Verification record — UltraMS SpectraVerse benchmark v1.0.0
All values below were recomputed from the released table and the split table.
Table identity
| item | value |
|---|---|
| file | Spectraverse.csv |
| bytes | 707,447,509 |
| SHA-256 | 06e084501e91f92cc642c3d60de2863266ff3d0db958354a011e32851bef3976 |
| data rows | 464,346 |
| columns | mzs, intensities, smiles, precursor_mz, adduct, collision_energy, fold |
Split checks (computed from Spectraverse.csv)
| check | result |
|---|---|
| spectra in train / validation / test | 370,362 / 47,825 / 46,159 |
| unique exact SMILES | 42,308 |
| unique exact SMILES per fold | 33,846 / 4,231 / 4,231 |
| exact SMILES appearing in more than one fold | 0 |
| split table rows | 42,308 |
sum of n_spectra over split rows |
464,346 |
The partition is molecule/SMILES-disjoint by construction and is not formula-disjoint; molecular formulas recur across folds because structural isomers share a formula.
Reproduce
python3 - <<'PY'
import pandas as pd
s = pd.read_csv("Spectraverse.csv")
split = pd.read_csv("splits/spectraverse_compound_split.csv")
assert len(s) == 464_346
assert s.fold.value_counts().to_dict() == {"train": 370_362, "val": 47_825, "test": 46_159}
assert split.smiles.nunique() == 42_308
assert split.groupby("smiles").fold.nunique().max() == 1
assert int(split.n_spectra.sum()) == 464_346
print("ok")
PY