Data Utils

Input Example

Input Features

Fewshot Sampler

class FewShotSampler(num_examples_total: Optional[int] = None, num_examples_per_label: Optional[int] = None, also_sample_dev: Optional[bool] = False, num_examples_total_dev: Optional[int] = None, num_examples_per_label_dev: Optional[int] = None)[source]

Few-shot learning is an important scenario for prompt-learning, this is a sampler that samples few examples over each class.

Parameters

  • num_examples_total (int, optional) – Sampling strategy I: Use total number of examples for few-shot sampling.

  • num_examples_per_label (int, optional) – Sampling strategy II: Use the number of examples for each label for few-shot sampling.

  • also_sample_dev (bool, optional) – Whether to apply the sampler to the dev data.

  • num_examples_total_dev (int, optional) – Sampling strategy I: Use total number of examples for few-shot sampling.

  • num_examples_per_label_dev (int, optional) – Sampling strategy II: Use the number of examples for each label for few-shot sampling.

sample_total(indices: List, num_examples_total)[source]

Use the total number of examples for few-shot sampling (Strategy I).

Parameters

  • indices (List) – The random indices of the whole datasets.

  • num_examples_total (int) – The total number of examples.

Returns

The selected indices with the size of num_examples_total.

Return type

List

sample_per_label(indices: List, labels, num_examples_per_label)[source]

Use the number of examples per class for few-shot sampling (Strategy II). If the number of examples is not enough, a warning will pop up.

Parameters

  • indices (List) – The random indices of the whole datasets.

  • labels (List) – The list of the labels.

  • num_examples_per_label (int) – The total number of examples for each class.

Returns

The selected indices with the size of num_examples_total.

Return type

List