How I can convert row of vector back to words

nlp
1

Dear friends,

I use BucketIterator to generate vector as below:

train_iterator, valid_iterator, test_iterator = BucketIterator.splits(
(train_data, valid_data, test_data), sort_key=lambda x: len(x.src),
batch_size = BATCH_SIZE,
device = device)

successfully, I get ( train_iterator, valid_iterator, test_iterator), and I can access each vector as :

Get_vector_row = next(iter(train_iterator))

vars(Get_vector_row).keys()

dict_keys([‘batch_size’, ‘dataset’, ‘fields’, ‘input_fields’, ‘target_fields’, ‘src’, ‘trg’])

Get_vector_row.src[0:]

tensor([[ 2, 293, 1674, …, 1, 1, 1],
[ 2, 720, 2373, …, 1, 1, 1],
[ 2, 114, 6271, …, 1, 1, 1],
…,
[ 2, 135, 5987, …, 2345, 4, 3],
[ 2, 26, 1306, …, 1, 1, 1],
[ 2, 1186, 1780, …, 1, 1, 1]], device=‘cuda:0’)

My question is how to convert the tensor in the previous line back to words?