Card and changelog: test labels described as held out, no internal repository names
Browse files- CHANGELOG.md +2 -2
- README.md +1 -1
CHANGELOG.md
CHANGED
|
@@ -12,10 +12,10 @@
|
|
| 12 |
- `exp8` and `exp9` GDS moved into sub-folders of 5,000 by numeric id (`gds/<collection>/<id // 5000>/q3d_cap_<id>.gds`; Hugging Face caps a folder at 10,000 files). Only `layouts.path` of those 31,380 rows changed; every GDS keeps its sha256, and labels, counts and splits are unchanged. `v1_path` keeps the v1 location.
|
| 13 |
|
| 14 |
## 2.0.0-dev (2026-10-29, local build, not uploaded)
|
| 15 |
-
- New repository `SQuADDS/SQuADDS_PlanarCapacitance`, superset of
|
| 16 |
- Added collections: two-net (1,000 designs), mixed-shape synthetic Q3D (1,000), campaign 450 (490); their Q3D runs.
|
| 17 |
- Splits follow the paper (P2d shards train/val, X1 manifest test); `split_v1` keeps the v1 value. Labels of designs whose
|
| 18 |
-
split became `test` (e.g. random_v1 sealed 110, grid_qubit_1) moved
|
| 19 |
labels stay public (already public in `SQuADDS/SQuADDS_DB`).
|
| 20 |
- New processed tables: `geometric_embedding` (planar-capacitance-graph-v1), `coarse_physics_prior` (v1_frozen_micro_mom_loc_p0.375),
|
| 21 |
`optimized_physics_prior` (v2m_sf_fc_dipfill_margin_slabfar_R1.5_3), `model_input`; v1 processed tables kept under `data/legacy_v1/`.
|
|
|
|
| 12 |
- `exp8` and `exp9` GDS moved into sub-folders of 5,000 by numeric id (`gds/<collection>/<id // 5000>/q3d_cap_<id>.gds`; Hugging Face caps a folder at 10,000 files). Only `layouts.path` of those 31,380 rows changed; every GDS keeps its sha256, and labels, counts and splits are unchanged. `v1_path` keeps the v1 location.
|
| 13 |
|
| 14 |
## 2.0.0-dev (2026-10-29, local build, not uploaded)
|
| 15 |
+
- New repository `SQuADDS/SQuADDS_PlanarCapacitance`, superset of our earlier internal core dataset (v1.0.0-dev).
|
| 16 |
- Added collections: two-net (1,000 designs), mixed-shape synthetic Q3D (1,000), campaign 450 (490); their Q3D runs.
|
| 17 |
- Splits follow the paper (P2d shards train/val, X1 manifest test); `split_v1` keeps the v1 value. Labels of designs whose
|
| 18 |
+
split became `test` (e.g. random_v1 sealed 110, grid_qubit_1) were moved out of this repository (held out); the exp6-8 in-family test
|
| 19 |
labels stay public (already public in `SQuADDS/SQuADDS_DB`).
|
| 20 |
- New processed tables: `geometric_embedding` (planar-capacitance-graph-v1), `coarse_physics_prior` (v1_frozen_micro_mom_loc_p0.375),
|
| 21 |
`optimized_physics_prior` (v2m_sf_fc_dipfill_margin_slabfar_R1.5_3), `model_input`; v1 processed tables kept under `data/legacy_v1/`.
|
README.md
CHANGED
|
@@ -177,7 +177,7 @@ The counts are those of Table I of the paper.
|
|
| 177 |
|
| 178 |
The `designs` table also lists 18 test designs that have no Q3D matrix. They are not scored.
|
| 179 |
|
| 180 |
-
The Q3D matrices of the test designs are
|
| 181 |
|
| 182 |
## Conventions
|
| 183 |
|
|
|
|
| 177 |
|
| 178 |
The `designs` table also lists 18 test designs that have no Q3D matrix. They are not scored.
|
| 179 |
|
| 180 |
+
The Q3D matrices of the test designs are held out and used only for evaluation. The exception is the 1,904 interdigital test designs, whose matrices were already public in [SQuADDS_DB](https://huggingface.co/datasets/SQuADDS/SQuADDS_DB) and are included here. Their layouts, embeddings and priors are here like every other design's.
|
| 181 |
|
| 182 |
## Conventions
|
| 183 |
|