Description

Dataset.zip() has strict requirements for underlying data layout. It requires both datasets to be zipped, have same number of blocks, and same number of rows per block. It also requires the block formats to be same. It would throw exceptions without clear action items: Cannot zip .... Based on several users feedback, the API is not easy to use.

We should either:

• Handle all the blocks alignment and format conversion internally in zip(). E.g. align the number of blocks, and convert to same block format (e.g. simple/Pandas -> Arrow block). Add example in our documentation for zip.
• Improve error message with action items.

Use case

as above.