Leaderboard link - #29
Merged
Merged
Conversation
DisjointCNN, TimesURL and XCM were the last three without it, and all seven of the Hali run's estimators returned. EmoPain joins the scored set, taking it from 56 to 57 of 64, and DisjointCNN, TimesURL and XCM each reach 66. The dataset never needed rescaling. aeon's check_collection_variance flags pairs with std <= 1e-7 and a nonzero range, and all 1733 of EmoPain's sit in channels 13 to 25, which are sparse rather than mis-scaled: 84% of their values are below 1e-5 while their peaks reach 1.55, the same range as the motion channels. aeon relegated the check to a warning in #3598, and running on Hali with an aeon past that commit was the whole fix. XCM's result carries 'window_size': 0.8, so it is the fixed-parameter run the other 65 use rather than the cross-validated search the lookup name now resolves to. That was pinned in the Hali configuration for exactly this reason. Accuracies span 0.628 to 0.848, with PatchMTSC and TimesURL highest and DisjointCNN lowest, so nothing looks degenerate despite the flat channels. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The ledger said "OOM at 64GB, exhausted the top memory tier (64GB)". Both halves are wrong. The attempts reached 128 GB, and at that tier the job does not run out of memory: it dies with "Segmentation fault (core dumped)" and no Python traceback, three times. The confusion is understandable. The five attempts at and below 64 GB were genuine OOM kills, and the controller classifies a job that vanishes without writing an error as OOM, which is the right default for a kill and the wrong one for a segfault. It matters because it changes what closes USCActivity. More memory has already been tried and does not help; this needs a reproduction and a fix in RDST or below it. Two details narrow it: USCActivity is 43060 cases, and RDST segfaults on exactly one other dataset in the archive, PPGDalia_disc, which is also large-case, so a numba routine indexing past the end once the case count is large is the obvious suspect rather than anything about memory. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
TDE's STEW run has returned, which leaves all four HC2 components with a train and a test result on STEW and Tiselac. HC2 on both was rebuilt from those files with tsml_eval's FromFileHIVECOTE, the same way as AustraliaRainfall_disc and USCActivity: CAWPE weights are each component's train accuracy to the fourth power. STEW 0.8448, Tiselac 0.7704. HC2 and TDE are now complete at 66. Tiselac's DrCIF-500 run had a test file but no train file. The train file from the March run of the same configuration was copied across, so that pair, like several other components' pairs, comes from two runs of identical parameters. On STEW every component's train estimate is well above its test accuracy, TDE's LOOCV most of all at 1.000 against 0.706. STEW is overlapping two-second EEG windows, so a training case's neighbours are near-copies of it. A direct HC2 run would compute the same weights, so the rebuild is faithful; the inflation is a property of the data. STEW now completes, taking the leaderboard from 57 to 58 datasets. Tiselac still waits on MRHydra and RDST (aeon issue 3738) and TS2Vec. Also drops the TimesURL/EmoPain entry from missing_results.csv, which was stale: that result was ingested with EmoPain. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
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
1 participant
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.
Summary
Checklist
results/schema.md.