本文整理汇总了Python中tensorboard_logger.log_value方法的典型用法代码示例。如果您正苦于以下问题:Python tensorboard_logger.log_value方法的具体用法?Python tensorboard_logger.log_value怎么用?Python tensorboard_logger.log_value使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类tensorboard_logger
的用法示例。
在下文中一共展示了tensorboard_logger.log_value方法的15个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: validate_caption_only
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def validate_caption_only(opt, val_loader, model):
# compute the encoding for all the validation images and captions
img_embs, cap_embs = encode_data(
model, val_loader, opt.log_step, logging.info)
# caption retrieval
(r1, r5, r10, medr, meanr) = i2t_text_only(img_embs, cap_embs, measure=opt.measure)
logging.info("Image to text: %.1f, %.1f, %.1f, %.1f, %.1f" %
(r1, r5, r10, medr, meanr))
# sum of recalls to be used for early stopping
currscore = r1 + r5 + r10
# record metrics in tensorboard
tb_logger.log_value('r1', r1, step=model.Eiters)
tb_logger.log_value('r5', r5, step=model.Eiters)
tb_logger.log_value('r10', r10, step=model.Eiters)
tb_logger.log_value('medr', medr, step=model.Eiters)
tb_logger.log_value('meanr', meanr, step=model.Eiters)
tb_logger.log_value('rsum', currscore, step=model.Eiters)
return currscore
示例2: __call__
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def __call__(self, job):
import tensorboard_logger as tl
# id = job.id
budget = job.kwargs['budget']
# config = job.kwargs['config']
timestamps = job.timestamps
result = job.result
exception = job.exception
time_step = int(timestamps['finished'] - self.start_time)
if result is not None:
tl.log_value('BOHB/all_results', result['loss'] * -1, time_step)
if result['loss'] < self.incumbent:
self.incumbent = result['loss']
tl.log_value('BOHB/incumbent_results', self.incumbent * -1, time_step)
示例3: optimize_pipeline
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def optimize_pipeline(self, config, config_id, budget, optimize_start_time):
"""Fit the pipeline using the sampled hyperparameter configuration.
Arguments:
config {dict} -- The sampled hyperparameter configuration.
config_id {tuple} -- An ID for the configuration. Assigned by BOHB.
budget {float} -- The budget to evaluate the hyperparameter configuration.
optimize_start_time {float} -- The time when optimization started.
Returns:
dict -- The result of fitting the pipeline.
"""
try:
self.autonet_logger.info("Fit optimization pipeline")
return self.pipeline.fit_pipeline(hyperparameter_config=config, pipeline_config=self.pipeline_config,
X_train=self.X_train, Y_train=self.Y_train, X_valid=self.X_valid, Y_valid=self.Y_valid,
budget=budget, budget_type=self.budget_type, max_budget=self.max_budget, optimize_start_time=optimize_start_time,
refit=False, rescore=False, hyperparameter_config_id=config_id, dataset_info=self.dataset_info)
except Exception as e:
if 'use_tensorboard_logger' in self.pipeline_config and self.pipeline_config['use_tensorboard_logger']:
import tensorboard_logger as tl
tl.log_value('Exceptions/' + str(e), budget, int(time.time()))
self.autonet_logger.info(str(e))
raise e
示例4: optimize_pipeline
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def optimize_pipeline(self, config, budget, config_id, random_state):
random.setstate(random_state)
if self.permutations is not None:
current_sh_run = config_id[0]
self.pipeline_config["dataset_order"] = self.permutations[current_sh_run%len(self.permutations)].tolist()
try:
self.autonet_logger.info("Fit optimization pipeline")
return self.pipeline.fit_pipeline(hyperparameter_config=config, pipeline_config=self.pipeline_config,
