[AIR - Datasets] [Hotfix] Tensor extension column concatenation fixes. #29479
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
clarkzinzow
requested review from
ericl,
scv119,
jjyao,
jianoaix and
c21
as code owners
clarkzinzow
changed the title
jianoaix
reviewed
Oct 20, 2022
clarkzinzow
force-pushed
the
datasets/fix/concatenating-tensor-columns
branch
from
cf9eec1
to
c382d5c
Compare
|
@jianoaix Feedback implemented, PTAL. I opted to add test coverage for the particular |
|
I've been pushing for full unit test coverage of the block layer for a while, would definitely still like to do that in a dedicated PR. |
jianoaix
approved these changes
Oct 21, 2022
|
Failing tests appear to be unrelated, merging. |
13 tasks
WeichenXu123
pushed a commit
to WeichenXu123/ray
that referenced
this pull request
Dec 19, 2022
ray-project#29479) Fixes the concatenation of tensor columns (and extension types in general). Previously, concatenating tensor columns may produce a variable-shaped tensor but would be represented with a broken homogeneous-shaped tensor. This PR ensures that the correct tensor extension type is produced post-concatenation. Issue Details: The core issue is that we weren’t accounting for homogeneous-shaped tensor column concatenation resulting in a heterogeneous-shaped tensor column. In master, this currently fails silently until accessing the data fails somewhere downstream of the concatenation. E.g., if you have 5 images of shape (32, 32) in one block, and 5 images of shape (64, 64) in another block, the image column in each individual block is homogeneous-shaped. But if you concatenate those blocks, you have 10 images, half with shape (32, 32) and half with shape (64, 64), which needs our heterogeneous-shaped (variable-shaped) tensor column representation. Signed-off-by: Weichen Xu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes the concatenation of tensor columns (and extension types in general). Previously, concatenating tensor columns may produce a variable-shaped tensor but would be represented with a broken homogeneous-shaped tensor. This PR ensures that the correct tensor extension type is produced post-concatenation.
Issue Details
The core issue is that we weren’t accounting for homogeneous-shaped tensor column concatenation resulting in a heterogeneous-shaped tensor column. In master, this currently fails silently until accessing the data fails somewhere downstream of the concatenation.
E.g., if you have 5 images of shape (32, 32) in one block, and 5 images of shape (64, 64) in another block, the image column in each individual block is homogeneous-shaped. But if you concatenate those blocks, you have 10 images, half with shape (32, 32) and half with shape (64, 64), which needs our heterogeneous-shaped (variable-shaped) tensor column representation.
Issue
Closes #29489
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.