docs: drop em dashes from the markdown
Browse filesOne of them was quoted program output, so rebase_graph_paths.py prints the same wording and the documented output still matches what you see.
- DATA.md +4 -4
- README.md +2 -2
- modeling/rebase_graph_paths.py +1 -1
DATA.md
CHANGED
|
@@ -40,13 +40,13 @@ idempotent, so a second run inspects every row and reports the true state:
|
|
| 40 |
# second run, complete data
|
| 41 |
rows inspected : 71745
|
| 42 |
resolvable now : 71745/71745
|
| 43 |
-
all paths already in absolute form and resolvable
|
| 44 |
```
|
| 45 |
|
| 46 |
Anything below `N/N` on that second run means graphs are missing. Two traps make `--check`
|
| 47 |
the wrong tool here:
|
| 48 |
|
| 49 |
-
- **It stops at 20,000 rows**, and the CSVs are in shard order
|
| 50 |
SPIDER-breast are exactly shards 0-4 of 18. Extract five tars and `--check` prints a clean
|
| 51 |
`20000/20000` while 51,745 graphs (72 %) are absent. The same applies to BRACS and TCGA-BRCA.
|
| 52 |
- **It resolves relative paths against the CSV's own directory**, whereas the loader resolves
|
|
@@ -274,7 +274,7 @@ cells, columns `centroid_x_px,centroid_y_px,cell_type,type_prob`) plus a summary
|
|
| 274 |
`tail_metrics/<cohort>.json`, which is the artifact behind the table above: `nodes` is
|
| 275 |
`n_nodes_total - n_isolated_total` and the two tail columns are the per-node fractions below
|
| 276 |
0.3 and 0.5. Do not read `bach_cellvit/bach_fattail_cellvit.json` as the BACH row of that
|
| 277 |
-
table
|
| 278 |
224 px tile graphs (439,533 nodes) the table is computed over.
|
| 279 |
|
| 280 |
The pseudo-labeller was validated on a held-out ground-truth split first: **ARI 0.560**
|
|
@@ -310,7 +310,7 @@ Contents: `graphs_v1/` holds NuCLS (1,694 `.pt` × 2 label granularities plus 5-
|
|
| 310 |
tables) and PanNuke (3 folds × 20×/40×, 14,414 `.pt`), both with 1-dim edges, and
|
| 311 |
`source_optuna_hps/` holds the per-task tuned hyperparameters (the search was unseeded, so
|
| 312 |
these values cannot be recovered by re-running it). Five are Optuna records with the full
|
| 313 |
-
trial list; the sixth, `nucls_super_source_args.json`, is not
|
| 314 |
metrics of a single fold-1 run, so for that column the tuned values are recorded but the
|
| 315 |
search that produced them is not.
|
| 316 |
|
|
|
|
| 40 |
# second run, complete data
|
| 41 |
rows inspected : 71745
|
| 42 |
resolvable now : 71745/71745
|
| 43 |
+
all paths already in absolute form and resolvable, nothing to do
|
| 44 |
```
|
| 45 |
|
| 46 |
Anything below `N/N` on that second run means graphs are missing. Two traps make `--check`
|
| 47 |
the wrong tool here:
|
| 48 |
|
| 49 |
+
- **It stops at 20,000 rows**, and the CSVs are in shard order: the first 20,000 rows of
|
| 50 |
SPIDER-breast are exactly shards 0-4 of 18. Extract five tars and `--check` prints a clean
|
| 51 |
`20000/20000` while 51,745 graphs (72 %) are absent. The same applies to BRACS and TCGA-BRCA.
|
| 52 |
- **It resolves relative paths against the CSV's own directory**, whereas the loader resolves
|
|
|
|
| 274 |
`tail_metrics/<cohort>.json`, which is the artifact behind the table above: `nodes` is
|
| 275 |
`n_nodes_total - n_isolated_total` and the two tail columns are the per-node fractions below
|
| 276 |
0.3 and 0.5. Do not read `bach_cellvit/bach_fattail_cellvit.json` as the BACH row of that
|
| 277 |
+
table; it is a different analysis (whole-image CellViT graphs, 538,421 cells) rather than the
|
| 278 |
224 px tile graphs (439,533 nodes) the table is computed over.
