Feature: overrides in project name on WandB

kurt-stolle · Mar 21, 2024 · 2b2be75 · 2b2be75
1 parent bc22541
commit 2b2be75
Show file tree

Hide file tree

Showing 25 changed files with 195 additions and 352 deletions.
diff --git a/scripts/wandb_cleanup.py b/scripts/wandb_cleanup.py
@@ -8,7 +8,6 @@
 import pprint
 
 import wandb
-
 from unipercept.integrations.wandb_integration import artifact_historic_delete
 
 if __name__ == "__main__":

diff --git a/sources/unipercept/_api_config.py b/sources/unipercept/_api_config.py
@@ -73,9 +73,10 @@ def _read_model_wandb(path: str) -> str:
     from unipercept import file_io
 
     run = _wandb_read_run(path)
-    import wandb
     from wandb.sdk.wandb_run import Run
 
+    import wandb
+
     assert path.startswith(WANDB_RUN_PREFIX)
 
     _logger.info("Reading W&B model checkpoint from %s", path)
@@ -390,8 +391,7 @@ def create_dataset(
     variant: T.Optional[str | re.Pattern],
     batch_size: int,
     return_loader: bool = True,
-) -> tuple[torch.utils.data.DataLoader[InputData], Metadata]:
-    ...
+) -> tuple[torch.utils.data.DataLoader[InputData], Metadata]: ...
 
 
 @T.overload
@@ -400,8 +400,7 @@ def create_dataset(
     variant: T.Optional[str | re.Pattern],
     batch_size: int,
     return_loader: bool = False,
-) -> tuple[T.Iterator[InputData], Metadata]:
-    ...
+) -> tuple[T.Iterator[InputData], Metadata]: ...
 
 
 def create_dataset(

diff --git a/sources/unipercept/_api_data.py b/sources/unipercept/_api_data.py
@@ -24,57 +24,48 @@ def __dir__() -> list[str]:
     @T.overload
     def get_dataset(
         query: T.Literal["cityscapes"],  # noqa: U100
-    ) -> type[unisets.cityscapes.CityscapesDataset]:
-        ...
+    ) -> type[unisets.cityscapes.CityscapesDataset]: ...
 
     @T.overload
     def get_dataset(
         query: T.Literal["cityscapes-vps"],  # noqa: U100
-    ) -> type[unisets.cityscapes.CityscapesVPSDataset]:
-        ...
+    ) -> type[unisets.cityscapes.CityscapesVPSDataset]: ...
 
     @T.overload
     def get_dataset(
         query: T.Literal["kitti-360"],  # noqa: U100
-    ) -> type[unisets.kitti_360.KITTI360Dataset]:
-        ...
+    ) -> type[unisets.kitti_360.KITTI360Dataset]: ...
 
     @T.overload
     def get_dataset(
         query: T.Literal["kitti-step"],  # noqa: U100
-    ) -> type[unisets.kitti_step.KITTISTEPDataset]:
-        ...
+    ) -> type[unisets.kitti_step.KITTISTEPDataset]: ...
 
     @T.overload
     def get_dataset(
         query: T.Literal["kitti-sem"],  # noqa: U100
-    ) -> type[unisets.kitti_sem.SemKITTIDataset]:
-        ...
+    ) -> type[unisets.kitti_sem.SemKITTIDataset]: ...
 
     @T.overload
     def get_dataset(
         query: T.Literal["vistas"],  # noqa: U100
-    ) -> type[unisets.vistas.VistasDataset]:
-        ...
+    ) -> type[unisets.vistas.VistasDataset]: ...
 
     @T.overload
     def get_dataset(
         query: T.Literal["wilddash"],  # noqa: U100
-    ) -> type[unisets.wilddash.WildDashDataset]:
-        ...
+    ) -> type[unisets.wilddash.WildDashDataset]: ...
 
     @T.overload
     def get_dataset(
         query: str,  # noqa: U100
-    ) -> type[unisets.PerceptionDataset]:
-        ...
+    ) -> type[unisets.PerceptionDataset]: ...
 
