I am reading the NLP tutorial of official documents in https://pytorch.org/tutorials/intermediate/char_rnn_classification_tutorial.html and I think this tutorial is very good to understand RNN. However, I think it is not enough to learn about the pytorch with text processing. In this tutorial, it don’t use the recommend way to build the dataset such as torch.utils.data.Dataset, also it don’t show how to use dataloader to load the batch data to train the network.
I recommend add this part into the tutorial and this Blog has introduced some details of implementing it. Also, the tutorial can introduce more about pack_padded_sequence, opad_packed_sequence because when it makes me really confused when I want to create the dataloader with variable length tensors. Although your forums has many questions on this topic, maybe this kind of common problems can be added into the professional tutorials other than letting user to identify the solutions from dozons of reference documents.
Thanks!