Question data for E5 and ColBERT online and formatted

create_json_data.py

@@ -0,0 +1,37 @@
+import json
+import csv
+
+with open("question_data.csv", "r") as f:
+ reader = csv.reader(f)
+ questions = []
+ for row in reader:
+ questions.append(row)
+
+questions = questions[2:] # cut off top two (labels, passage #s)
+
+# qid,prompt,question,a,b,c,d,answer,gold_passage,top10_colbert,,,,,,,,,,generation_colbert,top10_e5,,,,,,,,,,generation_e5,gold_passage_generation
+# 0 1 2 3 4 5 6 7 8 9-18 19 20-29 30 31
+# See example.json for how these files will be ported
+
+full_question_dict = {} # stores all "id":q_data pairs
+for entry in questions:
+ # Create individual question data
+ q_data = {}
+ if not entry[1] == "":
+ entry[2] = entry[1] + " " + entry[2]
+ q_data["question"] = entry[2]
+ q_data["answers"] = entry[3:7] # inclusive of (3, 6) -> A, B, C, D
+ answer_map = {"A": 0, "B": 1, "C": 2, "D": 3}
+ q_data["correct_answer_index"] = answer_map[entry[7]] # entry[7] = "A" -> index = 0
+ q_data["top10_colbert"] = entry[9:19] # inclusive of (9-18) -> 10 retrievals
+ q_data["generation_colbert"] = entry[19]
+ q_data["top10_e5"] = entry[20:30] # inclusive of (20-29) -> 10 retrievals
+ q_data["generation_e5"] = entry[30]
+ q_data["top10_contains_gold_passage"] = False # this is always the case b/c of programming. Does not reflect reality
+ q_data["gold_passage"] = entry[8]
+ q_data["gold_passage_generation"] = entry[31]
+ # Add to full question dictionary
+ full_question_dict[entry[0]] = q_data # entry[0] is qid
+
+with open("question_data.json", "w") as f:
+ json.dump(full_question_dict, f)