def __init__(self, max_iter):
self.max_iter = max_iter
self.iter_timer = Timer()
# Current minibatch errors (smoothed over a window)
self.mb_top1_err = ScalarMeter(cfg.LOG_PERIOD)
# Min errors (over the full Val set)
self.min_top1_err = 100.0
# Number of misclassified examples
self.num_top1_mis = 0
self.num_samples = 0
def __init__(self, epoch_iters):
self.epoch_iters = epoch_iters
self.max_iter = cfg.OPTIM.MAX_EPOCH * epoch_iters
self.iter_timer = Timer()
self.loss = ScalarMeter(cfg.LOG_PERIOD)
self.loss_total = 0.0
self.lr = None
# Current minibatch errors (smoothed over a window)
self.mb_top1_err = ScalarMeter(cfg.LOG_PERIOD)
# Number of misclassified examples
self.num_top1_mis = 0
self.num_samples = 0
def compute_fw_test_time(model, inputs):
"""Computes forward test time (no grad, eval mode)."""
# Use eval mode
model.eval()
# Warm up the caches
for _cur_iter in range(cfg.PREC_TIME.WARMUP_ITER):
model(inputs)
# Make sure warmup kernels completed
torch.cuda.synchronize()
# Compute precise forward pass time
timer = Timer()
for _cur_iter in range(cfg.PREC_TIME.NUM_ITER):
timer.tic()
model(inputs)
torch.cuda.synchronize()
timer.toc()
# Make sure forward kernels completed
torch.cuda.synchronize()
return timer.average_time
def compute_fw_bw_time(model, loss_fun, inputs, labels):
"""Computes forward backward time."""
# Use train mode
model.train()
# Warm up the caches
for _cur_iter in range(cfg.PREC_TIME.WARMUP_ITER):
preds = model(inputs)
loss = loss_fun(preds, labels)
loss.backward()
# Make sure warmup kernels completed
torch.cuda.synchronize()
# Compute precise forward backward pass time
fw_timer = Timer()
bw_timer = Timer()
for _cur_iter in range(cfg.PREC_TIME.NUM_ITER):
# Forward
fw_timer.tic()
preds = model(inputs)
loss = loss_fun(preds, labels)
torch.cuda.synchronize()
fw_timer.toc()
# Backward
bw_timer.tic()
loss.backward()
torch.cuda.synchronize()
bw_timer.toc()
# Make sure forward backward kernels completed
torch.cuda.synchronize()
return fw_timer.average_time, bw_timer.average_time
class TrainMeter(object):
"""Measures training stats."""
def __init__(self, epoch_iters):
self.epoch_iters = epoch_iters
self.max_iter = cfg.OPTIM.MAX_EPOCH * epoch_iters
self.iter_timer = Timer()
self.loss = ScalarMeter(cfg.LOG_PERIOD)
self.loss_total = 0.0
self.lr = None
# Current minibatch errors (smoothed over a window)
self.mb_top1_err = ScalarMeter(cfg.LOG_PERIOD)
self.mb_top5_err = ScalarMeter(cfg.LOG_PERIOD)
# Number of misclassified examples
self.num_top1_mis = 0
self.num_top5_mis = 0
self.num_samples = 0
def reset(self, timer=False):
if timer:
self.iter_timer.reset()
self.loss.reset()
self.loss_total = 0.0
self.lr = None
self.mb_top1_err.reset()
self.mb_top5_err.reset()
self.num_top1_mis = 0
self.num_top5_mis = 0
self.num_samples = 0
def iter_tic(self):
self.iter_timer.tic()
def iter_toc(self):
self.iter_timer.toc()
def update_stats(self, top1_err, top5_err, loss, lr, mb_size):
# Current minibatch stats
self.mb_top1_err.add_value(top1_err)
self.mb_top5_err.add_value(top5_err)
self.loss.add_value(loss)
self.lr = lr
# Aggregate stats
self.num_top1_mis += top1_err * mb_size
self.num_top5_mis += top5_err * mb_size
self.loss_total += loss * mb_size
self.num_samples += mb_size
def get_iter_stats(self, cur_epoch, cur_iter):
eta_sec = self.iter_timer.average_time * (
self.max_iter - (cur_epoch * self.epoch_iters + cur_iter + 1))
eta_td = datetime.timedelta(seconds=int(eta_sec))
mem_usage = metrics.gpu_mem_usage()
stats = {
'_type': 'train_iter',
'epoch': '{}/{}'.format(cur_epoch + 1, cfg.OPTIM.MAX_EPOCH),
'iter': '{}/{}'.format(cur_iter + 1, self.epoch_iters),
'time_avg': self.iter_timer.average_time,
'time_diff': self.iter_timer.diff,
'eta': eta_str(eta_td),
'top1_err': self.mb_top1_err.get_win_median(),
'top5_err': self.mb_top5_err.get_win_median(),
'loss': self.loss.get_win_median(),
'lr': self.lr,
'mem': int(np.ceil(mem_usage))
}
return stats
def log_iter_stats(self, cur_epoch, cur_iter):
if (cur_iter + 1) % cfg.LOG_PERIOD != 0:
return
stats = self.get_iter_stats(cur_epoch, cur_iter)
lu.log_json_stats(stats)
def get_epoch_stats(self, cur_epoch):
eta_sec = self.iter_timer.average_time * (
self.max_iter - (cur_epoch + 1) * self.epoch_iters)
eta_td = datetime.timedelta(seconds=int(eta_sec))
mem_usage = metrics.gpu_mem_usage()
top1_err = self.num_top1_mis / self.num_samples
top5_err = self.num_top5_mis / self.num_samples
avg_loss = self.loss_total / self.num_samples
stats = {
'_type': 'train_epoch',
'epoch': '{}/{}'.format(cur_epoch + 1, cfg.OPTIM.MAX_EPOCH),
'time_avg': self.iter_timer.average_time,
'eta': eta_str(eta_td),
'top1_err': top1_err,
'top5_err': top5_err,
'loss': avg_loss,
'lr': self.lr,
'mem': int(np.ceil(mem_usage))
}
return stats
def log_epoch_stats(self, cur_epoch):
stats = self.get_epoch_stats(cur_epoch)
lu.log_json_stats(stats)
class TestMeter(object):
"""Measures testing stats."""
def __init__(self, max_iter):
self.max_iter = max_iter
self.iter_timer = Timer()
# Current minibatch errors (smoothed over a window)
self.mb_top1_err = ScalarMeter(cfg.LOG_PERIOD)
self.mb_top5_err = ScalarMeter(cfg.LOG_PERIOD)
# Min errors (over the full test set)
self.min_top1_err = 100.0
self.min_top5_err = 100.0
# Number of misclassified examples
self.num_top1_mis = 0
self.num_top5_mis = 0
self.num_samples = 0
def reset(self, min_errs=False):
if min_errs:
self.min_top1_err = 100.0
self.min_top5_err = 100.0
self.iter_timer.reset()
self.mb_top1_err.reset()
self.mb_top5_err.reset()
self.num_top1_mis = 0
self.num_top5_mis = 0
self.num_samples = 0
def iter_tic(self):
self.iter_timer.tic()
def iter_toc(self):
self.iter_timer.toc()
def update_stats(self, top1_err, top5_err, mb_size):
self.mb_top1_err.add_value(top1_err)
self.mb_top5_err.add_value(top5_err)
self.num_top1_mis += top1_err * mb_size
self.num_top5_mis += top5_err * mb_size
self.num_samples += mb_size
def get_iter_stats(self, cur_epoch, cur_iter):
mem_usage = metrics.gpu_mem_usage()
iter_stats = {
'_type': 'test_iter',
'epoch': '{}/{}'.format(cur_epoch + 1, cfg.OPTIM.MAX_EPOCH),
'iter': '{}/{}'.format(cur_iter + 1, self.max_iter),
'time_avg': self.iter_timer.average_time,
'time_diff': self.iter_timer.diff,
'top1_err': self.mb_top1_err.get_win_median(),
'top5_err': self.mb_top5_err.get_win_median(),
'mem': int(np.ceil(mem_usage))
}
return iter_stats
def log_iter_stats(self, cur_epoch, cur_iter):
if (cur_iter + 1) % cfg.LOG_PERIOD != 0:
return
stats = self.get_iter_stats(cur_epoch, cur_iter)
lu.log_json_stats(stats)
def get_epoch_stats(self, cur_epoch):
top1_err = self.num_top1_mis / self.num_samples
top5_err = self.num_top5_mis / self.num_samples
self.min_top1_err = min(self.min_top1_err, top1_err)
self.min_top5_err = min(self.min_top5_err, top5_err)
mem_usage = metrics.gpu_mem_usage()
stats = {
'_type': 'test_epoch',
'epoch': '{}/{}'.format(cur_epoch + 1, cfg.OPTIM.MAX_EPOCH),
'time_avg': self.iter_timer.average_time,
'top1_err': top1_err,
'top5_err': top5_err,
'min_top1_err': self.min_top1_err,
'min_top5_err': self.min_top5_err,
'mem': int(np.ceil(mem_usage))
}
return stats
def log_epoch_stats(self, cur_epoch):
stats = self.get_epoch_stats(cur_epoch)
lu.log_json_stats(stats)
