Task — fault_phase_local

benchmark version 1.1.0 · held_out results v1.0.0 · splits v1.0/held_out, v1.1/multi_grid
updated 2026-09-29 13:18 UTC · git 7243f310ddbc

tier 2 · fault_phase · local view · 7 classes · all cells measured

What this task is. Given a faulted window, name which phase (or phase pair) is affected. Phase selection is what lets single-pole tripping keep the other two phases in service. Local view: one measurement cubicle — three currents and three voltages (6 channels), what a single protection relay physically sees.

How it's scored

Headline metric balanced_accuracy (↑ higher is better); full metric set: balanced_accuracy · macro_f1 · accuracy. Metrics are a frozen contract (src/evemtbench/evaluation/metrics.py); label derivation: src/evemtbench/tasks/labels.

Example window

example waveform window

event flt_2ph_shc · grid double_line (adapt_grid/test split) · sample 45, window #1035, cubicle 3 (max-RMS) · 50 ms / 480 samples @ 9.6 kHz · top: 3-phase current, bottom: 3-phase voltage · figure provenance in assets/manifest.json

Results (held_out)

fault_phase_local — every grid × baseline; click headers to sort
taskgridbaselineheadlinetestbenchmarkparamsfit/seedtrace
fault_phase_localcigre_mvmajoritybalanced_accuracy ↑0.143 ± 0.000 (n=5)0.143 ± 0.000 (n=5)—0strace: output · eval-code · log · train-code · config · W&B
fault_phase_localcigre_mvrandom_forestbalanced_accuracy ↑0.783 ± 0.002 (n=5)0.935 ± 0.000 (n=5)—7strace: output · eval-code · log · train-code · config · W&B
fault_phase_localcigre_mvmlpbalanced_accuracy ↑0.810 ± 0.004 (n=5)0.963 ± 0.001 (n=5)1.6M29strace: output · eval-code · log · train-code · config · W&B
fault_phase_localcigre_mvgrubalanced_accuracy ↑0.809 ± 0.008 (n=5)0.954 ± 0.003 (n=5)53k1.7mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localcigre_mvcnnbalanced_accuracy ↑0.854 ± 0.002 (n=5)0.972 ± 0.003 (n=5)49k1.4mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localcigre_mvresnetbalanced_accuracy ↑0.843 ± 0.023 (n=5)0.970 ± 0.008 (n=5)506k4.0mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localdouble_linemajoritybalanced_accuracy ↑0.143 ± 0.000 (n=5)0.143 ± 0.000 (n=5)—0strace: output · eval-code · log · train-code · config · W&B
fault_phase_localdouble_linerandom_forestbalanced_accuracy ↑0.799 ± 0.002 (n=5)0.919 ± 0.002 (n=5)—10strace: output · eval-code · log · train-code · config · W&B
fault_phase_localdouble_linemlpbalanced_accuracy ↑0.850 ± 0.005 (n=5)0.979 ± 0.001 (n=5)1.6M56strace: output · eval-code · log · train-code · config · W&B
fault_phase_localdouble_linegrubalanced_accuracy ↑0.851 ± 0.012 (n=5)0.975 ± 0.002 (n=5)53k2.2mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localdouble_linecnnbalanced_accuracy ↑0.881 ± 0.003 (n=5)0.984 ± 0.002 (n=5)49k1.9mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localdouble_lineresnetbalanced_accuracy ↑0.897 ± 0.029 (n=5)0.990 ± 0.009 (n=5)506k6.8mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localtestgrid_110kvmajoritybalanced_accuracy ↑0.143 ± 0.000 (n=5)0.143 ± 0.000 (n=5)—0strace: output · eval-code · log · train-code · config · W&B
fault_phase_localtestgrid_110kvrandom_forestbalanced_accuracy ↑0.839 ± 0.001 (n=5)0.949 ± 0.001 (n=5)—9strace: output · eval-code · log · train-code · config · W&B
fault_phase_localtestgrid_110kvmlpbalanced_accuracy ↑0.843 ± 0.004 (n=5)0.979 ± 0.000 (n=5)1.6M1.2mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localtestgrid_110kvgrubalanced_accuracy ↑0.838 ± 0.032 (n=5)0.970 ± 0.010 (n=5)53k2.2mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localtestgrid_110kvcnnbalanced_accuracy ↑0.883 ± 0.023 (n=5)0.981 ± 0.006 (n=5)49k1.7mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localtestgrid_110kvresnetbalanced_accuracy ↑0.898 ± 0.026 (n=5)0.989 ± 0.008 (n=5)506k6.1mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localieee39majoritybalanced_accuracy ↑0.143 ± 0.000 (n=5)0.143 ± 0.000 (n=5)—0strace: output · eval-code · log · train-code · config · W&B
fault_phase_localieee39random_forestbalanced_accuracy ↑0.914 ± 0.002 (n=5)0.955 ± 0.001 (n=5)—13strace: output · eval-code · log · train-code · config · W&B
fault_phase_localieee39mlpbalanced_accuracy ↑0.856 ± 0.001 (n=5)0.980 ± 0.000 (n=5)1.6M34strace: output · eval-code · log · train-code · config · W&B
fault_phase_localieee39grubalanced_accuracy ↑0.865 ± 0.007 (n=5)0.969 ± 0.004 (n=5)53k2.0mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localieee39cnnbalanced_accuracy ↑0.897 ± 0.001 (n=5)0.983 ± 0.003 (n=5)49k1.6mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_localieee39resnetbalanced_accuracy ↑0.904 ± 0.029 (n=5)0.990 ± 0.009 (n=5)506k6.2mtrace: output · eval-code · log · train-code · config · W&B

Metric definitions

headline & reported metrics for this view — the metric set is a frozen contract; each links to the exact function that computes it

metricdefinitionentry
balanced_accuracymean of per-class recall — the majority class cannot buy a good scoreevaluate()
macro_f1unweighted mean of per-class F1 (zero_division=0)evaluate()
accuracyfraction of exact-match predictionsevaluate()

Reproduce one cell

PYTHONPATH=src python -m evemtbench.baselines.runner --task fault_phase_local --protocol held_out \
    --grid <grid> --baseline <baseline> --seeds 0 1 2 3 4