README.md

metadata

task_categories:
  - text-generation
language:
  - en
size_categories:
  - 10K<n<100K

Magic the gathering dataset

This dataset contains text of all magic the gathering cards. Example usage:

from datasets import load_dataset

dataset = load_dataset('augustoperes/mtg_text')
dataset

# outputs:
# DatasetDict({
# train: Dataset({
# features: ['card_name', 'type_line', 'oracle_text'],
# num_rows: 20063
# })
# validation: Dataset({
# features: ['card_name', 'type_line', 'oracle_text'],
# num_rows: 5016
# })
# })

Elements of the dataset are, for example:

train_dataset = dataset['train']
train_dataset[0]

# Outputs
# {'card_name': 'Recurring Insight',
# 'type_line': 'Sorcery',
# 'oracle_text': "Draw cards equal to the number of cards in target opponent's hand.\nRebound (If you cast this spell from your hand, exile it as it resolves. At the beginning of your next upkeep, you may cast this card from exile without paying its mana cost.)"}

Example usage with Pytorch

You can easily tokenize, convert and pad this dataset to be usable in pytorch with:

from transformers import AutoTokenizer

import torch
from torch.nn.utils.rnn import pad_sequence

tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

def tokenize(sample):
    sample["card_name"] = tokenizer(sample["card_name"])["input_ids"]
    sample["type_line"] = tokenizer(sample["type_line"])["input_ids"]
    sample["oracle_text"] = tokenizer(sample["oracle_text"])["input_ids"]
    return sample

tokenized_dataset = train_dataset.map(tokenize)

def collate_fn(sequences):
    # Pad the sequences to the maximum length in the batch
    card_names = [torch.tensor(sequence['card_name']) for sequence in sequences]
    type_line = [torch.tensor(sequence['type_line']) for sequence in sequences]
    oracle_text = [torch.tensor(sequence['oracle_text']) for sequence in sequences]

    padded_card_name = pad_sequence(card_names, batch_first=True, padding_value=0)
    padded_type_line = pad_sequence(type_line, batch_first=True, padding_value=0)
    padded_oracle_text = pad_sequence(oracle_text, batch_first=True, padding_value=0)

    return {'card_name': padded_card_name, 'type_line': padded_type_line, 'padded_oracle_text': padded_oracle_text}

loader = torch.utils.data.DataLoader(tokenized_dataset, collate_fn=collate_fn, batch_size=4)

for e in loader:
    print(e)
    break

# Will output:
# {'card_name': tensor([[ 101, 10694, 12369, 102, 0],
# [ 101, 3704, 9881, 102, 0],
# [ 101, 22639, 20066, 7347, 102],
# [ 101, 25697, 1997, 6019, 102]]),
# 'type_line': tensor([[ 101, 2061, 19170, 2854, 102, 0, 0],
# [ 101, 6492, 1517, 4743, 102, 0, 0],
# [ 101, 6492, 1517, 22639, 102, 0, 0],
# [ 101, 4372, 14856, 21181, 1517, 15240, 102]]),
# 'padded_oracle_text': [ommited for readability])}