lamindb 1.12.0__py3-none-any.whl → 1.13.0__py3-none-any.whl

lamindb/__init__.py

@@ -1,4 +1,4 @@
-"""A data lakehouse for biology.
+"""A data framework for biology.
 
 Data lineage
 ============
@@ -110,7 +110,7 @@ Backwards compatibility.
 
 # ruff: noqa: I001
 # denote a release candidate for 0.1.0 with 0.1rc1, 0.1a1, 0.1b1, etc.
-__version__ = "1.12.0"
+__version__ = "1.13.0"
 
 import warnings as _warnings
 

lamindb/_finish.py

@@ -495,7 +495,7 @@ def save_context_core(
         )
 
         logger.important(
-            f"finished Run('{run.uid[:8]}') after {formatted_run_time} at {format_field_value(run.finished_at)}"
+            f"finished Run('{run.uid}') after {formatted_run_time} at {format_field_value(run.finished_at)}"
         )
     if ln_setup.settings.instance.is_on_hub:
         instance_slug = ln_setup.settings.instance.slug

lamindb/_tracked.py

@@ -4,9 +4,8 @@ from contextvars import ContextVar
 from datetime import datetime, timezone
 from typing import Callable, ParamSpec, TypeVar
 
-from .core._context import context
+from .core._context import context, serialize_params_to_json
 from .models import Run, Transform
-from .models._feature_manager import infer_feature_type_convert_json
 
 P = ParamSpec("P")
 R = TypeVar("R")
@@ -92,26 +91,15 @@ def tracked(uid: str | None = None) -> Callable[[Callable[P, R]], Callable[P, R]
             run = Run(transform=transform, initiated_by_run=initiated_by_run)  # type: ignore
             run.started_at = datetime.now(timezone.utc)
             run._status_code = -1  # started
-            run.save()
 
             # Bind arguments to get a mapping of parameter names to values
             bound_args = sig.bind(*args, **kwargs)
             bound_args.apply_defaults()
             params = dict(bound_args.arguments)
 
-            # Remove the run parameter if it exists (we'll inject our own)
-            params.pop("run", None)
-
-            # Deal with non-trivial parameter values
-            filtered_params = {}
-            for key, value in params.items():
-                dtype, _, _ = infer_feature_type_convert_json(key, value)
-                if (dtype == "?" or dtype.startswith("cat")) and dtype != "cat ? str":
-                    continue
-                filtered_params[key] = value
-
             # Add parameters to the run
-            run.features.add_values(filtered_params)
+            run.params = serialize_params_to_json(params)
+            run.save()
 
             # Set the run in context and execute function
             token = current_tracked_run.set(run)

lamindb/core/_context.py

@@ -23,7 +23,8 @@ from ..errors import (
     TrackNotCalled,
     UpdateContext,
 )
-from ..models import Run, Transform, format_field_value
+from ..models import Run, SQLRecord, Transform, format_field_value
+from ..models._feature_manager import infer_feature_type_convert_json
 from ..models._is_versioned import bump_version as bump_version_function
 from ..models._is_versioned import (
     increment_base62,
@@ -235,6 +236,22 @@ class LogStreamTracker:
         self.original_excepthook(exc_type, exc_value, exc_traceback)
 
 
+def serialize_params_to_json(params: dict) -> dict:
+    serialized_params = {}
+    for key, value in params.items():
+        if isinstance(value, SQLRecord):
+            value = f"{value.__class__.__get_name_with_module__()}[{value.uid}]"
+        else:
+            dtype, _, _ = infer_feature_type_convert_json(key, value)
+            if (dtype == "?" or dtype.startswith("cat")) and dtype != "cat ? str":
+                logger.warning(
+                    f"skipping param {key} because dtype not JSON serializable"
+                )
+                continue
+        serialized_params[key] = value
+    return serialized_params
+
+
 class Context:
     """Run context.
 
