Bug in Bipartite Method

[roquelopez]

I found a bug in Magneto when I tried to get the top1 match using the fine-tuned small language model as retriever with the bipartite algorithm as reranker. It happens when initial_matches is empty https://github.com/VIDA-NYU/data-integration-eval/blob/38a232855a2753204c6a6bc411657115e6d7a3b0/algorithms/magneto/magneto/bp_reranker.py#L73 and it becomes empty because of this: https://github.com/VIDA-NYU/data-integration-eval/blob/38a232855a2753204c6a6bc411657115e6d7a3b0/algorithms/magneto/magneto/bp_reranker.py#L67

File ../magneto/magneto/bp_reranker.py:73, in arrange_bipartite_matches(initial_matches, source_table, source_table_name, target_table, target_table_name)
     70 min_filtered_score = min(filtered_matches.values())
     72 # Calculate the scaling factor to keep scores in initial_matches just below min_filtered_score
---> 73 initial_max_score = max(initial_matches.values())
     74 scaling_factor = (
     75     (min_filtered_score - 0.01) / initial_max_score if initial_max_score > 0 else 1
     76 )
     78 # Adjust scores in initial_matches, maintaining relative differences

ValueError: max() arg is an empty sequence```