Datasets:
Tasks:
Question Answering
Modalities:
Image
Formats:
imagefolder
Languages:
English
Size:
1K - 10K
DOI:
License:
File size: 4,453 Bytes
dcdfcf0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 |
import argparse
import json
import csv
import math
import os
def compute_dcg(pred_docs, gold_docs):
dcg_score = 0.0
for i, doc in enumerate(pred_docs):
position = i + 1
discount = 1.0 / math.log2(position + 1)
relevance = 0.0
if doc in gold_docs:
# If predicted image is present in gold list, set relevance to 1.0
relevance = 1.0
else:
for gdoc in gold_docs:
# If predicted image is a sub-image or parent image of an image in gold list,
# we set relevance to 0.5 to provide partial credit
if doc in gdoc or gdoc in doc:
relevance = 0.5
break
dcg_score += (discount * relevance)
return dcg_score
def compute_idcg(relevance_ranking, rank):
sorted_relevance_ranking = list(sorted(relevance_ranking.items(), key=lambda x: x[1], reverse=True))
# Only consider top k relevant items for IDCG@k
sorted_relevance_ranking = sorted_relevance_ranking[:min(len(sorted_relevance_ranking), rank)]
idcg_score = sum([ (1.0 / (math.log2(i + 2))) * x[1] for i, x in enumerate(sorted_relevance_ranking)])
return idcg_score
def run_eval(pred_labels, gold_labels, parse_folder, claim_citekeys, debug):
ranks_to_eval = [5, 10]
ndcg_scores = {n: {} for n in ranks_to_eval}
non_empty_samples = 0
for claim_id in pred_labels:
if claim_id not in gold_labels:
print(f"Warning: Claim ID {claim_id} not found in gold data - skipping!")
continue
if not gold_labels[claim_id]:
print(f"Warning: Claim ID {claim_id} has no associated evidence figures/tables - skipping!")
continue
non_empty_samples += 1
for rank in ranks_to_eval:
# If #predictions < rank in predicted ranking, include all for evaluation
pred_images = pred_labels[claim_id][:min(len(pred_labels[claim_id]), rank)]
gold_images = gold_labels[claim_id]
# Compute DCG score
dcg_score = compute_dcg(pred_images, gold_images)
# Compute ideal DCG score
# First need to get relevance scores for all possible images
# Images in gold list get relevance score of 1.0
relevance_ranking = {x: 1.0 for x in gold_images}
for file in os.listdir(os.path.join(parse_folder, claim_citekeys[claim_id])):
if 'CAPTION' in file:
continue
image_id = file.split('.png')[0]
if image_id not in gold_images:
relevance_ranking[image_id] = 0.0
# All images that are parent/sub-images of a gold image get relevance of 0.5
for gold_image in gold_images:
if image_id in gold_image or gold_image in image_id:
relevance_ranking[image_id] = 0.5
break
idcg_score = compute_idcg(relevance_ranking, rank)
# Finally compute and store NDCG score@k
ndcg_score = dcg_score / idcg_score
ndcg_scores[rank][claim_id] = ndcg_score
# Display final evaluation scores
for rank in ranks_to_eval:
final_ndcg = sum(list(ndcg_scores[rank].values())) / len(gold_labels)
print(f'NDCG@{rank}: {final_ndcg}')
if debug:
json.dump(ndcg_scores, open("task1_scores.json", "w"))
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--pred_file", type=str, required=True, help="Path to prediction file")
parser.add_argument("--gold_file", type=str, required=True, help="Path to gold data file")
parser.add_argument("--parse_folder", type=str, required=True, help="Path to folder containing parsed images/tables")
parser.add_argument("--debug", type=bool, default=False, help="Dump per-prediction scores for debuggin/analysis")
args = parser.parse_args()
gold_data = json.loads(open(args.gold_file).read())
gold_labels = {x["id"]: x["findings"] for x in gold_data}
claim_citekeys = {x["id"]: x["citekey"] for x in gold_data}
reader = csv.reader(open(args.pred_file))
next(reader, None)
pred_labels = {}
for row in reader:
pred_labels[row[0]] = row[1].split(',')
run_eval(pred_labels, gold_labels, args.parse_folder, claim_citekeys, args.debug)
|