def construct_seq_class_transformer(options:
KaggleEvaluationOptions) -> Reranker:
try:
model = MonoBERT.get_model(options.model_name, device=options.device)
except OSError:
try:
model = MonoBERT.get_model(
options.model_name,
from_tf=True,
device=options.device)
except AttributeError:
# Hotfix for BioBERT MS MARCO. Refactor.
BertForSequenceClassification.bias = torch.nn.Parameter(
torch.zeros(2))
BertForSequenceClassification.weight = torch.nn.Parameter(
torch.zeros(2, 768))
model = BertForSequenceClassification.from_pretrained(
options.model_name, from_tf=True)
model.classifier.weight = BertForSequenceClassification.weight
model.classifier.bias = BertForSequenceClassification.bias
device = torch.device(options.device)
model = model.to(device).eval()
tokenizer = MonoBERT.get_tokenizer(
options.tokenizer_name, do_lower_case=options.do_lower_case)
return MonoBERT(model, tokenizer)
def construct_seq_class_transformer(
options: DocumentRankingEvaluationOptions) -> Reranker:
model = MonoBERT.get_model(options.model,
from_tf=options.from_tf,
device=options.device)
tokenizer = MonoBERT.get_tokenizer(options.tokenizer_name)
return MonoBERT(model, tokenizer)
def build_bert_reranker(
name_or_path: str = "castorini/monobert-large-msmarco-finetune-only",
device: str = None,
):
"""Returns a BERT reranker using the provided model name or path to load from"""
model = MonoBERT.get_model(name_or_path, device=device)
tokenizer = MonoBERT.get_tokenizer(name_or_path)
return MonoBERT(model, tokenizer)
def construct_seq_class_transformer(
options: DocumentRankingEvaluationOptions) -> Reranker:
model = AutoModelForSequenceClassification.from_pretrained(
options.model, from_tf=options.from_tf)
device = torch.device(options.device)
model = model.to(device).eval()
tokenizer = AutoTokenizer.from_pretrained(options.tokenizer_name)
return MonoBERT(model, tokenizer)
def construct_seq_class_transformer(
options: PassageRankingEvaluationOptions) -> Reranker:
try:
model = AutoModelForSequenceClassification.from_pretrained(
options.model, from_tf=options.from_tf)
except AttributeError:
# Hotfix for BioBERT MS MARCO. Refactor.
BertForSequenceClassification.bias = torch.nn.Parameter(torch.zeros(2))
BertForSequenceClassification.weight = torch.nn.Parameter(
torch.zeros(2, 768))
model = BertForSequenceClassification.from_pretrained(
options.model, from_tf=options.from_tf)
model.classifier.weight = BertForSequenceClassification.weight
model.classifier.bias = BertForSequenceClassification.bias
device = torch.device(options.device)
model = model.to(device).eval()
tokenizer = AutoTokenizer.from_pretrained(options.tokenizer_name)
return MonoBERT(model, tokenizer)
import nltk
from bs4 import BeautifulSoup
from nltk.tokenize import word_tokenize
from collections import Counter
from nltk.corpus import stopwords
from pygaggle.rerank.base import Query, Text
from pygaggle.rerank.transformer import MonoT5
from nltk.tokenize import word_tokenize
from pygaggle.rerank.transformer import MonoBERT
from pygaggle.rerank.base import hits_to_texts
from expanders.relevancefeedback import RelevanceFeedback
from cmn import utils
reranker = MonoBERT()
#@inproceedings{zheng-etal-2020-bert,
# title = "{BERT-QE}: {C}ontextualized {Q}uery {E}xpansion for {D}ocument {R}e-ranking",
# author = "Zheng, Zhi and Hui, Kai and He, Ben and Han, Xianpei and Sun, Le and Yates, Andrew",
# booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2020",
# month = nov,
# year = "2020",
# address = "Online",
# publisher = "Association for Computational Linguistics",
# url = "https://www.aclweb.org/anthology/2020.findings-emnlp.424",
# pages = "4718--4728",
#}
class BertQE(RelevanceFeedback):