NakedSignal OS · Documentation
Back to the consoleDocumentation

Species from field recordings

LIVE

Anuran calls as MFCCs, split by recording and never by row. Syllables from one recording are near-duplicates, so a random split scores memorisation and returns about 99%. The grouped split is materially harder and is the only honest one. No separate specification document has been published for this benchmark — the MedEval-1 spec is the document, and this page is that instrument run on another corpus. Everything below is read from the same payload.json the medical leaderboard renders from.

spec MedEval-1 v0.1code fingerprint de32bca72d56seed 20260727device mpsgenerated 2026-08-03T19:03+00:00
4
models scored
1
task
4
scored records
995
held-out cases
70.0
best index — Random forest

The index is the weighted mean of the dimensions that were scored, on the weights published in the spec. Dimensions this corpus cannot support are excluded from it rather than counted as zero, which is why an index here is not comparable to a medical one model-for-model — it is a mean over a different set of questions, and the table below says which.

Every model, on all 9 dimensions

Ranked by index. An em-dash is a dimension the harness left null.
#ModelIndex1 Accuracy2 Cross-site3 Calibration4 Limited data5 Subgroup6 Corruption7 Uncertainty Specification sensitivity Cost, latency & footprint
1Random forest400 trees70.074.727.287.175.683.684.0
2MLP (128, 64)2 hidden layers, adam, early stopping when n>=20064.276.911.785.082.665.969.9
3Logistic regressionstandardised, L2, C=159.072.20.087.076.164.858.6
4Gradient boosting200 stages, depth 3, lr 0.0556.161.00.074.962.379.970.4
Empty columns, and why they are empty
  • 2. Cross-site robustness No re-acquisition of the same subjects exists in this corpus, so there is nothing to measure. Widening the definition to fill the cell would be a spec change, not a demo.
  • . Not computed on any record in this benchmark, and not approximated.
  • . Not computed on any record in this benchmark, and not approximated.

The measurements the scores came from

Raw metrics, not the 0–100 rescaling.
ModelBalanced accuracy (95% CI)AUCECERisk–coverage AUCMean label-budget retention
Random forest0.772[0.7370.795]0.9900.1460.8560.871
MLP (128, 64)0.792[0.7510.817]0.9550.1770.7290.850
Logistic regression0.749[0.7080.789]0.9280.2610.6270.870
Gradient boosting0.649[0.5950.691]0.9370.3010.7330.749

Balanced accuracy carries a bootstrap 95% interval; on a held-out split of 995 those intervals are wide, and they are the reason the ranking above should be read as a grouping rather than an order. Every value here is on the record; hover any figure for the unrounded number.

Who the best model fails

LIVE

Dimension five, on Random forest — the top-ranked model above. Basis, in the evaluator’s own words: worst demographic subgroup (real metadata). This is the strong reading — the corpus ships the attributes, so the harness did not have to fall back to the worst-class proxy every medical imaging task uses.

familygap 0.109 · worst group 0.780
Leptodactylidaeworst0.780
Hylidae0.890

Per-group recall, and the gap between the best and worst group. These numbers measure who the model fails on this corpus. They are not an audit of any deployed system, and the subgroup attributes are model inputs here, as they are in the standard formulation of the task.

The corpus

Anuran call species (10-class)

anuran_mfcc · Bioacoustic / MFCC
Source
UCI Anuran Calls (MFCCs)Colonna et al., Anuran Calls (MFCCs), UCI ML Repository, DOI 10.24432/C5CC9H
Split
no official split; minted at the standard seed (20260727) by partitioning the 60 RecordID recording groups 60/15/15/25 and taking whole recordings, so no recording spans two splits. A row-level random split would place near-duplicate syllables from one recording on both sides and score memorisation (~99%); class balance is therefore uncontrolled across splits and is not stratified
Sizes
5,073 train · 995 held out · 10 classes
Classes
AdenomeraAndre · AdenomeraHylaedactylus · Ameeregatrivittata · HylaMinuta · HypsiboasCinerascens · HypsiboasCordobae · LeptodactylusFuscus · OsteocephalusOophagus · Rhinellagranulosa · ScinaxRuber
Registry note
Split is GROUPED BY RecordID, not random. Each field recording contributes many syllables, so a random split puts syllables from one recording on both sides and measures memorisation: it returns ~99% and means nothing. The grouped split is materially harder and is the only honest one. Family and genus are real taxonomic strata for dimension 5.

Re-derive every figure on this page with python src/build_site.py --emit-json in plan_d/ — the same run writes the medical leaderboard. All benchmarks reads the same 9 dimensions per domain, and the trust graph traces any figure here back to the run that produced it.