Transformer Classifier¶
TransformerClassifier base on transformers library. This is a wrapper of
transformers.AutoModelForSequenceClassification, language model should be one of shortcut in transformers
pretrained models or using one in ['vinai/phobert-base',
'vinai/phobert-large']
TransformerClassifier(num_labels=3, language_model_shortcut='vinai/phobert-base', device='cuda')
-
class
sentivi.classifier.TransformerClassifier(num_labels: Optional[int] = 3, language_model_shortcut: Optional[str] = 'vinai/phobert', freeze_language_model: Optional[bool] = True, batch_size: Optional[int] = 2, warmup_steps: Optional[int] = 100, weight_decay: Optional[float] = 0.01, accumulation_steps: Optional[int] = 50, save_steps: Optional[int] = 100, learning_rate: Optional[float] = 3e-05, device: Optional[str] = 'cpu', optimizer=None, criterion=None, num_epochs: Optional[int] = 10, num_workers: Optional[int] = 2, *args, **kwargs)¶ -
class
TransformerDataset(batch_encodings, labels)¶ -
__init__(batch_encodings, labels)¶ Initialize transformer dataset
- Parameters
batch_encodings –
labels –
-
-
class
TransformerPredictedDataset(batch_encodings)¶ -
__init__(batch_encodings)¶ Initialize transformer dataset
- Parameters
batch_encodings –
-
-
__init__(num_labels: Optional[int] = 3, language_model_shortcut: Optional[str] = 'vinai/phobert', freeze_language_model: Optional[bool] = True, batch_size: Optional[int] = 2, warmup_steps: Optional[int] = 100, weight_decay: Optional[float] = 0.01, accumulation_steps: Optional[int] = 50, save_steps: Optional[int] = 100, learning_rate: Optional[float] = 3e-05, device: Optional[str] = 'cpu', optimizer=None, criterion=None, num_epochs: Optional[int] = 10, num_workers: Optional[int] = 2, *args, **kwargs)¶ Initialize TransformerClassifier instance
- Parameters
num_labels – number of polarities
language_model_shortcut – language model shortcut
freeze_language_model – whether language model is freeze or not
batch_size – training batch size
warmup_steps – learning rate warm up step
weight_decay – learning rate weight decay
accumulation_steps – optimizer accumulation step
save_steps – saving step
learning_rate – training learning rate
device – training and evaluating rate
optimizer – training optimizer
criterion – training criterion
num_epochs – maximum number of epochs
num_workers – number of DataLoader workers
args – arbitrary arguments
kwargs – arbitrary keyword arguments
-
forward(data, *args, **kwargs)¶ Training and evaluating TransformerClassifier instance
- Parameters
data – TransformerTextEncoder output
args – arbitrary arguments
kwargs – arbitrary keyword arguments
- Returns
training and evaluating results
- Return type
str
-
get_overall_result(loader)¶ Get overall result
- Parameters
loader – DataLoader
- Returns
overall result
- Return type
str
-
load(model_path, *args, **kwargs)¶ Load model from disk
- Parameters
model_path – path to model path
args – arbitrary arguments
kwargs – arbitrary keyword arguments
- Returns
-
predict(X, *args, **kwargs)¶ Predict polarities with given list of sentences
- Parameters
X – list of sentences
args – arbitrary arguments
kwargs – arbitrary keyword arguments
- Returns
list of polarities
- Return type
str
-
save(save_path, *args, **kwargs)¶ Save model to disk
- Parameters
save_path – path to saved model
args – arbitrary arguments
kwargs – arbitrary keyword arguments
- Returns
-
class