X_train=self.X_train, Y_train=self.Y_train, X_valid=self.X_valid, Y_valid=self.Y_valid,
budget=budget, budget_type=self.budget_type, max_budget=self.max_budget,
config_id=config_id, working_directory=self.working_directory), random.getstate()
except Exception as e:
if 'use_tensorboard_logger' in self.pipeline_config and self.pipeline_config['use_tensorboard_logger']:
import tensorboard_logger as tl
tl.log_value('Exceptions/' + str(e), budget, int(time.time()))
#self.autonet_logger.exception('Exception occurred')
raise e
示例5: log_value
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def log_value(self, name, value):
log_value(name, value, self.global_step)
return self
示例6: validate
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def validate(opt, val_loader, model):
# compute the encoding for all the validation images and captions
img_embs, cap_embs = encode_data(
model, val_loader, opt.log_step, logging.info)
# caption retrieval
(r1, r5, r10, medr, meanr) = i2t(img_embs, cap_embs, measure=opt.measure)
logging.info("Image to text: %.1f, %.1f, %.1f, %.1f, %.1f" %
(r1, r5, r10, medr, meanr))
# image retrieval
(r1i, r5i, r10i, medri, meanr) = t2i(
img_embs, cap_embs, measure=opt.measure)
logging.info("Text to image: %.1f, %.1f, %.1f, %.1f, %.1f" %
(r1i, r5i, r10i, medri, meanr))
# sum of recalls to be used for early stopping
currscore = r1 + r5 + r10 + r1i + r5i + r10i
# record metrics in tensorboard
tb_logger.log_value('r1', r1, step=model.Eiters)
tb_logger.log_value('r5', r5, step=model.Eiters)
tb_logger.log_value('r10', r10, step=model.Eiters)
tb_logger.log_value('medr', medr, step=model.Eiters)
tb_logger.log_value('meanr', meanr, step=model.Eiters)
tb_logger.log_value('r1i', r1i, step=model.Eiters)
tb_logger.log_value('r5i', r5i, step=model.Eiters)
tb_logger.log_value('r10i', r10i, step=model.Eiters)
tb_logger.log_value('medri', medri, step=model.Eiters)
tb_logger.log_value('meanr', meanr, step=model.Eiters)
tb_logger.log_value('rsum', currscore, step=model.Eiters)
return currscore
示例7: train
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def train(opt, train_loader, model, epoch):
# average meters to record the training statistics
batch_time = AverageMeter()
data_time = AverageMeter()
train_logger = LogCollector()
# switch to train mode
model.train_start()
progbar = Progbar(len(train_loader.dataset))
end = time.time()
for i, train_data in enumerate(train_loader):
# measure data loading time
data_time.update(time.time() - end)
# make sure train logger is used
model.logger = train_logger
# Update the model
b_size, loss = model.train_emb(*train_data)
progbar.add(b_size, values=[('loss', loss)])
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
# Record logs in tensorboard
tb_logger.log_value('epoch', epoch, step=model.Eiters)
tb_logger.log_value('step', i, step=model.Eiters)
tb_logger.log_value('batch_time', batch_time.val, step=model.Eiters)
tb_logger.log_value('data_time', data_time.val, step=model.Eiters)
model.logger.tb_log(tb_logger, step=model.Eiters)
示例8: log_value
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def log_value(harn, key, value, n_iter):
if False:
print('{}={} @ {}'.format(key, value, n_iter))
if tensorboard_logger:
tensorboard_logger.log_value(key, value, n_iter)
示例9: train_epoch
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def train_epoch(harn):
ave_metrics = defaultdict(lambda: 0)
# change learning rate (modified optimizer inplace)
lr = harn.lr_scheduler(harn.epoch, harn.optimizer)
# train batch
for batch_idx, input_batch in enumerate(harn.train_loader):
input_batch = harn._to_xpu(*input_batch)
