improves dataloader performance#687
Merged
Merged
Conversation
Member
Author
|
quick push to try #699 where tiling is vectorized, removed the need for
|
This was referenced Sep 3, 2024
**Bugs fixed in datasets.py:** - rasterize=True path was broken: __getitem__ always called image.sel() regardless of rasterize flag, bypassing rasterize_fn entirely. Fixed by storing self._rasterize and branching in __getitem__. - ad.concat(*tables_l) unpacked the list as positional args, failing with >1 region. Fixed to ad.concat(tables_l). - Vectorized selection pre-computation was always run even for rasterize=True where it is unused. Fixed by guarding with `if not rasterize`. - Removed stale commented-out pandas.apply fallback code. **Fixes in _utils.py:** - Removed redundant nopython=True from @nb.njit (njit implies nopython=True, and the argument caused a RuntimeWarning). - Replaced invalid nb.types.Array[nb.float64, nb.float64] annotations with np.ndarray. **Fixes in spatial_query.py:** - Restored BoundingBoxRequest validation that was commented out. The validator's __post_init__ already handles both 1-D (single box) and 2-D (multi-box) arrays. **Benchmark (benchmark_dataloader.py):** Synthetic 2048x2048 image, 500 circle regions (32 px radius), 3-channel. Phase main PR (fixed) speedup init ~162 ms ~20 ms ~8x fetch 500 ~618 ms ~118 ms ~5x per-tile ~1237 us ~235 us ~5x Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Member
|
I picked this up, performance indeed improves significantly with the new vectorized bounding box approach. Thanks @giovp
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #687 +/- ##
==========================================
+ Coverage 92.28% 92.55% +0.26%
==========================================
Files 51 51
Lines 7804 7763 -41
==========================================
- Hits 7202 7185 -17
+ Misses 602 578 -24
🚀 New features to boost your workflow:
|
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
2 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.
replaces #565 #622
iteration over 20 batches, single worker
new implementation
main
one annoying thing is that the "apply" method of the dataframe to get the bounding box selection is quite slow.