MMSRL

Our Multimodal Semantic Role Labeling model for CONSTRAINT 2022
git clone https://esimon.eu/repos/MMSRL.git
Log | Files | Refs | Submodules | README

submission.py (2396B)

      1 from typing import Any, Dict
      2 import argparse
      3 import collections
      4 import json
      5 import os
      6 import pathlib
      7 import pickle
      8 import zipfile
      9 
     10 import mmsrl.utils
     11 
     12 
     13 def make_submission_file(config: mmsrl.utils.dotdict) -> None:
     14     inputs: Dict[str, Dict[str, Any]] = collections.defaultdict(dict)
     15     with config.prediction.open("rb") as input_file:
     16         try:
     17             while True:
     18                 image, entities = pickle.load(input_file)
     19                 for entity, prediction in entities.items():
     20                     inputs[image][entity] = prediction
     21         except EOFError:
     22             pass
     23 
     24     data_path = pathlib.Path(os.environ['DATA_PATH'])
     25     annot_file = data_path / "mmsrl" / "all" / "annotations" / f"unseen_test.jsonl"
     26     with annot_file.open('r') as json_file:
     27         dataset = list(map(json.loads, json_file))
     28 
     29     outputfile_path = config.output /  pathlib.Path(config.prediction.stem).with_suffix('.json')
     30     with outputfile_path.open("w") as output_file:
     31         for sample in dataset:
     32             image = sample["image"]
     33             output: Dict[str, Any] = {"image": image}
     34             for label in mmsrl.utils.LABELS:
     35                 output[label] = []
     36 
     37             for i, entity in enumerate(sample["entity_list"]):
     38                 if entity == "":
     39                     prediction = mmsrl.utils.LABELS.index("other")
     40                 else:
     41                     predictions = [(p, i) for i, p in enumerate(inputs[image][entity].tolist())]
     42                     predictions.sort(reverse=True)
     43                     if len(set(entities)) == len(entities) or entities.count(entity) == 1:
     44                         prediction = predictions[0][1]
     45                     else:
     46                         prediction = predictions[entities[:i].count(entity)][1]
     47                 output[mmsrl.utils.LABELS[prediction]].append(entity)
     48 
     49             print(json.dumps(output), file=output_file)
     50 
     51     with zipfile.ZipFile(str(outputfile_path.with_suffix('.zip')), mode="w") as archive:
     52         archive.write(str(outputfile_path), arcname="answer.jsonl")
     53 
     54 if __name__ == "__main__":
     55     parser = argparse.ArgumentParser()
     56     parser.add_argument("output", type=pathlib.Path, help="Path were we write the submission json file.")
     57     parser.add_argument("prediction", type=pathlib.Path, help="Prediction file to use.")
     58     args = mmsrl.utils.dotdict(vars(parser.parse_args()))
     59     make_submission_file(args)