_size,
parser_params.img_size, parser_params.img_channel,
parser_params.filter_size, parser_params.stride)
self.network2 = DataParallel(self.network2, device_ids=[0, 1, 2])
self.network2.to(device)
else:
raise ValueError('Name of network unknown {}'.format(
parser_params.model_name))
self.criterion = Criterion.Loss()
self.optimizer2 = Adam(self.network2.parameters(), lr=parser_params.lr)
def fillUpdate(self, **kw):
assessment = Assessment.get(cherrypy.session['current_assessment'])
category = cherrypy.session['current_category']
fill = kw['fill']
value = kw['values']
cherrypy.session['fill'] = fill
cherrypy.session['values'] = value
scores = assessment.scores
criteria = Criterion.selectBy(category=category, level=assessment.level)
if value == 'previous':
sorted_assessments = Assessment.select(AND(Assessment.q.childID==assessment.childID,
Assessment.q.id!=assessment.id), orderBy='-dateEntered')
previous = sorted_assessments[0]
for c in criteria:
score = Score.selectBy(criterionID=c.id, assessmentID=assessment.id, type='score')
if score.count() == 0:
if value == 'previous':
temp = Score.selectBy(criterionID=c.id, assessmentID=previous.id, type='score')
if temp.count() > 0:
Score(value=temp[0].value, criterionID=c.id, assessmentID=assessment.id, type='score')
elif value == 'zeros':
Score(value="0", criterionID=c.id, assessmentID=assessment.id, type='score')
elif value == 'twos':
Score(value="2", criterionID=c.id, assessmentID=assessment.id, type='score')
elif score[0].value not in ['0','1','2','#'] or fill == 'all':
if value == 'previous':
temp = Score.selectBy(criterionID=c.id, assessmentID=previous.id, type='score')
if temp.count() > 0:
score[0].value = temp[0].value
elif value == 'zeros':
score[0].value = "0"
elif value == 'twos':
score[0].value = "2"
user = identity.current.user
eh = EditHistory.selectBy(assessmentID=assessment.id, category=category)
if eh.count() > 0:
eh[0].editor = user.firstlast()
eh[0].program = user.program.name
else:
EditHistory(assessmentID=assessment.id, category=category, editor=user.firstlast(), program=user.program.name)
raise redirect("/assessment")
def __init__(self, parser_params, device):
"""
:param parser_params: parser args
"""
num_hidden = [int(x) for x in parser_params.num_hidden.split(',')]
self.seq_length = parser_params.seq_length
self.batch_size = parser_params.batch_size
self.patch_size = parser_params.patch_size
self.num_layers = len(num_hidden)
self.CA_patch_size = parser_params.CA_patch_size
networks_map = {'BiLSTM': BiLSTM.RNN, 'Bi-LSTM3': BiLSTM3.RNN}
if parser_params.model_name in networks_map:
if parser_params.LSTM_pretrained:
print("Loading LSTM pretrained model...")
self.network = torch.load(parser_params.LSTM_pretrained)
#### freeze weight
for param in self.network.parameters():
param.requires_grad = False
else:
Network = networks_map[parser_params.model_name]
self.network = Network(
self.num_layers, num_hidden, parser_params.seq_length,
parser_params.patch_size, parser_params.batch_size,
parser_params.img_size, parser_params.img_channel,
parser_params.filter_size, parser_params.stride)
self.network = DataParallel(self.network, device_ids=[0, 1])
self.network.to(device)
self.CA = ResModule.RES(n_resgroups=parser_params.n_resgroups,
n_resblocks=parser_params.n_resblocks,
n_channel=parser_params.img_channel *
(parser_params.CA_patch_size**2))
self.CA = DataParallel(self.CA, device_ids=[0, 1])
self.CA.to(device)
else:
raise ValueError('Name of network unknown {}'.format(
parser_params.model_name))
# self.optimizer = Adam(self.network.parameters(), lr=parser_params.lr)
self.criterion = Criterion.Loss()
self.optimizer_CA = Adam(self.CA.parameters(), lr=parser_params.lr)
def __main__():
args = get_args_parser()
dist.init_process_group(backend='nccl')
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
set_random_seed(args.random_seed + dist.get_rank())
torch.cuda.set_device(torch.device('cuda:{}'.format(dist.get_rank())))
dist_logger = DistributedLogger(args.name, args.output_base_path,
args.master_rank, args.use_tensorboard)
train_dataset = TrainDataset(args.dataset_root, args.dataset_year,
(args.input_size_h, args.input_size_w),
args.pooler_size)
train_sampler = data.distributed.DistributedSampler(train_dataset)
train_dataloader = data.DataLoader(train_dataset,
batch_size=args.batch_size,
