File size: 2,264 Bytes
6074409 | 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 | from transformers import PreTrainedModel
from modules import SmartphoneBERT
import torch
from .configuration_vnsabsa import VnSmartphoneBERTConfig
from typing import Tuple
class VnSmartphoneBERTModel(PreTrainedModel):
config_class = VnSmartphoneBERTConfig
def __init__(
self,
config: VnSmartphoneBERTConfig
):
super().__init__(config)
self.model = SmartphoneBERT(
vocab_size=config.vocab_size,
embed_dim=config.embed_dim,
num_heads=config.num_heads,
num_encoders=config.num_encoders,
encoder_dropout=config.encoder_dropout,
fc_dropout=config.fc_dropout,
fc_hidden_size=config.fc_hidden_size
)
self.ASPECT_LOOKUP = {
i: a
for i, a in enumerate(["CAMERA", "FEATURES", "BATTERY", "PRICE", "GENERAL", "SER&ACC", "PERFORMANCE", "SCREEN", "DESIGN", "STORAGE", "OTHERS"])
}
self.POLARITY_LOOKUP = {
i: p
for i, p in enumerate(["Negative", "Neutral", "Positive"])
}
def forward(
self,
input_ids: torch.Tensor,
attention_mask: torch.Tensor,
aspect_thresholds: float | torch.Tensor = 0.5
):
pred = self.model(input_ids, attention_mask)
result = self.decode_absa(
pred,
aspect_thresholds=aspect_thresholds
)
return result
def decode_absa(
self,
pred: Tuple[torch.Tensor, torch.Tensor],
aspect_thresholds: float | torch.Tensor = 0.5
):
if isinstance(aspect_thresholds, float):
aspect_thresholds = torch.full((11,), aspect_thresholds)
a, p = pred
a = a.sigmoid().cpu()
p = p.argmax(dim=-1).cpu()
results = []
for a_i, p_i in zip(a, p):
res_i = {}
for i in range(10):
a = self.ASPECT_LOOKUP[i]
p = self.POLARITY_LOOKUP[p_i[i].item()]
if a_i[i] >= aspect_thresholds[i]:
res_i[a] = p
results.append(res_i)
# OTHERS
if a_i[-1] >= aspect_thresholds[-1]:
res_i["OTHERS"] = ""
return results |