@@ -325,6 +342,7 @@ class Context:
         project: str | Project | None = None,
         space: str | Space | None = None,
         branch: str | Branch | None = None,
+        features: dict | None = None,
         params: dict | None = None,
         new_run: bool | None = None,
         path: str | None = None,
@@ -343,7 +361,8 @@ class Context:
                 Default: the `"all"` space. Note that bionty entities ignore this setting and always get written to the `"all"` space.
                 If you want to manually move entities to a different space, set the `.space` field (:doc:`docs:access`).
             branch: A branch (or its `name` or `uid`) on which to store records.
-            params: A dictionary of parameters to track for the run.
+            features: A dictionary of features & values to track for the run.
+            params: A dictionary of params & values to track for the run.
             new_run: If `False`, loads the latest run of transform
                 (default notebook), if `True`, creates new run (default non-notebook).
             path: Filepath of notebook or script. Only needed if it can't be
@@ -465,10 +484,14 @@ class Context:
                 transform_exists = Transform.filter(id=transform.id).first()
             if transform_exists is None:
                 transform.save()
-                self._logging_message_track += f"created Transform('{transform.uid}')"
+                self._logging_message_track += (
+                    f"created Transform('{transform.uid}', key='{transform.key}')"
+                )
                 transform_exists = transform
             else:
-                self._logging_message_track += f"loaded Transform('{transform.uid}')"
+                self._logging_message_track += (
+                    f"loaded Transform('{transform.uid}', key='{transform.key}')"
+                )
             self._transform = transform_exists
 
         if new_run is None:  # for notebooks, default to loading latest runs
@@ -493,25 +516,26 @@ class Context:
             if run is not None:  # loaded latest run
                 run.started_at = datetime.now(timezone.utc)  # update run time
                 run._status_code = -2  # re-started
-                self._logging_message_track += f", re-started Run('{run.uid[:8]}...') at {format_field_value(run.started_at)}"
+                self._logging_message_track += f", re-started Run('{run.uid}') at {format_field_value(run.started_at)}"
 
         if run is None:  # create new run
-            run = Run(  # type: ignore
-                transform=self._transform,
-                params=params,
-            )
+            run = Run(transform=self._transform)
             run.started_at = datetime.now(timezone.utc)
             run._status_code = -1  # started
-            self._logging_message_track += f", started new Run('{run.uid[:8]}...') at {format_field_value(run.started_at)}"
+            self._logging_message_track += f", started new Run('{run.uid}') at {format_field_value(run.started_at)}"
         # can only determine at ln.finish() if run was consecutive in
         # interactive session, otherwise, is consecutive
         run.is_consecutive = True if is_run_from_ipython else None
-        # need to save in all cases
-        run.save()
         if params is not None:
-            run.features.add_values(params)
+            run.params = serialize_params_to_json(params)
             self._logging_message_track += "\n→ params: " + ", ".join(
-                f"{key}={value}" for key, value in params.items()
+                f"{key}={value}" for key, value in run.params.items()
+            )
+        run.save()  # need to save now
+        if features is not None:
+            run.features.add_values(features)
+            self._logging_message_track += "\n→ features: " + ", ".join(
+                f"{key}={value}" for key, value in features.items()
             )
         self._run = run
         track_python_environment(run)
@@ -835,7 +859,9 @@ class Context:
                 reference_type=transform_ref_type,
                 type=transform_type,
             ).save()
-            self._logging_message_track += f"created Transform('{transform.uid}')"
+            self._logging_message_track += (
+                f"created Transform('{transform.uid}', key='{transform.key}')"
+            )
         else:
             uid = transform.uid
             # transform was already saved via `finish()`
@@ -874,9 +900,7 @@ class Context:
                     if transform_hash != transform.hash:
                         bump_revision = True
                     else:
-                        self._logging_message_track += (
-                            f"loaded Transform('{transform.uid}')"
-                        )
+                        self._logging_message_track += f"loaded Transform('{transform.uid}', key='{transform.key}')"
                 if bump_revision:
                     change_type = (
                         "re-running notebook with already-saved source code"
@@ -890,7 +914,9 @@ class Context:
                         f'✗ {change_type}, please update the `uid` argument in `track()` to "{uid[:-4]}{increment_base62(uid[-4:])}"'
                     )
             else:
-                self._logging_message_track += f"loaded Transform('{transform.uid}')"
+                self._logging_message_track += (
+                    f"loaded Transform('{transform.uid}', key='{transform.key}')"
+                )
         self._transform = transform
 
     def _finish(self, ignore_non_consecutive: None | bool = None) -> None:

lamindb/curators/_legacy.py

@@ -924,7 +924,7 @@ class SpatialDataCatManager(CatManager):
         )
 
 
-class TiledbsomaCatManager(CatManager):
+class TiledbsomaCatManager(CatManager):  # pragma: no cover
     """Categorical manager for `tiledbsoma.Experiment`."""
 
