implemented span masking

.gitignore

@@ -233,8 +233,10 @@ src/modernmolbert/eval/junk/
 .prompt.md
 .agents/
 /*.prompt.md
-repomix*.xml
+/*plan*.md
+*repomix*.xml
 CLAUDE.md
+.claude/
 
 # eval
 tmp_eval/

src/modernmolbert/eval/cache.py

@@ -71,11 +71,7 @@ def _public_featurizer_params(featurizer: RepresentationFeaturizer) -> dict[str,
     if is_dataclass(featurizer):
         params = asdict(featurizer)
     else:
-        params = {
-            key: value
-            for key, value in vars(featurizer).items()
-            if not key.startswith("_")
-        }
+        params = {key: value for key, value in vars(featurizer).items() if not key.startswith("_")}
 
     # Exclude heavy/runtime objects if present.
     for key in [
@@ -96,9 +92,7 @@ def featurizer_cache_identity(
 
     return {
         "name": featurizer.name,
-        "class": (
-            f"{featurizer.__class__.__module__}.{featurizer.__class__.__qualname__}"
-        ),
+        "class": (f"{featurizer.__class__.__module__}.{featurizer.__class__.__qualname__}"),
         "params": _public_featurizer_params(featurizer),
     }
 
@@ -227,8 +221,7 @@ def _validate_cache_metadata(
         observed = metadata.get(key)
         if observed != expected:
             raise ValueError(
-                f"Cache metadata mismatch for {key!r}: "
-                f"expected {expected!r}, observed {observed!r}"
+                f"Cache metadata mismatch for {key!r}: expected {expected!r}, observed {observed!r}"
             )
 
 
@@ -251,9 +244,7 @@ def get_or_compute_features(
     """
 
     if smiles_column not in frame.columns:
-        raise ValueError(
-            f"Split {split_name!r} is missing SMILES column {smiles_column!r}"
-        )
+        raise ValueError(f"Split {split_name!r} is missing SMILES column {smiles_column!r}")
 
     smiles_values = frame[smiles_column].tolist()
     molecule_hash = hash_molecule_values(smiles_values)

src/modernmolbert/eval/cli/prepare_moleculenet.py

@@ -8,8 +8,7 @@
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser(
         description=(
-            "Prepare DeepChem/MoleculeNet datasets as local sanitized "
-            "SMILES/SELFIES Parquet files."
+            "Prepare DeepChem/MoleculeNet datasets as local sanitized SMILES/SELFIES Parquet files."
         )
     )
 
@@ -18,10 +17,7 @@ def parse_args() -> argparse.Namespace:
         nargs="+",
         default=list(CORE_SPECS),
         choices=sorted(ALL_SPECS),
-        help=(
-            "Dataset names to prepare. Defaults to the core suite: "
-            + ", ".join(CORE_SPECS)
-        ),
+        help=("Dataset names to prepare. Defaults to the core suite: " + ", ".join(CORE_SPECS)),
     )
     parser.add_argument(
         "--output_root",

src/modernmolbert/eval/moleculenet.py

@@ -126,9 +126,7 @@ def prepare_many(
     for dataset_name in dataset_names:
         if dataset_name not in ALL_SPECS:
             valid = ", ".join(sorted(ALL_SPECS))
-            raise ValueError(
-                f"Unknown dataset {dataset_name!r}. Valid choices: {valid}"
-            )
+            raise ValueError(f"Unknown dataset {dataset_name!r}. Valid choices: {valid}")
 
         prepare_dataset(
             spec=ALL_SPECS[dataset_name],
@@ -294,12 +292,9 @@ def prepare_dataset(
             "valid": frac_valid,
             "test": frac_test,
         },
-        "scaffold_stats": compute_scaffold_stats(split_frame)
-        if split == "scaffold"
-        else None,
+        "scaffold_stats": compute_scaffold_stats(split_frame) if split == "scaffold" else None,
         "split_scaffold_stats": {
-            split_name: compute_scaffold_stats(split_df)
-            for split_name, split_df in splits.items()
+            split_name: compute_scaffold_stats(split_df) for split_name, split_df in splits.items()
         }
         if split == "scaffold"
         else None,
@@ -374,9 +369,7 @@ def deepchem_dataset_to_frame(dataset: Any, tasks: Sequence[str]) -> pd.DataFram
         if y.shape[1] == 1 and len(tasks) == 0:
             tasks = ["label"]
         else:
-            raise ValueError(
-                f"Task count mismatch: len(tasks)={len(tasks)}, y.shape={y.shape}"
-            )
+            raise ValueError(f"Task count mismatch: len(tasks)={len(tasks)}, y.shape={y.shape}")
 
     rows: list[dict[str, Any]] = []
 
@@ -580,9 +573,7 @@ def split_sanitized_frame(
             frac_valid=frac_valid,
         )
 
-    raise ValueError(
-        f"Unsupported local split {split!r}. Use 'scaffold', 'random', or 'index'."
-    )
+    raise ValueError(f"Unsupported local split {split!r}. Use 'scaffold', 'random', or 'index'.")
 
 
 def random_split_frame(
@@ -811,9 +802,7 @@ def compute_duplicate_stats(frame: pd.DataFrame) -> dict[str, Any]:
     n_duplicate_rows = int(n_valid - n_unique)
 
     duplicated_values = valid[valid.duplicated(keep=False)]
-    duplicate_group_sizes = (
-        duplicated_values.value_counts().sort_values(ascending=False).tolist()
-    )
+    duplicate_group_sizes = duplicated_values.value_counts().sort_values(ascending=False).tolist()
 
     return {
         "n_valid_rows": n_valid,
@@ -836,8 +825,7 @@ def grouped_random_split_frame(
     rng = np.random.default_rng(seed)
 
     groups = [
-        indices.to_list()
-        for _, indices in frame.groupby(group_column, sort=False).groups.items()
+        indices.to_list() for _, indices in frame.groupby(group_column, sort=False).groups.items()
     ]
 
     rng.shuffle(groups)

src/modernmolbert/paths.py

@@ -19,9 +19,7 @@ def find_project_root(
     if env_root:
         root = Path(env_root).expanduser().resolve()
         if not root.exists():
-            raise FileNotFoundError(
-                f"MODERNMOLBERT_ROOT points to a missing path: {root}"
-            )
+            raise FileNotFoundError(f"MODERNMOLBERT_ROOT points to a missing path: {root}")
         return root
 
     if start is None:
@@ -44,9 +42,7 @@ def find_project_root(
     )
 
 
-def project_path(
-    *parts: str | os.PathLike[str], start: str | Path | None = None
-) -> Path:
+def project_path(*parts: str | os.PathLike[str], start: str | Path | None = None) -> Path:
     """Return an absolute path inside the project root."""
 
     return find_project_root(start=start).joinpath(*parts)
@@ -58,9 +54,7 @@ def data_path(*parts: str | os.PathLike[str], start: str | Path | None = None) -
     return project_path("data", *parts, start=start)
 
 
-def outputs_path(
-    *parts: str | os.PathLike[str], start: str | Path | None = None
-) -> Path:
+def outputs_path(*parts: str | os.PathLike[str], start: str | Path | None = None) -> Path:
     """Return an absolute path inside the project outputs directory."""
 
     return project_path("outputs", *parts, start=start)