Fix O(n²) model load time for TreeEnsemble with categorical feature chains#27391
Merged
tianleiwu merged 1 commit intomicrosoft:mainfrom Mar 1, 2026
Merged
Fix O(n²) model load time for TreeEnsemble with categorical feature chains#27391tianleiwu merged 1 commit intomicrosoft:mainfrom
tianleiwu merged 1 commit intomicrosoft:mainfrom
Conversation
Some LightGBM-exported models that make heavy use of categorical features exhibit extremely slow load times (minutes for a single 2.5mb model). Profiling shows that CheckIfSubtreesAreEqual is invoked recursively for many node pairs. A significant portion of this work consists of self-comparisons (left_id == right_id), leading to effectively O(n²) behavior during model loading. This change adds a fast-path for the trivial equality, avoiding unnecessary recursive comparisons. Example results (test models) - 7K BRANCH_EQ nodes: 527 ms → 47 ms (~11× faster) - 106K BRANCH_EQ nodes: 141 s → 80 ms (~1760× faster)
Contributor
Author
|
@xadupre could you take a look 🙏🏻 |
Contributor
Thanks so much for looking into this! IIRC, the current v5 reuses large parts of the v3 implementation. I think one could make a good case for eventually rewriting this operator from scratch at some point 🙈. |
xadupre
approved these changes
Feb 23, 2026
Contributor
|
Can this bei merged and possibly be included in 1.24.3 via #27501 ? |
Contributor
|
/azp run Linux QNN CI Pipeline, Win_TRT_Minimal_CUDA_Test_CI, Windows ARM64 QNN CI Pipeline, Windows GPU Doc Gen CI Pipeline |
|
Azure Pipelines successfully started running 4 pipeline(s). |
tianleiwu
pushed a commit
that referenced
this pull request
Mar 2, 2026
…hains (#27391) ### Description Profiling shows that CheckIfSubtreesAreEqual is invoked recursively for many node pairs for LightGBM models with categorical features. A significant portion of this work consists of self-comparisons (left_id == right_id), leading to effectively O(n²) comparing trees to themselves during model loading. This change adds a fast-path for trivial equality, avoiding unnecessary recursive comparisons. Example results: - model with 7K BRANCH_EQ nodes: 527 ms → 47 ms (~11× faster) - model with 106K BRANCH_EQ nodes: 141 s → 80 ms (~1760× faster) ### Motivation and Context We have some LightGBM exported models that make heavy use of categorical features and exhibit extremely slow load times (minutes for a single 2.5mb model). Heres a diagram to illustrate the issue: <img width="1008" height="1229" alt="image" src="https://github.com/user-attachments/assets/348e16cb-9eec-448f-ac5c-e1edb60e2a3d" /> the 106K model has much longer "member of" chains, with chains that lead into more chains: <details> <summary>"trees"</summary> <img width="1405" height="593" alt="image" src="https://github.com/user-attachments/assets/12f0c43f-5987-4b33-9001-2a2b526e537f" /> </details> Interestingly we did also try using the new onnx.ml opset 5 node that has MEMBER, but it seems even slower as it recreates these branch EQ chains.
tianleiwu
added a commit
that referenced
this pull request
Mar 2, 2026
This cherry-picks the following commits for the release: | Commit ID | PR Number | Commit Title | |-----------|-----------|-------------| | 6e72d31 | #27295 | Remove s_kernel_registry_vitisaiep.reset() in deinitialize_vitisai_ep() | | 41f9056 | #27391 | Fix O(n²) model load time for TreeEnsemble with categorical feature chains | --------- Co-authored-by: zz002 <zhenzew@amd.com> Co-authored-by: Adam Scarr <adam@vektah.net>
This was referenced Mar 9, 2026
This was referenced Mar 16, 2026
deps(nuget): Bump the microsoft-packages group with 8 updates
Ellerbach/azure-ai-search-simulator#76
Closed
This was referenced Mar 23, 2026
deps(nuget): Bump the microsoft-packages group with 8 updates
Ellerbach/azure-ai-search-simulator#80
Closed
This was referenced Mar 30, 2026
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Description
Profiling shows that CheckIfSubtreesAreEqual is invoked recursively for many node pairs for LightGBM models with categorical features. A significant portion of this work consists of self-comparisons (left_id == right_id), leading to effectively O(n²) comparing trees to themselves during model loading.
This change adds a fast-path for trivial equality, avoiding unnecessary recursive comparisons.
Example results:
Motivation and Context
We have some LightGBM exported models that make heavy use of categorical features and exhibit extremely slow load times (minutes for a single 2.5mb model).
Heres a diagram to illustrate the issue:
the 106K model has much longer "member of" chains, with chains that lead into more chains:
"trees"
Interestingly we did also try using the new onnx.ml opset 5 node that has MEMBER, but it seems even slower as it recreates these branch EQ chains.