send evaluation table to wandb when enabled

[lfoppiano]

This PR allow to send the results of the evaluation to wandb. When running train_eval, it will be attached to the run, but with the eval command, it will be attached to a separate run in wandb. Nothing we can do about that.

[Copilot]

Pull Request Overview

This PR enables logging of evaluation metrics to Weights & Biases by propagating a new report_to_wandb flag and formatting results into a wandb.Table.

• Adds a report_to_wandb parameter through the evaluation and training pipeline.
• Implements to_wandb_table utility to convert metric maps into table format.
• Updates the Grobid tagger CLI to accept and forward the report_to_wandb flag.

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
delft/sequenceLabelling/wrapper.py	Propagate report_to_wandb into evaluation and log tables to wandb
delft/sequenceLabelling/trainer.py	Rename wandb flag, implement report_to_wandb in Scorer, add helper
delft/applications/grobidTagger.py	Update eval_ signature and forwarding of report_to_wandb flag

Comments suppressed due to low confidence (3)

delft/sequenceLabelling/trainer.py:33

• The constructor parameter is now report_to_wandb but the instance attribute remains self.enable_wandb. Rename the attribute to self.report_to_wandb (or the parameter back to enable_wandb) for consistency.

report_to_wandb = False

delft/sequenceLabelling/trainer.py:497

• This new utility function isn’t covered by tests. Consider adding unit tests for to_wandb_table to verify correct table structure for various input maps.

def to_wandb_table(report_as_map):

delft/applications/grobidTagger.py:521

• Passing the wandb module here will lead to a NameError and is semantically incorrect—this parameter expects a boolean. Replace report_to_wandb=wandb with report_to_wandb=True (or the appropriate flag) and ensure you import wandb if needed.

report_to_wandb=wandb

[Copilot]

[nitpick] Using an empty string for the first column header is unclear. Consider renaming it to "label" or "class" to improve readability in the wandb table.

Suggested change

	columns = ["", "precision", "recall", "f1-score", "support"]
	columns = ["label", "precision", "recall", "f1-score", "support"]

[Copilot]

[nitpick] Reusing the variable micro for macro-averaged metrics can be confusing. Rename it to macro_metrics (or similar) when handling the 'macro' key.

Suggested change

	micro = report_as_map['macro']
	micro_row = [
	"all (macro avg.)",
	round(micro['precision'], 4),
	round(micro['recall'], 4),
	round(micro['f1'], 4),
	int(micro['support'])
	]
	data.append(micro_row)
	macro_metrics = report_as_map['macro']
	macro_row = [
	"all (macro avg.)",
	round(macro_metrics['precision'], 4),
	round(macro_metrics['recall'], 4),
	round(macro_metrics['f1'], 4),
	int(macro_metrics['support'])
	]
	data.append(macro_row)