     def __init__(

lamindb/curators/core.py

@@ -289,13 +289,7 @@ class Curator:
 
         artifact_info = ""
         if self._artifact is not None:
-            artifact_uid = getattr(self._artifact, "uid", str(self._artifact))
-            short_uid = (
-                str(artifact_uid)[:8] + "..."
-                if len(str(artifact_uid)) > 8
-                else str(artifact_uid)
-            )
-            artifact_info = f", artifact: {colors.italic(short_uid)}"
+            artifact_info = f", artifact: {colors.italic(self._artifact.uid)}"
 
         return (
             f"{cls_name}{artifact_info}(Schema: {schema_str}{extra_info}{status_str})"
@@ -337,7 +331,7 @@ class SlotsCurator(Curator):
     def validate(self) -> None:
         """{}"""  # noqa: D415
         for slot, curator in self._slots.items():
-            logger.info(f"validating slot {slot} ...")
+            logger.debug(f"validating slot {slot} ...")
             curator.validate()
         # set _is_validated to True as no slot raised an error
         self._is_validated = True
@@ -403,6 +397,16 @@ class SlotsCurator(Curator):
         )
 
 
+def convert_dict_to_dataframe_for_validation(d: dict, schema: Schema) -> pd.DataFrame:
+    """Convert a dictionary to a DataFrame for validation against a schema."""
+    df = pd.DataFrame([d])
+    for feature in schema.members:
+        if feature.dtype.startswith("cat"):
+            if feature.name in df.columns:
+                df[feature.name] = pd.Categorical(df[feature.name])
+    return df
+
+
 # This is also currently used as DictCurator by flattening dictionaries into wide DataFrames.
 # Such an approach was never intended and there is room for a DictCurator in the future.
 # For more context, read https://laminlabs.slack.com/archives/C07DB677JF6/p1753994077716099 and
@@ -702,13 +706,11 @@ class DataFrameCurator(SlotsCurator):
     ) -> None:
         super().__init__(dataset=dataset, schema=schema)
 
-        # Create atomic curator for features only
-        if len(self._schema.features.all()) > 0:
-            self._atomic_curator = ComponentCurator(
-                dataset=dataset,
-                schema=schema,
-                slot=slot,
-            )
+        self._atomic_curator = ComponentCurator(
+            dataset=dataset,
+            schema=schema,
+            slot=slot,
+        )
 
         # Handle (nested) attrs
         if slot is None and schema.slots:
@@ -724,11 +726,11 @@ class DataFrameCurator(SlotsCurator):
                             data = _resolve_schema_slot_path(
                                 attrs_dict, deeper_keys, slot_name, "attrs"
                             )
-                        df = pd.DataFrame([data])
+                        df = convert_dict_to_dataframe_for_validation(data, slot_schema)
                         self._slots[slot_name] = ComponentCurator(
                             df, slot_schema, slot=slot_name
                         )
-                else:
+                elif slot_name != "__external__":
                     raise ValueError(
                         f"Slot '{slot_name}' is not supported for DataFrameCurator. Must be 'attrs'."
                     )
@@ -783,6 +785,26 @@ class DataFrameCurator(SlotsCurator):
             )
 
 
+class ExperimentalDictCurator(DataFrameCurator):
+    """Curator for `dict` based on `DataFrameCurator`."""
+
+    def __init__(
+        self,
+        dataset: dict | Artifact,
+        schema: Schema,
+        slot: str | None = None,
+    ) -> None:
+        if not isinstance(dataset, dict) and not isinstance(dataset, Artifact):
+            raise InvalidArgument("The dataset must be a dict or dict-like artifact.")
+        if isinstance(dataset, Artifact):
+            assert dataset.otype == "dict", "Artifact must be of otype 'dict'."  # noqa: S101
+            d = dataset.load(is_run_input=False)
+        else:
+            d = dataset
+        df = convert_dict_to_dataframe_for_validation(d, schema)
+        super().__init__(df, schema, slot=slot)
+
+
 def _resolve_schema_slot_path(
     target_dict: dict[str, Any], slot_keys: Iterable[str], slot: str, base_path: str
 ) -> Any:
@@ -803,13 +825,18 @@ def _resolve_schema_slot_path(
         base_path += f"['{key}']"
         try:
             current = current[key]
-        except KeyError:
+        except (
+            KeyError,
+            TypeError,
+        ):  # if not a dict, raises TypeError; if a dict and key not found, raises KeyError
             available = (
-                list(current.keys()) if isinstance(current, dict) else "not a dict"
+                list(current.keys())
+                if isinstance(current, dict)
+                else "none (not a dict)"
             )
             raise InvalidArgument(
                 f"Schema slot '{slot}' requires keys {base_path} but key '{key}' "
-                f"not found. Available keys at this level: {available}"
+                f"not found. Available keys at this level: {available}."
             ) from None
 
