def main():
apb = ArgumentParserBuilder()
apb.add_opts(
opt('--task', type=str, default='msmarco'),
opt('--dataset', type=Path, required=True),
opt('--index-dir', type=Path, required=True),
opt('--method', required=True, type=str, choices=METHOD_CHOICES),
opt('--model',
required=True,
type=str,
help='Path to pre-trained model or huggingface model name'),
opt('--duo_model',
type=str,
help='Path to pre-trained model or huggingface model name'),
opt('--mono_hits',
type=int,
default=50,
help='Top k candidates from mono for duo reranking'),
opt('--output-file', type=Path, default='.'),
opt('--overwrite-output', action='store_true'),
opt('--split', type=str, default='dev', choices=('dev', 'eval')),
opt('--batch-size', '-bsz', type=int, default=96),
opt('--device', type=str, default='cuda:0'),
opt('--is-duo', action='store_true'),
opt('--from-tf', action='store_true'),
opt('--metrics',
type=str,
nargs='+',
default=metric_names(),
choices=metric_names()), opt('--model-type', type=str),
opt('--tokenizer-name', type=str))
args = apb.parser.parse_args()
options = PassageRankingEvaluationOptions(**vars(args))
logging.info("Preprocessing Queries & Passages:")
ds = MsMarcoDataset.from_folder(str(options.dataset),
split=options.split,
is_duo=options.is_duo)
examples = ds.to_relevance_examples(str(options.index_dir),
is_duo=options.is_duo)
logging.info("Loading Ranker & Tokenizer:")
construct_map = dict(transformer=construct_transformer,
bm25=construct_bm25,
t5=construct_t5,
duo_t5=construct_duo_t5,
seq_class_transformer=construct_seq_class_transformer,
random=lambda _: RandomReranker())
reranker = construct_map[options.method](options)
writer = MsMarcoWriter(args.output_file, args.overwrite_output)
if options.method == 'duo_t5':
evaluator = DuoRerankerEvaluator(mono_reranker=reranker[0],
duo_reranker=reranker[1],
metric_names=options.metrics,
mono_hits=options.mono_hits,
writer=writer)
else:
evaluator = RerankerEvaluator(reranker, options.metrics, writer=writer)
width = max(map(len, args.metrics)) + 1
logging.info("Reranking:")
for metric in evaluator.evaluate(examples):
logging.info(f'{metric.name:<{width}}{metric.value:.5}')
def main():
apb = ArgumentParserBuilder()
apb.add_opts(opt('--dataset',
type=str,
default='msmarco'),
opt('--data-dir', type=Path, default='/content/data/msmarco'),
opt('--method',
required=True,
type=str,
choices=METHOD_CHOICES),
opt('--model-name-or-path', type=str),
opt('--output-file', type=Path, default='.'),
opt('--overwrite-output', action='store_true'),
opt('--split',
type=str,
default='dev',
choices=('dev', 'eval')),
opt('--batch-size', '-bsz', type=int, default=96),
opt('--device', type=str, default='cuda:0'),
opt('--is-duo', action='store_true'),
opt('--metrics',
type=str,
nargs='+',
default=metric_names(),
choices=metric_names()),
opt('--model-type', type=str, default='bert-base'),
opt('--tokenizer-name', type=str),
opt('--index-dir', type=Path))
args = apb.parser.parse_args()
options = PassageRankingEvaluationOptions(**vars(args))
ds = MsMarcoDataset.from_folder(str(options.data_dir), split=options.split,
is_duo=options.is_duo)
examples = ds.to_relevance_examples(SETTINGS.msmarco_index_path,
is_duo=options.is_duo)
construct_map = dict(transformer=construct_transformer,
bm25=construct_bm25,
t5=construct_t5,
seq_class_transformer=construct_seq_class_transformer,
random=lambda _: RandomReranker())
reranker = construct_map[options.method](options)
writer = MsMarcoWriter(args.output_file, args.overwrite_output)
evaluator = RerankerEvaluator(reranker, options.metrics, writer=writer)
width = max(map(len, args.metrics)) + 1
stdout = []
for metric in evaluator.evaluate(examples):
logging.info(f'{metric.name:<{width}}{metric.value:.5}')
stdout.append(f'{metric.name}\t{metric.value}')
print('\n'.join(stdout))
