The proposal is to modify its API to allow users to pass in any number (two or more) IterDataPipe, rather than just strictly two. We will have to think carefully about how the key_fn and ref_key_fn will change.
Motivation, pitch
There are situations where users may want to zip multiple DataPipes together.
Alternatives
Users will have to write their own custom DataPipes.
Additional context
There are similar DataPipes and we should keep in mind that their APIs should be as consistent as possible.
🚀 The feature
Currently,
IterKeyZipper
can only zip two DataPipes together.https://github.com/pytorch/data/blob/198cffe7e65a633509ca36ad744f7c3059ad1190/torchdata/datapipes/iter/util/combining.py#L13
The proposal is to modify its API to allow users to pass in any number (two or more)
IterDataPipe
, rather than just strictly two. We will have to think carefully about how thekey_fn
andref_key_fn
will change.Motivation, pitch
There are situations where users may want to zip multiple DataPipes together.
Alternatives
Users will have to write their own custom DataPipes.
Additional context
There are similar DataPipes and we should keep in mind that their APIs should be as consistent as possible.