import torch
from experimaestro import Param
from xpmir.distributed import DistributableModel
from xpmir.learning.batchers import Batcher
from xpmir.learning.context import TrainerContext
from xpmir.letor.records import (
BaseRecords,
PairwiseRecords,
)
from xpmir.rankers import LearnableScorer
from xpmir.text.encoders import DualTextEncoder, TripletTextEncoder
from xpmir.rankers import (
DuoLearnableScorer,
DuoTwoStageRetriever,
Retriever,
)
from xpmir.utils.utils import easylog
logger = easylog()
[docs]class CrossScorer(LearnableScorer, DistributableModel):
"""Query-Document Representation Classifier
Based on a query-document representation representation (e.g. BERT [CLS] token).
AKA Cross-Encoder
Attributes:
encoder: Document (and query) encoder
query_encoder: Query encoder; if null, uses the document encoder
"""
encoder: Param[DualTextEncoder]
"""an encoder for encoding the concatenated query-document tokens which
doesn't contains the final linear layer"""
def __validate__(self):
super().__validate__()
assert not self.encoder.static(), "The vocabulary should be learnable"
def __initialize__(self, options):
super().__initialize__(options)
self.encoder.initialize(options)
self.classifier = torch.nn.Linear(self.encoder.dimension, 1)
def forward(self, inputs: BaseRecords, info: TrainerContext = None):
# Encode queries and documents
pairs = self.encoder(
[
(tr.topic.get_text(), dr.document.get_text())
for tr, dr in zip(inputs.topics, inputs.documents)
]
) # shape (batch_size * dimension)
return self.classifier(pairs).squeeze(1)
def distribute_models(self, update):
self.encoder = update(self.encoder)
[docs]class DuoCrossScorer(DuoLearnableScorer, DistributableModel):
"""Preference based classifier
This scorer can be used to train a DuoBERT-type model.
"""
encoder: Param[TripletTextEncoder]
"""The encoder to use for the Duobert model"""
def __validate__(self):
assert not self.encoder.static(), "The vocabulary should be learnable"
def __initialize__(self, options):
super().__initialize__(options)
self.encoder.initialize(options)
self.classifier = torch.nn.Linear(self.encoder.dimension, 1)
def forward(self, inputs: PairwiseRecords, info: TrainerContext = None):
"""Encode the query-document-document"""
triplets = self.encoder(
[
(q.text, d_1.text, d_2.text)
for q, d_1, d_2 in zip(
inputs.unique_queries, inputs.positives, inputs.negatives
)
]
)
return self.classifier(triplets).squeeze(1)
def distribute_models(self, update):
self.encoder.model = update(self.encoder.model)
def getRetriever(
self,
retriever: "Retriever",
batch_size: int,
batcher: Batcher = Batcher(),
top_k=None,
device=None,
):
"""The given the base_retriever and return a two stage retriever
specialized for Duobert"""
return DuoTwoStageRetriever(
retriever=retriever,
scorer=self,
batchsize=batch_size,
batcher=batcher,
device=device,
top_k=top_k,
)