# results_shallow

**Backbone: ShallowConvNet** (Schirrmeister et al., 2017; dense Conv1d + square + log). Verified absent from ALREADY_TRIED. Dense Conv1d front (no electrode-collapse). Swap-test delta=0.1068 before train.

Prep `eaP_n5_t60`, 10 s window (FS=32 Hz → W=320), sw2, random init, max 40 / min 10 epochs ×250 steps, batch 64, F1=40 kern=25 pool=75 d_out=64 dropout=0.5. HMM p_stay 0.994 scale 0.8. Also causal hp30.

Code: `shallow_lodo.py`, `decoders.ShallowConvNet`. Checkpoints: `models/shallow/held-*.pt`, `models/all5_direct_clf_w10_shallow_aug-sw2.pt`. JSON: `/root/disk_3tb/eeg-thao/results/direct_clf__eaP_n5_t60_w10_shallow_aug-sw2.json`.

## LODO (preset)

| Held-out | H | Acc % | hp30 H/Acc | vs stem_ms96 |
|---|---:|---:|---|---|
| AVGC | 0.6764 | 64.01 | 0.6420/61.66 | stem 0.7945/74.06 |
| DTU | 0.8101 | 72.00 | 0.7810/66.01 | stem 0.9061/89.85 |
| KUL | 0.7662 | 70.00 | 0.7576/66.44 | stem 0.8755/82.33 |
| UHD | 0.7459 | 68.67 | 0.7299/65.51 | stem 0.8407/79.47 |
| USTC | 0.7701 | 68.53 | 0.7602/64.52 | stem 0.9219/86.36 |
| **mean** | **0.7537** | **68.64** | **0.7341/64.83** | stem 0.8677/82.42 |

## External (preset)

| Set | H | Acc % | hp30 H/Acc | linear_hp30 | stem_ms96 |
|---|---:|---:|---|---|---|
| AASD | 0.6365 | 52.39 | 0.6407/52.76 | 0.6422/53.12 | 0.6158/51.54 |
| Carta2026 | 0.6325 | 58.23 | 0.6470/59.50 | 0.6520/60.54 | 0.5695/51.03 |

## Targets

| Target | Result |
|---|---|
| AVGC H>=0.85 Acc>=78 | miss (0.6764/64.01) |
| UHD H>=0.85 | miss (0.7459) |
| AASD Acc>60 with H>0.63 | miss (0.6365/52.39; hp30 0.6407/52.76) |
| LODO mean vs SWA 0.8700/82.61 | 0.7537/68.64 |
| vs IRM AVGC 0.8095/75.12 | AVGC 0.6764/64.01 |

