Upload pipeline.py
Browse files- pipeline.py +29 -0
pipeline.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import BertTokenizer
|
3 |
+
from foody_bert import FoodyBertForSequenceClassification
|
4 |
+
|
5 |
+
class PreTrainedPipeline():
|
6 |
+
|
7 |
+
def __init__(self, path=""):
|
8 |
+
|
9 |
+
"""
|
10 |
+
Initialize model
|
11 |
+
"""
|
12 |
+
self.bert_tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
|
13 |
+
self.model = FoodyBertForSequenceClassification.from_pretrained(".")
|
14 |
+
def __call__(self, inputs: str) -> List[float]:
|
15 |
+
|
16 |
+
"""
|
17 |
+
Args:
|
18 |
+
inputs (:obj:`str`):
|
19 |
+
a string to get the features of.
|
20 |
+
Return:
|
21 |
+
A :obj:`list` of floats: The features computed by the model.
|
22 |
+
"""
|
23 |
+
input_ids = self.bert_tokenizer.encode(inputs, add_special_tokens=True)
|
24 |
+
X = torch.tensor([input_ids])
|
25 |
+
with torch.no_grad():
|
26 |
+
predicted_class_id = self.model(X).logits.argmax().item()
|
27 |
+
labels = ['positive','neutral','negative']
|
28 |
+
reps = labels[predicted_class_id]
|
29 |
+
return self.model.get_sentence_vector(inputs).tolist()
|