⚡️ Speed up function aggregate_embedded_text_by_block by 15% in PR #4169 (fix/partially-filled-inferred-layout-mark-as-not-extracted)
#4172
+66
−19
Conversation
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
…inferred-layout-mark-as-not-extracted
…inferred-layout-mark-as-not-extracted
The optimized code achieves a **14% speedup** by eliminating redundant calls to `source_regions.slice(mask)`. **Key optimization:** The original code called `source_regions.slice(mask)` three separate times: 1. To extract `.texts` for text aggregation 2. To get `.element_coords` for IoU calculation 3. To access `.is_extracted_array` for extraction verification The optimized version caches this result in `masked_source = source_regions.slice(mask)` and reuses it. This single change reduces execution time from 377μs to 328μs. **Why this matters:** According to the line profiler, each `slice()` call has non-trivial overhead (~70-90μs). By caching the result, we eliminate two redundant slice operations when `sum(mask) > 0`. The profiler shows the cached access to `masked_source.element_coords` takes only 1.4μs vs 71μs for the repeated `source_regions.slice(mask).element_coords` call, and accessing `masked_source.is_extracted_array` takes 15.5μs vs 81.8μs. **Impact context:** Based on the function reference, `aggregate_embedded_text_by_block` is called within a loop in `merge_out_layout_with_ocr_layout`, iterating over `invalid_text_indices`. For PDF documents with many text regions requiring OCR aggregation, this optimization compounds significantly. Each iteration saves ~100μs (from avoided redundant slicing), which adds up when processing documents with dozens or hundreds of text blocks. **Performance characteristics:** The optimization is most effective when: - `mask` contains True values (common case where text regions overlap) - The source_regions contain multiple elements to slice - The function is called repeatedly in hot paths (as evidenced by the OCR merge workflow)
codeflash-ai
bot
added
⚡️ codeflash
Contributor
Author
|
Base automatically changed from
fix/partially-filled-inferred-layout-mark-as-not-extracted
to
main
January 7, 2026 22:17
codeflash-ai
bot
deleted the
codeflash/optimize-pr4169-2026-01-07T19.05.25
branch
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.
⚡️ This pull request contains optimizations for PR #4169
If you approve this dependent PR, these changes will be merged into the original PR branch
fix/partially-filled-inferred-layout-mark-as-not-extracted.📄 15% (0.15x) speedup for
aggregate_embedded_text_by_blockinunstructured/partition/pdf_image/pdfminer_processing.py⏱️ Runtime :
377 microseconds→328 microseconds(best of8runs)📝 Explanation and details
The optimized code achieves a 14% speedup by eliminating redundant calls to
source_regions.slice(mask).Key optimization:
The original code called
source_regions.slice(mask)three separate times:.textsfor text aggregation.element_coordsfor IoU calculation.is_extracted_arrayfor extraction verificationThe optimized version caches this result in
masked_source = source_regions.slice(mask)and reuses it. This single change reduces execution time from 377μs to 328μs.Why this matters:
According to the line profiler, each
slice()call has non-trivial overhead (~70-90μs). By caching the result, we eliminate two redundant slice operations whensum(mask) > 0. The profiler shows the cached access tomasked_source.element_coordstakes only 1.4μs vs 71μs for the repeatedsource_regions.slice(mask).element_coordscall, and accessingmasked_source.is_extracted_arraytakes 15.5μs vs 81.8μs.Impact context:
Based on the function reference,
aggregate_embedded_text_by_blockis called within a loop inmerge_out_layout_with_ocr_layout, iterating overinvalid_text_indices. For PDF documents with many text regions requiring OCR aggregation, this optimization compounds significantly. Each iteration saves ~100μs (from avoided redundant slicing), which adds up when processing documents with dozens or hundreds of text blocks.Performance characteristics:
The optimization is most effective when:
maskcontains True values (common case where text regions overlap)✅ Correctness verification report:
⚙️ Click to see Existing Unit Tests
partition/pdf_image/test_pdfminer_processing.py::test_aggregate_by_blockpartition/pdf_image/test_pdfminer_processing.py::test_aggregate_only_partially_fill_targetTo edit these changes
git checkout codeflash/optimize-pr4169-2026-01-07T19.05.25and push.