# print('Begin batch {}'.format(batch_idx))
t_cur_metrics = harn.train_batch(input_batch)
for k, v in t_cur_metrics.items():
ave_metrics[k] += v
# display training info
if (batch_idx + 1) % harn.config['displayInterval'] == 0:
for k in ave_metrics.keys():
ave_metrics[k] /= harn.config['displayInterval']
n_train = len(harn.train_loader)
harn.log('Epoch {0}: {1} / {2} | lr:{3} - tloss:{4:.5f} acc:{5:.2f} | sdis:{6:.3f} ddis:{7:.3f}'.format(
harn.epoch, batch_idx, n_train, lr,
ave_metrics['loss'], ave_metrics['accuracy'],
ave_metrics['pos_dist'], ave_metrics['neg_dist']))
iter_idx = harn.epoch * n_train + batch_idx
for key, value in ave_metrics.items():
harn.log_value('train ' + key, value, iter_idx)
# diagnoseGradients(model.parameters())
for k in ave_metrics.keys():
ave_metrics[k] = 0
示例10: validation_epoch
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def validation_epoch(harn):
ave_metrics = defaultdict(lambda: 0)
final_metrics = ave_metrics.copy()
for vali_idx, input_batch in enumerate(harn.vali_loader):
input_batch = harn._to_xpu(*input_batch)
# print('Begin batch {}'.format(vali_idx))
v_cur_metrics = harn.validation_batch(input_batch)
for k, v in v_cur_metrics.items():
ave_metrics[k] += v
final_metrics[k] += v
if (vali_idx + 1) % harn.config['vail_displayInterval'] == 0:
for k in ave_metrics.keys():
ave_metrics[k] /= harn.config['displayInterval']
harn.log('Epoch {0}: {1} / {2} | vloss:{3:.5f} acc:{4:.2f} | sdis:{5:.3f} ddis:{6:.3f}'.format(
harn.epoch, vali_idx, len(harn.vali_loader),
ave_metrics['loss'], ave_metrics['accuracy'],
ave_metrics['pos_dist'], ave_metrics['neg_dist']))
for k in ave_metrics.keys():
ave_metrics[k] = 0
for k in final_metrics.keys():
final_metrics[k] /= len(harn.vali_loader)
harn.log('Epoch {0}: final vloss:{1:.5f} acc:{2:.2f} | sdis:{3:.3f} ddis:{4:.3f}'.format(
harn.epoch, final_metrics['loss'], final_metrics['accuracy'],
final_metrics['pos_dist'], final_metrics['neg_dist']))
iter_idx = harn.epoch * len(harn.vali_loader) + vali_idx
for key, value in final_metrics.items():
harn.log_value('validation ' + key, value, iter_idx)
# def display_metrics():
# pass
示例11: train_one_epoch
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def train_one_epoch(model, train_loader, optimizer, epoch, lr_scheduler, total_it, tb_log, log_f):
model.train()
log_print('===============TRAIN EPOCH %d================' % epoch, log_f=log_f)
loss_func = DiceLoss(ignore_target=-1)
for it, batch in enumerate(train_loader):
optimizer.zero_grad()
pts_input, cls_labels = batch['pts_input'], batch['cls_labels']
pts_input = torch.from_numpy(pts_input).cuda(non_blocking=True).float()
cls_labels = torch.from_numpy(cls_labels).cuda(non_blocking=True).long().view(-1)
pred_cls = model(pts_input)
pred_cls = pred_cls.view(-1)
loss = loss_func(pred_cls, cls_labels)
loss.backward()
clip_grad_norm_(model.parameters(), 1.0)
optimizer.step()
total_it += 1
pred_class = (torch.sigmoid(pred_cls) > FG_THRESH)
fg_mask = cls_labels > 0
correct = ((pred_class.long() == cls_labels) & fg_mask).float().sum()
union = fg_mask.sum().float() + (pred_class > 0).sum().float() - correct
iou = correct / torch.clamp(union, min=1.0)
cur_lr = lr_scheduler.get_lr()[0]
tb_log.log_value('learning_rate', cur_lr, epoch)
if tb_log is not None:
tb_log.log_value('train_loss', loss, total_it)
tb_log.log_value('train_fg_iou', iou, total_it)
log_print('training epoch %d: it=%d/%d, total_it=%d, loss=%.5f, fg_iou=%.3f, lr=%f' %
(epoch, it, len(train_loader), total_it, loss.item(), iou.item(), cur_lr), log_f=log_f)
return total_it
示例12: eval_one_epoch
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def eval_one_epoch(model, eval_loader, epoch, tb_log=None, log_f=None):
