easynlp.fewshot

PET

class easynlp.fewshot_learning.fewshot_application.FewshotClassification(pretrained_model_name_or_path=None, user_defined_parameters=None, **kwargs)

An application class for supporting fewshot learning (PET and P-tuning).

forward(inputs)

compute_loss(forward_outputs, label_ids)

classmethod from_pretrained(pretrained_model_name_or_path, **kwargs)

class easynlp.fewshot_learning.fewshot_dataset.FewshotBaseDataset(pretrained_model_name_or_path, data_file, max_seq_length, first_sequence, input_schema=None, user_defined_parameters=None, label_name=None, second_sequence=None, label_enumerate_values=None, **kwargs)

This is a dataset class for supporting fewshot learning.

Parameters:

• pretrained_model_name_or_path -- The name of the path of th pretrained model for fewshot classification.
• data_file -- The input data file.
• max_seq_length -- The maximum sequence length for the transformer model.
• first_sequence -- The name of the first sequence in the input schema.
• input_schema -- The schema of the input data file.
• user_defined_parameters -- The dict of user defined parameters for fewshot classification.
• label_name -- The name of the label name in the input schema.
• second_sequence -- The name of the second sequence in the input schema.
• label_enumerate_values -- The string of all label values, seperated by comma.

eval_metrics

label_enumerate_values

convert_single_row_to_example(row)

Converting the examples into the dict of values.

batch_fn(features)

Divide examples into batches.

class easynlp.fewshot_learning.fewshot_evaluator.PromptEvaluator(valid_dataset, **kwargs)

An evaluator class for supporting fewshot learning (PET and P-tuning).

evaluate(model)

class easynlp.fewshot_learning.fewshot_predictor.FewshotPyModelPredictor(model_cls, saved_model_path, input_keys, user_defined_parameters, output_keys, **kwargs)

This is the predictor class for supporting fewshot learning.

Parameters:

• model_cls -- The classification model of the pretrained model for fewshot classification.
• saved_model_path -- The path of the saved model.
• input_keys -- The collection of input keys for prediction.
• user_defined_parameters -- The dict of user defined parameters for fewshot prediction.
• output_keys -- The collection of output keys for prediction.

predict(in_data)

The main predictor function for fewshot learning.

Parameters:in_data -- the dict of input data

Returns: the dict of output tensors containing the results

class easynlp.fewshot_learning.fewshot_predictor.PromptPredictor(model_dir, model_cls=None, user_defined_parameters=None, *args, **kwargs)

This is the predictor class for supporting prompt-based fewshot learning.

Parameters:

• model_dir -- The path of the model
• model_cls -- The classification model of the pretrained model for fewshot classification.
• user_defined_parameters -- The dict of user defined parameters for fewshot prediction.

preprocess(in_data)

The preprocess step of model prediction.

Parameters:in_data -- the dict of input data

Returns: the dict of output tensors containing the results

predict(in_data)

The main function calling the predictor.

postprocess(result)

The postprocess that converts logits to final results.

Parameters:result -- the dict of prediction results by the model

Returns: the list of the final results.

CPT

class easynlp.fewshot_learning.fewshot_application.CPTClassification(pretrained_model_name_or_path=None, user_defined_parameters=None, **kwargs)

An application class for supporting CPT fewshot learning.

forward(inputs, do_mlm=False)

compute_loss(forward_outputs, label_ids)

class easynlp.fewshot_learning.fewshot_application.CircleLoss(margin: float = 0.4, gamma: float = 64, k: float = 1, distance_function='cos')

forward(features, labels)

class easynlp.fewshot_learning.fewshot_evaluator.CPTEvaluator(valid_dataset, *args, **kwargs)

An evaluator class for supporting CPT fewshot learning.

evaluate(model)

class easynlp.fewshot_learning.fewshot_predictor.CPTPredictor(model_dir, model_cls=None, user_defined_parameters=None, *args, **kwargs)

This is the predictor class for supporting CPT fewshot learning.

Parameters:

• model_dir -- The path of the model
• model_cls -- The classification model of the pretrained model for fewshot classification.
• user_defined_parameters -- The dict of user defined parameters for fewshot prediction.

preprocess(in_data)

The preprocess step of model prediction.

Parameters:in_data -- the dict of input data

Returns: the dict of output tensors containing the results

predict(in_data)

The main function calling the predictor.

postprocess(result)

The postprocess that converts embeddings of CPT to final results.

Parameters:result -- the dict of prediction results by the model

Returns: the list of the final results.