Skip to content

feat(train): add validation metrics and distributed support #2574

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+48 −1
Open

feat(train): add validation metrics and distributed support #2574

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
49 changes: 48 additions & 1 deletion train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,12 @@
except ImportError as e:
has_functorch = False

try:
from sklearn.metrics import precision_score, recall_score, f1_score
has_sklearn = True
except ImportError:
has_sklearn = False

has_compile = hasattr(torch, 'compile')


Expand Down Expand Up @@ -400,6 +406,10 @@
help='wandb tags')
group.add_argument('--wandb-resume-id', default='', type=str, metavar='ID',
help='If resuming a run, the id of the run in wandb')
group.add_argument('--metrics-avg', type=str, default=None,
choices=['micro', 'macro', 'weighted'],
help='Enable precision, recall, F1-score calculation and specify the averaging method. '
'Requires scikit-learn. (default: None)')

# NaFlex scheduled loader arguments
group.add_argument('--naflex-loader', action='store_true', default=False,
Expand Down Expand Up @@ -1318,6 +1328,15 @@ def validate(
top1_m = utils.AverageMeter()
top5_m = utils.AverageMeter()

if args.metrics_avg:
if not has_sklearn:
_logger.warning(
"scikit-learn not installed, disabling extra metrics. Please install with 'pip install scikit-learn'.")
args.metrics_avg = None
else:
all_preds = []
all_targets = []

model.eval()

end = time.time()
Expand Down Expand Up @@ -1345,6 +1364,10 @@ def validate(
loss = loss_fn(output, target)
acc1, acc5 = utils.accuracy(output, target, topk=(1, 5))

if args.metrics_avg:
all_preds.append(torch.argmax(output, dim=1).cpu())
all_targets.append(target.cpu())

if args.distributed:
reduced_loss = utils.reduce_tensor(loss.data, args.world_size)
acc1 = utils.reduce_tensor(acc1, args.world_size)
Expand Down Expand Up @@ -1374,7 +1397,31 @@ def validate(
f'Acc@5: {top5_m.val:>7.3f} ({top5_m.avg:>7.3f})'
)

metrics = OrderedDict([('loss', losses_m.avg), ('top1', top1_m.avg), ('top5', top5_m.avg)])
metrics = OrderedDict([('loss', losses_m.avg), ('top1', top1_m.avg), ('top5', top5_m.avg)])

if args.metrics_avg:
all_preds = torch.cat(all_preds)
all_targets = torch.cat(all_targets)

if args.distributed:
# Gather list of tensors from all processes
pred_list = [torch.zeros_like(all_preds) for _ in range(args.world_size)]
target_list = [torch.zeros_like(all_targets) for _ in range(args.world_size)]
torch.distributed.all_gather(pred_list, all_preds)
torch.distributed.all_gather(target_list, all_targets)

if utils.is_primary(args):
all_preds = torch.cat(pred_list)
all_targets = torch.cat(target_list)

if utils.is_primary(args):
precision = precision_score(all_targets.numpy(), all_preds.numpy(), average=args.metrics_avg, zero_division=0)
recall = recall_score(all_targets.numpy(), all_preds.numpy(), average=args.metrics_avg, zero_division=0)
f1 = f1_score(all_targets.numpy(), all_preds.numpy(), average=args.metrics_avg, zero_division=0)

metrics[f'{args.metrics_avg}_precision'] = round(precision, 4)
metrics[f'{args.metrics_avg}_recall'] = round(recall, 4)
metrics[f'{args.metrics_avg}_f1_score'] = round(f1, 4)

return metrics

Expand Down