- GLiNER evaluation for all datasets except pile, with results (which are...

- GLiNER evaluation for all datasets except pile, with results (which are invalid for figer because it currently only returns one annotated entity for each sentence)
- fixed test_datasets inaccurate number of sentences error due to recent changes regarding limiting numbers of outputted instances

src/experiments/GLiNER_evalutaion/evaluation.py

@@ -4,6 +4,7 @@ Evaluates GLiNER as SotA  and plots results using reusable functions in plotter.
 
 from src.metrics import NER_metrics, read_NER_metrics
 
-# NER_metrics("GLiNER", "CoNLL", "results", test_instances=100)
-
-read_NER_metrics("results")
+for dataset in ["CoNLL", "FIGER-coarse", "FIGER-fine"]:
+    NER_metrics("GLiNER", dataset, f"results_{dataset}", test_instances=1000)
+    print(f"\nResults {dataset}:")
+    read_NER_metrics(f"results_{dataset}")

src/experiments/NER_with_LLMs/NER_with_LLMs.py

@@ -41,3 +41,5 @@ def find_entities(model_name, sentence, labels):
         except Exception as e:
             print(e)
     return "No answer was marked, here is the model's response: " + answer
+
+# todo find entities with NLI / MLM

tests/test_datasets.py

@@ -8,7 +8,7 @@ working_datasets = ["Pile-NER-type", "CoNLL", "FIGER-coarse", "FIGER-fine"]
 for dataset in working_datasets:
     print(dataset + ":")
     labels = data_manager.get_labels(dataset)
-    data = data_manager.get_annotated_sentences(dataset)
+    data = data_manager.get_annotated_sentences(dataset, 99999999)
 
     print("First (up to) ten labels: " + str(labels[:10]))
     print("Number of distinct labels: " + str(len(labels)))