num_workers=args.num_workers,
sampler=train_sampler,
pin_memory=True,
drop_last=True)
val_dataset = ValDataset(args.dataset_root, args.dataset_year,
(args.input_size_h, args.input_size_w))
val_sampler = data.distributed.DistributedSampler(val_dataset)
val_dataloader = data.DataLoader(val_dataset,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=True,
sampler=val_sampler)
model = BlendMask(len(COCO_CLASSES), args.fpn_channels,
args.bases_module_channels, args.num_bases,
args.atten_size, args.pooler_size).cuda()
# model.load_state_dict(torch.load(f'./output/{args.name}/model/param.pth'))
model = parallel.DistributedDataParallel(model,
device_ids=[dist.get_rank()],
find_unused_parameters=True)
criterion = Criterion(args.focal_alpha, args.focal_gamma)
optim_parameters = [{
'params': [
p for n, p in model.module.named_parameters()
if not n.endswith('bias') and p.requires_grad
]
}, {
'params': [
p for n, p in model.module.named_parameters()
if n.endswith('bias') and p.requires_grad
],
'lr':
args.lr * args.bias_lr_mul,
'weight_decay':
args.weight_decay * args.bias_weight_decay_mul
}]
optimizer = optim.SGD(optim_parameters,
lr=args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
lr_lambda = utils.lr_lambda.get_warm_up_multi_step_lr_lambda(
len(train_dataloader), args.warm_up_epoch, args.warm_up_ratio,
args.milestones, args.step_gamma)
lr_scheduler = optim.lr_scheduler.LambdaLR(optimizer, lr_lambda)
nms_cfg = {
'nms_pre': args.nms_pre,
'cls_score_thr': args.nms_cls_score_thr,
'iou_thr': args.nms_iou_thr
}
for epoch_idx in range(args.epochs):
train_sampler.set_epoch(epoch_idx)
val_sampler.set_epoch(epoch_idx)
engine.train_one_epoch(model, criterion, optimizer, lr_scheduler,
train_dataloader, epoch_idx, dist_logger)
def index1(self, id=None, category=None, osep='None', print_area='false'):
if identity.current.user.basketmode:
raise redirect(url("index2",id=id,category=category,osep=osep,print_area=print_area))
osep = True
if print_area == 'true':
print_area = True
else:
print_area = False
id = util.session('current_assessment', 0, id)
category = util.session('current_category', 'Fine Motor', category)
osep = util.session('osep', False, osep)
child = util.session('current_child', 0)
if id == 0: # no id given
raise redirect("/")
try:
current_assessment = Assessment.get(id)
except SQLObjectNotFound: # Child with given id does not exist
raise redirect("/")
current_child = current_assessment.child
all_assessments = current_child.assessments
sorted_assessments = Assessment.select(AND(Assessment.q.childID==current_assessment.childID,
Assessment.q.id!=id,
Assessment.q.level==current_assessment.level), orderBy='-dateEntered')
old = list(sorted_assessments[:7])
if osep:
criteria = Criterion.selectBy(category=category, level=current_assessment.level, osep=True)
else:
criteria = Criterion.selectBy(category=category, level=current_assessment.level)
criteria = criteria.orderBy('sort_order')
crit_ids = [c.id for c in criteria]
sums = {}
for t in old:
sum = 0;
for s in t.scores:
if s.criterionID in crit_ids and s.criterion.rank != "1":
try:
sum += int(s.value)
except:
pass
sums[t.id] = sum
sum = 0;
for s in current_assessment.scores:
if s.criterionID in crit_ids and s.criterion.rank != "1":
try:
sum += int(s.value)
except:
pass
sums[current_assessment.id] = sum
util.session('fill', 'missing')
if len(old) == 0:
temp = util.session('values', 'zeros')
if temp == 'previous':
cherrypy.session['values'] = 'zeros'
else:
util.session('values', 'previous')
subtotal = Criterion.selectBy(category=category, level=current_assessment.level, prefix='Z')[0]
_na = Criterion.selectBy(category=category, level=current_assessment.level, prefix='Y')[0]
oldtotals = {}
oldna = {}
percents = {}
for t in old:
stscore = Score.selectBy(assessmentID=t.id, criterionID=subtotal.id)
if stscore.count() > 0:
stscore = stscore[0]
else:
stscore = Score(assessmentID=t.id, criterionID=subtotal.id, value="0", type='subtotal')
oldtotals[t.id] = stscore.value
nascore = Score.selectBy(assessmentID=t.id, criterionID=_na.id)
if nascore.count() > 0:
nascore = nascore[0]
else:
nascore = Score(assessmentID=t.id, criterionID=_na.id, value="0", type='na')
oldna[t.id] = nascore.value
percents[t.id] = util.percent(stscore.value, util.scorePossible(category, current_assessment.level), nascore.value)
stscore = Score.selectBy(assessmentID=id, criterionID=subtotal.id)
nascore = Score.selectBy(assessmentID=id, criterionID=_na.id)
if stscore.count() > 0:
stscore = stscore[0]
else:
stscore = Score(assessmentID=id, criterionID=subtotal.id, value="0", type='subtotal')
if nascore.count() > 0:
nascore = nascore[0]
else:
nascore = Score(assessmentID=id, criterionID=_na.id, value="0", type='na')
percent = util.percent(stscore.value, util.scorePossible(category, current_assessment.level), nascore.value)
return dict(child=current_child, id=id, a=current_assessment, old=old, oldtotals=oldtotals, \
stc=subtotal, subtotal=stscore, criteria=criteria, osep=osep, sums=sums, \
catlist=self.category_list, category=category, nextCategory=self.nextCategory, \
print_area=print_area, getEditor=self.getEditor, pcheck=self.persistCheck, \
shaded=util.shaded, scoreclass=util.color, scorePossible=util.scorePossible, \
nac=_na, na=nascore, oldna=oldna, percents=percents, percent=percent)
def index2(self, id=None, category=None, osep='None', print_area='false'):
if not identity.current.user.basketmode:
raise redirect(url("index1",id=id,category=category,osep=osep,print_area=print_area))
osep = True
if print_area == 'true':
print_area = True
else:
print_area = False
id = util.session('current_assessment', 0, id)
child = util.session('current_child', 0)
if id == 0: # no id given
raise redirect("/")
try:
current_assessment = Assessment.get(id)
except SQLObjectNotFound: # Child with given id does not exist
raise redirect("/")
current_child = current_assessment.child
all_assessments = current_child.assessments
sorted_assessments = Assessment.select(AND(Assessment.q.childID==current_assessment.childID,
Assessment.q.id!=id,
Assessment.q.level==current_assessment.level), orderBy='-dateEntered')
old = list(sorted_assessments[:7])
categories = self.category_list
data = {}
for c in categories:
cdata = {}
subtotal = Criterion.selectBy(category=c, level=current_assessment.level, prefix='Z')[0]
_na = Criterion.selectBy(category=c, level=current_assessment.level, prefix='Y')[0]
cdata['subtotal'] = subtotal
cdata['na'] = _na
oldtotals = {}
oldna = {}
percents = {}
for t in old:
stscore = Score.selectBy(assessmentID=t.id, criterionID=subtotal.id)
if stscore.count() > 0:
stscore = stscore[0]
else:
stscore = Score(assessmentID=t.id, criterionID=subtotal.id, value="0", type='subtotal')
oldtotals[t.id] = stscore.value
nascore = Score.selectBy(assessmentID=t.id, criterionID=_na.id)
if nascore.count() > 0:
nascore = nascore[0]
else:
nascore = Score(assessmentID=t.id, criterionID=_na.id, value="0", type='na')
oldna[t.id] = nascore.value
percents[t.id] = util.percent(stscore.value, util.scorePossible(c, current_assessment.level), nascore.value)
cdata['oldtotals'] = oldtotals
cdata['oldna'] = oldna
cdata['percents'] = percents
stscore = Score.selectBy(assessmentID=id, criterionID=subtotal.id)
nascore = Score.selectBy(assessmentID=id, criterionID=_na.id)
if stscore.count() > 0:
stscore = stscore[0]
else:
stscore = Score(assessmentID=id, criterionID=subtotal.id, value="0", type='subtotal')
if nascore.count() > 0:
nascore = nascore[0]
else:
nascore = Score(assessmentID=id, criterionID=_na.id, value="0", type='na')
percent = util.percent(stscore.value, util.scorePossible(c, current_assessment.level), nascore.value)
cdata['stscore'] = stscore
cdata['nascore'] = nascore
cdata['percent'] = percent
data[c] = cdata
baskets = []
for i in [1,2,3]:
bdata = {}
prefix = "B" + str(i)
b = Criterion.selectBy(level=current_assessment.level, prefix=prefix)[0]
oldb = {}
for t in old:
bscore = Score.selectBy(assessmentID=t.id, criterionID=b.id)
if bscore.count() > 0:
bscore = bscore[0]
else:
bscore = Score(assessmentID=t.id, criterionID=b.id, value="0", type='basket')
oldb[t.id] = bscore
bdata['old'] = oldb
bscore = Score.selectBy(assessmentID=id, criterionID=b.id)
if bscore.count() > 0:
bscore = bscore[0]
else:
bscore = Score(assessmentID=id, criterionID=b.id, value="0", type='basket')
bdata['val'] = bscore
baskets.append(bdata)
return dict(child=current_child, id=id, a=current_assessment, old=old, categories=categories, \
print_area=print_area, getEditor=self.getEditor, pcheck=self.persistCheck, \
shaded=util.shaded, scoreclass=util.color, scorePossible=util.scorePossible, \
data=data, baskets=baskets)