model.train()
log_print('===============EVAL EPOCH %d================' % epoch, log_f=log_f)
iou_list = []
for it, batch in enumerate(eval_loader):
pts_input, cls_labels = batch['pts_input'], batch['cls_labels']
pts_input = torch.from_numpy(pts_input).cuda(non_blocking=True).float()
cls_labels = torch.from_numpy(cls_labels).cuda(non_blocking=True).long().view(-1)
pred_cls = model(pts_input)
pred_cls = pred_cls.view(-1)
pred_class = (torch.sigmoid(pred_cls) > FG_THRESH)
fg_mask = cls_labels > 0
correct = ((pred_class.long() == cls_labels) & fg_mask).float().sum()
union = fg_mask.sum().float() + (pred_class > 0).sum().float() - correct
iou = correct / torch.clamp(union, min=1.0)
iou_list.append(iou.item())
log_print('EVAL: it=%d/%d, iou=%.3f' % (it, len(eval_loader), iou), log_f=log_f)
iou_list = np.array(iou_list)
avg_iou = iou_list.mean()
if tb_log is not None:
tb_log.log_value('eval_fg_iou', avg_iou, epoch)
log_print('\nEpoch %d: Average IoU (samples=%d): %.6f' % (epoch, iou_list.__len__(), avg_iou), log_f=log_f)
return avg_iou
示例13: run_epoch
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def run_epoch(model, optimizer, train_ldr, it, avg_loss):
model_t = 0.0; data_t = 0.0
end_t = time.time()
tq = tqdm.tqdm(train_ldr)
for batch in tq:
start_t = time.time()
optimizer.zero_grad()
loss = model.loss(batch)
loss.backward()
grad_norm = nn.utils.clip_grad_norm(model.parameters(), 200)
loss = loss.data[0]
optimizer.step()
prev_end_t = end_t
end_t = time.time()
model_t += end_t - start_t
data_t += start_t - prev_end_t
exp_w = 0.99
avg_loss = exp_w * avg_loss + (1 - exp_w) * loss
tb.log_value('train_loss', loss, it)
tq.set_postfix(iter=it, loss=loss,
avg_loss=avg_loss, grad_norm=grad_norm,
model_time=model_t, data_time=data_t)
it += 1
return it, avg_loss
示例14: train
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def train(epoch):
rnet.train()
total = 0
correct = 0
train_loss = 0
total_batch = 0
for batch_idx, (inputs, targets) in tqdm.tqdm(enumerate(trainloader), total=len(trainloader)):
inputs, targets = inputs.to(device), targets.to(device)
probs = rnet(inputs, True)
optimizer.zero_grad()
loss = criterion(probs, targets)
loss.backward()
optimizer.step()
train_loss += loss.item()
_, predicted = probs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
total_batch += 1
print('E:%d Train Loss: %.3f Train Acc: %.3f LR %f'
% (epoch,
train_loss / total_batch,
correct / total,
optimizer.param_groups[0]['lr']))
tensorboard_logger.log_value('train_acc', correct/total, epoch)
tensorboard_logger.log_value('train_loss', train_loss / total_batch, epoch)
示例15: test
# 需要导入模块: import tensorboard_logger [as 别名]
# 或者: from tensorboard_logger import log_value [as 别名]
def test(epoch):
global best_test_acc
rnet.eval()
total = 0
correct = 0
test_loss = 0
total_batch = 0
for batch_idx, (inputs, targets) in tqdm.tqdm(enumerate(testloader), total=len(testloader)):
inputs, targets = inputs.to(device), targets.to(device)
probs = rnet(inputs)
loss = criterion(probs, targets)
test_loss += loss.item()
_, predicted = probs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
total_batch += 1
print('E:%d Test Loss: %.3f Test Acc: %.3f'
% (epoch, test_loss / total_batch, correct / total))
# save best model
acc = 100.*correct/total
if acc > best_test_acc:
best_test_acc = acc
print('saving best model...')
state = {
'net': rnet.state_dict(),
'acc': acc,
'epoch': epoch,
}
torch.save(state, 'resnet110.t7')
tensorboard_logger.log_value('test_acc', acc, epoch)
tensorboard_logger.log_value('test_loss', test_loss/total_batch, epoch)