| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
[docs] [ci] Update docs and source code for org change (#7188) * update organization reference, copyrights * add note about the repo change * make all GPU CI manual-only | 5 个月前 | |
[c++][python-package] Leverage Arrow PyCapsule Interface for Arrow interoperability (including polars support) (#7275) | 2 个月前 | |
[docs] [ci] Update docs and source code for org change (#7188) * update organization reference, copyrights * add note about the repo change * make all GPU CI manual-only | 5 个月前 | |
[docs] [ci] Update docs and source code for org change (#7188) * update organization reference, copyrights * add note about the repo change * make all GPU CI manual-only | 5 个月前 | |
[python-package] Allow to pass Arrow array & Polars series as position (#7260) | 1 个月前 | |
normalize line endings to LF, enforce that with gitattributes (#7291) | 3 个月前 | |
[c++] remove unused ArrayArgs::CheckAllZero(), reduce macro scope in socket_wrapper (#7201) | 5 个月前 | |
Fix calculation of percentiles for weighted training with regression objectives (including `mape`, `quantile`, and `regression_l1`) (#7224) * fix: use correct CDF segment in WeightedPercentileFun WeightedPercentileFun in src/objective/regression_objective.hpp used the segment [pos, pos+1] for linear interpolation, but upper_bound guarantees weighted_cdf[pos-1] <= threshold < weighted_cdf[pos], so the threshold lies in segment [pos-1, pos]. The old numerator (threshold - weighted_cdf[pos]) was always negative and could produce weighted percentile values below min(y) — e.g. weighted median of y=[2,3,4,5], w=[4,3,2,1] returned ~1.0 (or 2.0 in boundary cases) instead of the correct 2.333. This affects regression_l1, quantile, and mape objectives with sample weights (plus mape without weights, since its internal label_weight_ always drives WeightedPercentileFun). Both BoostFromScore and RenewTreeOutput share the macro. Align the weighted implementation with the unweighted PercentileFun and with the fix in #5848 for the same class of bug. Also: - Change the small-gap fallback from v2 to v1. Since threshold is strictly below weighted_cdf[pos], snapping toward v1 keeps the result inside [min(y), max(y)]. - Use 1.0 (double) instead of 1.0f for the gap sentinel so the literal matches the surrounding weighted_cdf double arithmetic. - Loosen the pred_mean lower bound in test_mape_for_specific_boosting_types from >8 to >5. The corrected MAPE training output for the synthetic regression dataset shifts from ~9 to ~6.8. The test's intent is to guard against MAPE being stuck in [0, 1] (see #1579); >5 still serves that purpose. Adds test_weighted_percentile_inside_label_range parametrized on regression_l1, quantile, and mape, using the example from the issue. Each variant asserts the resulting boosted score stays within [min(y), max(y)], and regression_l1 additionally asserts the exact weighted median 2 + 1/3. Fixes #7151 * fix linting errors --------- Co-authored-by: ColorfulData <125739631+ColorfulData@users.noreply.github.com> Co-authored-by: James Lamb <jaylamb20@gmail.com> | 2 个月前 | |
[cuda] Fix dense histogram offset issue for global-memory (#7261) [cuda] fix discretized dense histogram global-memory offset CUDAConstructDiscretizedHistogramDenseKernel_GlobalMemory mis-based its shared_hist_packed pointer on partition_column_start (column-index space) where partition_hist_start (bin space) is required. The two coincide only when every column has exactly one bin, which never holds in real histogram-based training. The three analogous kernels in the same file (non-discretized dense at line 156, non-discretized sparse at line 215, discretized sparse at line 472) all use partition_hist_start. Affects only the dense + use_quantized_grad + USE_GLOBAL_MEM_BUFFER dispatch path (line 615). Flips the regression test added in the previous commit from FAIL to PASS. Co-authored-by: shiyu1994 <shiyu_k1994@qq.com> | 3 个月前 | |
[docs] [ci] Update docs and source code for org change (#7188) * update organization reference, copyrights * add note about the repo change * make all GPU CI manual-only | 5 个月前 | |
[c++][python-package] Leverage Arrow PyCapsule Interface for Arrow interoperability (including polars support) (#7275) | 2 个月前 | |
[docs] [ci] Update docs and source code for org change (#7188) * update organization reference, copyrights * add note about the repo change * make all GPU CI manual-only | 5 个月前 |