Create multiple views of a single field

[yjxiao]

Is there a way to create multiple views of a single field in the v0.3.0? For example, getting both the numericalized sequences as well as the bag-of-words representations from the same text field.

[mttk]

Not yet, but this is something planned.

[akurniawan]

Hi @mttk, bumping up this request. Is this something that you are working on right now? I'm also happy to help if needed!

[mttk]

@akurniawan the repo is undergoing a large overhaul by @zhangguanheng66 and @cpuhrsch, and I think this might be one of their future goals. I'd suggest you wait until the new classes are fleshed out though. Either way, George & Christian will know more information.

[akurniawan]

@mttk thanks for the info! Is there any roadmap that all of us can see?

[zhangguanheng66]

@akurniawan. We appreciate the feedbacks from the open source community. There are a few things we want to do, including a C++ extension for the dictionary, sentencepiece.

[cpuhrsch]

Currently we're only adding separate features or improving existing features without changing the API surface. The upcoming release notes will include a detailed review of these changes. That means this isn't an overhaul just yet, since it barely touches anything currently available, but we are reviewing existing issues and are adding features with some of them in mind.

Both George and I have only recently started working on torchtext and only George is working on it full-time. A few of the items George mentioned are on the roadmap, but we can't make statements with respect to their priority or the corresponding tasks. We understand that torchtext has been lacking in the past and are working on improving the building blocks we provide in this library to make it easier to write datasets and build models. Stay tuned.