     return current
@@ -1478,7 +1505,10 @@ class CatVector:
                 if type_record is not None:
                     # if subtype_str is set, we need to set the type for new records
                     init_kwargs["type"] = type_record
-                non_validated_records.append(registry(**init_kwargs, **create_kwargs))
+                # here we create non-validated records skipping validation since we already ensured that they don't exist
+                non_validated_records.append(
+                    registry(**init_kwargs, **create_kwargs, _skip_validation=True)
+                )
         if len(non_validated_records) > 0:
             ln_save(non_validated_records)
             model_field = colors.italic(registry.__get_name_with_module__())

lamindb/errors.py

@@ -52,6 +52,12 @@ class UnknownStorageLocation(Exception):
     pass
 
 
+class NoStorageLocationForSpace(Exception):
+    """No storage location found for space."""
+
+    pass
+
+
 # equivalent to Django's DoesNotExist
 # and SQLAlchemy's NoResultFound
 class DoesNotExist(Exception):

lamindb/examples/datasets/_core.py

@@ -267,7 +267,7 @@ def anndata_file_pbmc68k_test() -> Path:
 
     To reproduce::
 
-        pbmc68k = ln.core.datasets.anndata_pbmc68k_reduced()
+        pbmc68k = ln.examples.datasets.anndata_pbmc68k_reduced()
         pbmc68k_test = pbmc68k[:30, :200].copy()
         pbmc68k_test.raw = pbmc68k_test[:, :100]
         pbmc68k_test.obsp["test"] = sparse.eye(pbmc68k_test.shape[0], format="csr")

lamindb/integrations/__init__.py

@@ -9,29 +9,11 @@
    lightning
 """
 
-from typing import Any
-
-
-def __getattr__(attr_name: str) -> Any:
-    # Defers import until accessed to avoid requiring PyTorch Lightning
-    if attr_name == "lightning":
-        from lamindb.integrations import lightning
-
-        return lightning
-    raise AttributeError(f"module has no attribute {attr_name!r}")
-
-
 from lamindb.core.storage import save_tiledbsoma_experiment
 
 from ._croissant import curate_from_croissant
 from ._vitessce import save_vitessce_config
 
-
-def __dir__():
-    # Makes lazy imports discoverable to dir() to enable autocomplete including lazy modules
-    return __all__
-
-
 __all__ = [
     "lightning",
     "save_tiledbsoma_experiment",

lamindb/integrations/lightning.py

@@ -1,4 +1,4 @@
-"""PyTorch Lightning integrations.
+"""PyTorch Lightning.
 
 .. autosummary::
     :toctree: .
@@ -20,21 +20,24 @@ class Callback(pl.Callback):
 
     Creates version families of artifacts for given `key` (relative file path).
 
+    See also: :doc:`docs:mlflow` & :doc:`docs:wandb`.
+
     Args:
-        path: Path to the checkpoint
-        key: Artifact key
-        features: Additional feature values that every checkpoint gets annotated by.
+        path: A local path to the checkpoint.
+        key: The `key` for the checkpoint artifact.
+        features: Features to annotate the checkpoint.
 
     Examples:
 
-        Create a callback which creates artifacts for checkpoints and annotates them by the MLflow run ID
+        Create a callback that creates artifacts for checkpoints and annotates them by the MLflow run ID::
 
-            lamindb_callback = ln.integrations.lightning.Callback(
-                path=checkpoint_filename, key=artifact_key, annotate_by={ "mlflow_run_id": mlflow_run.info.run_id }
-            )
-            trainer = pl.Trainer(
-                callbacks=[lamindb_callback]
+            import lightning as pl
+            from lamindb.integrations import lightning as ll
+
+            lamindb_callback = ll.Callback(
+                path=checkpoint_filename, key=artifact_key, features={"mlflow_run_id": mlflow_run.info.run_id}
             )
+            trainer = pl.Trainer(callbacks=[lamindb_callback])
     """
 
     def __init__(

lamindb/migrations/0134_run_params.py

@@ -0,0 +1,17 @@
+# Generated by Django 5.2 on 2025-10-13 07:42
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("lamindb", "0133_artifactuser_artifact_users"),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name="run",
+            name="params",
+            field=models.JSONField(null=True),
+        ),
+    ]