Add function to read Pretrain word vector

[gau820827]

Write a function to load in the dataset and transform to pre-train word embedding.

For example, build an iterator on train, eval, and test set.
train_iter = get_batch(train_set)
a, b, org_a, org_b = next(train_iter)

This should get a batch of pairs a, b where the dimensions are (batch_size, sentence_length, embedding_size). Add paddings to make the sentence length fixed in one batch if it's convenient.

Note that org_a and org_b are original tokens in a list. The dimensions are (batch_size, sentence_length)

[o9812]

Done