fix bug with propbank, fix dep-based

Riccorl · Nov 18, 2020 · 9725154 · 9725154
1 parent ccce441
commit 9725154
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 7 deletions.
diff --git a/training_config/bert_ml_conll2009.jsonnet b/training_config/bert_ml_conll2009.jsonnet
@@ -19,6 +19,7 @@
         "type": "transformer_srl_span",
         "embedding_dropout": 0.1,
         "bert_model": "bert-base-multilingual-cased",
+        "inventory": "propbank",
     },
 
     "trainer": {

diff --git a/training_config/bert_tiny_dep.jsonnet b/training_config/bert_tiny_dep.jsonnet
@@ -1,6 +1,6 @@
 {
     "dataset_reader": {
-      "type": "transformer_srl_span",
+      "type": "transformer_srl_dependency",
       "model_name": "mrm8488/bert-tiny-finetuned-squadv2",
     },
 
@@ -17,7 +17,7 @@
     "model": {
         "type": "transformer_srl_dependency",
         "embedding_dropout": 0.1,
-        "bert_model": "mrm8488/bert-tiny-finetuned-squadv2",
+        "model_name": "mrm8488/bert-tiny-finetuned-squadv2",
     },
 
     "trainer": {

diff --git a/transformer_srl/dataset_readers.py b/transformer_srl/dataset_readers.py
@@ -442,7 +442,7 @@ def _read(self, file_path: str):
 
             for annotation in parse_incr(
                 conll_file,
-                fields=conllu_fields if self.format == "conllu" else conll2009_fields,
+                fields=conll2009_fields if self.format == "conll2009" else conllu_fields,
                 field_parsers={"roles": lambda line, i: line[i:]},
             ):
                 # CoNLLU annotations sometimes add back in words that have been elided
@@ -460,7 +460,7 @@ def _read(self, file_path: str):
                     continue
                 frames = [x["frame"] for x in annotation]
                 # clean C-V tags (not needed yet)
-                frames = [f if f != "C-V" else "_" for f in frames]
+                frames = [f if f.isupper() else "_" for f in frames]
                 roles = [x["roles"] for x in annotation]
                 # transpose rolses, to have a list of roles per frame
                 roles = list(map(list, zip(*roles)))

diff --git a/transformer_srl/models.py b/transformer_srl/models.py
@@ -59,6 +59,7 @@ def __init__(
         srl_eval_path: str = DEFAULT_SRL_EVAL_PATH,
         restrict_frames: bool = False,
         restrict_roles: bool = False,
+        inventory: str = "verbatlas",
         **kwargs,
     ) -> None:
         # bypass SrlBert constructor
@@ -69,9 +70,10 @@ def __init__(
         self.restrict_roles = restrict_roles
         self.transformer = AutoModel.from_pretrained(bert_model)
         self.frame_criterion = nn.CrossEntropyLoss()
-        # add missing labels
-        frame_list = load_label_list(FRAME_LIST_PATH)
-        self.vocab.add_tokens_to_namespace(frame_list, "frames_labels")
+        if inventory == "verbatlas":
+            # add missing labels
+            frame_list = load_label_list(FRAME_LIST_PATH)
+            self.vocab.add_tokens_to_namespace(frame_list, "frames_labels")
         self.num_classes = self.vocab.get_vocab_size("labels")
         self.frame_num_classes = self.vocab.get_vocab_size("frames_labels")
         if srl_eval_path is not None: