Fix transformed metric format

Author: Marcin Kardas

sota_extractor2/models/linking/bm25_naive.py

@@ -231,11 +231,12 @@ def linked_proposals(proposals):
             df = taxonomy_linking(prop.dataset, datasets, desc, topk=topk, debug_info=prop)
             for _, row in df.iterrows():
                 raw_value = prop.raw_value
-                parsed = float(extract_value(raw_value, format))
+                parsed = extract_value(raw_value, format)
                 metric = row['metric']
                 if metric != row['true_metric']:
                     metric = row['true_metric']
                     parsed = 1 - parsed if parsed < 1 else 100 - parsed
+                parsed = float(parsed)
 
                 linked = {
                     'dataset': row['dataset'],

sota_extractor2/models/linking/context_search.py

@@ -211,7 +211,7 @@ def __call__(self, query, datasets, caption, topk=1, debug_info=None):
         cellstr = debug_info.cell.cell_ext_id
         pipeline_logger("linking::taxonomy_linking::call", ext_id=cellstr, query=query, datasets=datasets, caption=caption)
         datasets = " ".join(datasets)
-        key = (datasets, caption, query)
+        key = (datasets, caption, query, topk)
         ###print(f"[DEBUG] {cellstr}")
         ###print("[DEBUG]", debug_info)
         ###print("query:", query, caption)
@@ -229,7 +229,7 @@ def __call__(self, query, datasets, caption, topk=1, debug_info=None):
             ###print("Taking result from cache")
             p = self.queries[key]
         else:
-            dist = self.match(key)
+            dist = self.match((datasets, caption, query))
             top_results = sorted(dist, key=lambda x: x[1], reverse=True)[:max(topk, 5)]
 
             entries = []

sota_extractor2/models/linking/manual_dicts.py

@@ -156,6 +156,9 @@
 
     # Semantic Textual Similarity
     'STS Benchmark': ['sts b'],
+
+    # Paraphrase Identification
+    'Microsoft Research Paraphrase Corpus': ['MRPC'],
 }
 
 tasks = {}

sota_extractor2/models/structure/ulmfit.py

@@ -6,6 +6,9 @@ def __init__(self, path, file, sp_path=None, sp_model="spm.model", sp_vocab="spm
         path = Path(path)
         sp_path = path if sp_path is None else Path(sp_path)
         self.learner = load_learner(path=path, file=file)
+        import sys, os
+        print(f"[PID {os.getpid()}] Load model {file}", file=sys.stderr)
+        sys.stderr.flush()
         self._fix_sp_processor(sp_path, sp_model, sp_vocab)
 
         # disable multiprocessing to avoid celery deamon issues