def make_dataset(language: Language, mapping: dict, num_positive: int,
num_negative: int):
positive_strings = language.generate_batch_data(
num_datapoints=num_positive, truth_value=True)
negative_strings = language.generate_batch_data(
num_datapoints=num_negative, truth_value=False)
# TODO: fix language._max_length
max_length = max(
[len(string) for string in positive_strings + negative_strings])
# print(max_length)
positive_tensor, positive_lengths = strings_to_tensor(
positive_strings, max_length, mapping)
positive_labels = torch.zeros(len(positive_strings), dtype=torch.long)
negative_tensor, negative_lengths = strings_to_tensor(
negative_strings, max_length, mapping)
negative_labels = torch.ones(len(negative_strings), dtype=torch.long)
all_tensors = torch.cat((positive_tensor, negative_tensor))
all_labels = torch.cat((positive_labels, negative_labels))
all_lengths = torch.cat((positive_lengths, negative_lengths))
return LanguageDataset(all_tensors, all_labels,
all_lengths) # unshuffled, must be shuffled at use
