Display 5 worst predictions at the end of evaluation

Based on !339 (merged)

Closes #245 (closed)

dan/ocr/evaluate.py

@@ -94,7 +94,7 @@ def print_worst_predictions(all_inferences: Dict[str, List[Inference]]):
             inference.prediction,
         )
         alignment_str = f'{alignment["query_aligned"]}\n{alignment["matched_aligned"]}\n{alignment["target_aligned"]}'
-        table.add_row([inference.image, inference.wer, alignment_str])
+        table.add_row([inference.image, round(inference.wer * 100, 2), alignment_str])
 
     print(f"\n#### {NB_WORST_PREDICTIONS} worst prediction(s)\n")
     print(table)
@@ -126,7 +126,14 @@ def eval_nerval(
         if not (ground_truths and predictions):
             continue
 
-        scores = evaluate(ground_truths, predictions, threshold)
+        scores = {
+            key: {
+                k: round(value * 100, 2) if k in ["P", "R", "F1"] else value
+                for k, value in values.items()
+            }
+            for key, values in evaluate(ground_truths, predictions, threshold).items()
+        }
+
         print(f"\n##### {split_name}\n")
         print_results(scores)