F1 by keyword

evaluation.py

@@ -115,10 +115,24 @@ def f1_by_input_length(split: int = 5) -> dict[tuple[int, int], float]:
     return results
 
 
+def f1_by_keyword() -> dict[str, float]:
+    keywords = combined["keyword"].unique()
+    result_dict = {}
+    f1 = load("f1")
+
+    for keyword in keywords:
+        cur = combined[combined["keyword"] == keyword]
+        print(f'{keyword} has {len(cur)} entries')
+        result_dict[keyword] = f1.compute(predictions=cur["pcl"].to_list(), references=cur["is_patronising"].to_list())
+
+    return result_dict
+
+
 if __name__ == '__main__':
     print("F1:", f1_score())
     print("Recall:", recall())
     print("Precision:", precision())
     print("F1 by input length (fixed):", f1_by_input_length_fixed())
+    print("F1 by keyword:", f1_by_keyword())
     for i in range(2, 17):
         print(f1_by_input_length(i))