Skip to content

Fix broken APIs and hardcoded wandb entity in Classification Transformers#231

Open
ArneshBanerjee wants to merge 1 commit into
ML4SCI:mainfrom
ArneshBanerjee:fix/classification-transformers-bugs
Open

Fix broken APIs and hardcoded wandb entity in Classification Transformers#231
ArneshBanerjee wants to merge 1 commit into
ML4SCI:mainfrom
ArneshBanerjee:fix/classification-transformers-bugs

Conversation

@ArneshBanerjee

Copy link
Copy Markdown

Fixes #191, #192

Several things in DeepLense_Classification_Transformers_Archil_Srivastava that break with current library versions: torchmetrics 0.11+ needs task="multiclass" on auroc/accuracy calls, micro_auroc was declared but never appended to so it silently returned NaN, softmax on eval.py:172 was missing dim=-1, verbose=True was removed from CosineAnnealingWarmRestarts in newer PyTorch, and torch.has_mps is deprecated. Also removed the hardcoded entity="_archil" from both train and eval so W&B runs work for other contributors.

…mers

torchmetrics 0.11+ requires task= for auroc and accuracy, and micro_auroc
was never appended to so it always returned NaN. softmax on eval.py:172
was missing dim=-1. CosineAnnealingWarmRestarts no longer takes verbose=
in newer PyTorch. torch.has_mps is deprecated, replaced with
torch.backends.mps.is_built(). Hardcoded entity="_archil" in train.py
and eval.py broke W&B for anyone else, so removed it.

Fixes ML4SCI#191, ML4SCI#192
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug] DeepLense_Classification_Transformers: Missing dependencies, broken APIs, and deprecated calls prevent running

1 participant