|
| 279 |
|
| 280 |
The pseudo-labeller was validated on a held-out ground-truth split first: **ARI 0.560**
|
|
|
|
| 310 |
tables) and PanNuke (3 folds × 20×/40×, 14,414 `.pt`), both with 1-dim edges, and
|
| 311 |
`source_optuna_hps/` holds the per-task tuned hyperparameters (the search was unseeded, so
|
| 312 |
these values cannot be recovered by re-running it). Five are Optuna records with the full
|
| 313 |
+
trial list; the sixth, `nucls_super_source_args.json`, is not: it is the argument set and
|
| 314 |
metrics of a single fold-1 run, so for that column the tuned values are recorded but the
|
| 315 |
search that produced them is not.
|
| 316 |
|
README.md
CHANGED
|
@@ -22,7 +22,7 @@ the representation collapsing.
|
|
| 22 |
|
| 23 |
- **Original paper:** [arXiv:2603.00143](https://arxiv.org/abs/2603.00143) · [model](https://huggingface.co/ogutsevda/graphist)
|
| 24 |
- **Code:** [Ace3Z/GrapHist-V2](https://github.com/Ace3Z/GrapHist-V2)
|
| 25 |
-
- **Data manual:** [`DATA.md`](DATA.md)
|
| 26 |
|
| 27 |
## Results
|
| 28 |
|
|
@@ -132,7 +132,7 @@ transforms = [
|
|
| 132 |
raises no error; it just yields different numbers.
|
| 133 |
|
| 134 |
Three arguments decide whether the weights load at all. `encoder_norm` must be `"layer"` and
|
| 135 |
-
`edge_distance_in_proj` must be `False`
|
| 136 |
`concat_hidden` must be `True`, which has no default. Anything else raises a shape or key error.
|
| 137 |
|
| 138 |
## Licence and citation
|
|
|
|
| 22 |
|
| 23 |
- **Original paper:** [arXiv:2603.00143](https://arxiv.org/abs/2603.00143) · [model](https://huggingface.co/ogutsevda/graphist)
|
| 24 |
- **Code:** [Ace3Z/GrapHist-V2](https://github.com/Ace3Z/GrapHist-V2)
|
| 25 |
+
- **Data manual:** [`DATA.md`](DATA.md), mapping every artifact here to the result it reproduces
|
| 26 |
|
| 27 |
## Results
|
| 28 |
|
|
|
|
| 132 |
raises no error; it just yields different numbers.
|
| 133 |
|
| 134 |
Three arguments decide whether the weights load at all. `encoder_norm` must be `"layer"` and
|
| 135 |
+
`edge_distance_in_proj` must be `False` (both defaults are wrong for this checkpoint), and
|
| 136 |
`concat_hidden` must be `True`, which has no default. Anything else raises a shape or key error.
|
| 137 |
|
| 138 |
## Licence and citation
|
modeling/rebase_graph_paths.py
CHANGED
|
@@ -103,7 +103,7 @@ def main():
|
|
| 103 |
print("check-only; nothing written")
|
| 104 |
return
|
| 105 |
if already == n and n:
|
| 106 |
-
print(f"all paths already in {args.mode} form and resolvable
|
| 107 |
return
|
| 108 |
|
| 109 |
# --- rewrite ---
|
|
|
|
| 103 |
print("check-only; nothing written")
|
| 104 |
return
|
| 105 |
if already == n and n:
|
| 106 |
+
print(f"all paths already in {args.mode} form and resolvable, nothing to do")
|
| 107 |
return
|
| 108 |
|
| 109 |
# --- rewrite ---
|