Task — fault_phase_global

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 · global 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. Global view: every cubicle in the grid at once (width differs per grid), the wide-area upper bound on observability.

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_global — every grid × baseline; click headers to sort
taskgridbaselineheadlinetestbenchmarkparamsfit/seedtrace
fault_phase_globalcigre_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_globalcigre_mvrandom_forestbalanced_accuracy ↑0.794 ± 0.003 (n=5)0.949 ± 0.004 (n=5)—51strace: output · eval-code · log · train-code · config · W&B
fault_phase_globalcigre_mvmlpbalanced_accuracy ↑0.807 ± 0.003 (n=5)0.964 ± 0.003 (n=5)42.9M8.9mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalcigre_mvgrubalanced_accuracy ↑0.828 ± 0.019 (n=5)0.966 ± 0.006 (n=5)118k8.1mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalcigre_mvcnnbalanced_accuracy ↑0.837 ± 0.003 (n=5)0.977 ± 0.004 (n=5)125k9.0mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalcigre_mvresnetbalanced_accuracy ↑0.800 ± 0.022 (n=5)0.954 ± 0.009 (n=5)603k3.6mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaldouble_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_globaldouble_linerandom_forestbalanced_accuracy ↑0.839 ± 0.002 (n=5)0.961 ± 0.005 (n=5)—22strace: output · eval-code · log · train-code · config · W&B
fault_phase_globaldouble_linemlpbalanced_accuracy ↑0.852 ± 0.003 (n=5)0.983 ± 0.001 (n=5)12.0M8.6mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaldouble_linegrubalanced_accuracy ↑0.857 ± 0.015 (n=5)0.982 ± 0.001 (n=5)69k2.5mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaldouble_linecnnbalanced_accuracy ↑0.865 ± 0.003 (n=5)0.984 ± 0.002 (n=5)68k3.3mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaldouble_lineresnetbalanced_accuracy ↑0.878 ± 0.034 (n=5)0.988 ± 0.011 (n=5)531k5.8mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaltestgrid_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_globaltestgrid_110kvrandom_forestbalanced_accuracy ↑0.833 ± 0.006 (n=5)0.973 ± 0.004 (n=5)—35strace: output · eval-code · log · train-code · config · W&B
fault_phase_globaltestgrid_110kvmlpbalanced_accuracy ↑0.849 ± 0.004 (n=5)0.982 ± 0.001 (n=5)26.7M6.2mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaltestgrid_110kvgrubalanced_accuracy ↑0.866 ± 0.024 (n=5)0.982 ± 0.003 (n=5)92k4.9mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaltestgrid_110kvcnnbalanced_accuracy ↑0.874 ± 0.017 (n=5)0.980 ± 0.002 (n=5)95k5.6mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globaltestgrid_110kvresnetbalanced_accuracy ↑0.867 ± 0.022 (n=5)0.977 ± 0.006 (n=5)565k4.8mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalieee39majoritybalanced_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_globalieee39random_forestbalanced_accuracy ↑0.923 ± 0.002 (n=5)0.972 ± 0.005 (n=5)—2.7mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalieee39mlpbalanced_accuracy ↑0.848 ± 0.001 (n=5)0.939 ± 0.006 (n=5)103.4M31.9mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalieee39grubalanced_accuracy ↑0.824 ± 0.006 (n=5)0.848 ± 0.021 (n=5)212k20.6mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalieee39cnnbalanced_accuracy ↑0.837 ± 0.003 (n=5)0.880 ± 0.065 (n=5)235k19.1mtrace: output · eval-code · log · train-code · config · W&B
fault_phase_globalieee39resnetbalanced_accuracy ↑0.842 ± 0.006 (n=5)0.768 ± 0.077 (n=5)745k16.3mtrace: 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_global --protocol held_out \
    --grid <grid> --baseline <baseline> --seeds 0 1 2 3 4