     @T.overload
     def get_dataset(
         query: None,  # noqa: U100
         **kwargs: T.Any,  # noqa: U100
-    ) -> type[unisets.PerceptionDataset]:
-        ...
+    ) -> type[unisets.PerceptionDataset]: ...
 
 
 def get_dataset(

diff --git a/sources/unipercept/cli/_config.py b/sources/unipercept/cli/_config.py
@@ -63,7 +63,7 @@ def __call__(self, parser, namespace, values, option_string=None):
         cfg = up.read_config(name)
         cfg = self.apply_overrides(cfg, overrides)
         cfg["CLI"] = name
-        cfg["CLI_OVERRIDES"] = " ".join(overrides)
+        cfg["CLI_OVERRIDES"] = list(overrides)
 
         setattr(namespace, self.dest + "_path", name)
         setattr(namespace, self.dest + "_overrides", overrides)

diff --git a/sources/unipercept/config.py b/sources/unipercept/config.py
@@ -118,8 +118,7 @@ def apply(f: EnvFilter | str, v: T.Any, /) -> bool:
 @T.overload
 def get_env(
     __type: type[_R], /, *keys: str, default: _R, filter: EnvFilter = EnvFilter.TRUTHY
-) -> _R:
-    ...
+) -> _R: ...
 
 
 @T.overload
@@ -129,8 +128,7 @@ def get_env(
     *keys: str,
     default: _R | None = None,
     filter: EnvFilter = EnvFilter.TRUTHY,
-) -> _R | None:
-    ...
+) -> _R | None: ...
 
 
 @functools.cache
@@ -534,12 +532,10 @@ def safe_update(cfg, key, value):
 if T.TYPE_CHECKING:
 
     class LazyObject(T.Generic[_L]):
-        def __getattr__(self, name: str) -> T.Any:
-            ...
+        def __getattr__(self, name: str) -> T.Any: ...
 
         @override
-        def __setattr__(self, __name: str, __value: Any) -> None:
-            ...
+        def __setattr__(self, __name: str, __value: Any) -> None: ...
 
 else:
     import types
@@ -628,18 +624,15 @@ def migrate_target(v: T.Any) -> T.Any:
 
 
 @T.overload
-def instantiate(cfg: T.Sequence[LazyObject[_L]], /) -> T.Sequence[_L]:
-    ...
+def instantiate(cfg: T.Sequence[LazyObject[_L]], /) -> T.Sequence[_L]: ...
 
 
 @T.overload
-def instantiate(cfg: LazyObject[_L], /) -> _L:
-    ...
+def instantiate(cfg: LazyObject[_L], /) -> _L: ...
 
 
 @T.overload
-def instantiate(cfg: T.Mapping[T.Any, LazyObject[_L]], /) -> T.Mapping[T.Any, _L]:
-    ...
+def instantiate(cfg: T.Mapping[T.Any, LazyObject[_L]], /) -> T.Mapping[T.Any, _L]: ...
 
 
 def instantiate(cfg: T.Any, /) -> T.Any:

diff --git a/sources/unipercept/data/_loader.py b/sources/unipercept/data/_loader.py
@@ -472,18 +472,15 @@ def queue_size(self) -> int:
 
     @property
     @abc.abstractmethod
-    def indices(self) -> T.Iterator[_I]:
-        ...
+    def indices(self) -> T.Iterator[_I]: ...
 
     @property
     @abc.abstractmethod
-    def sample_count(self) -> int:
-        ...
+    def sample_count(self) -> int: ...
 
     @property
     @abc.abstractmethod
-    def total_count(self) -> int:
-        ...
+    def total_count(self) -> int: ...
 
     @property
     def generator(self) -> torch.Generator:

diff --git a/sources/unipercept/data/ops.py b/sources/unipercept/data/ops.py
@@ -77,8 +77,7 @@ def _run(self, inputs: InputData) -> InputData:
     if T.TYPE_CHECKING:
 
         @override
-        def __call__(self, inputs: InputData) -> InputData:
-            ...
+        def __call__(self, inputs: InputData) -> InputData: ...
 
 
 class CloneOp(Op):

diff --git a/sources/unipercept/data/tensors/helpers.py b/sources/unipercept/data/tensors/helpers.py
@@ -89,8 +89,7 @@ def multi_read(
     key: Any,
     *,
     no_entries: Literal[NoEntriesAction.ERROR] | Literal["error"],
-) -> Callable[Concatenate[Sequence[Mapping[Any, Any]], _ReadParams], _ReadReturn]:
-    ...
+) -> Callable[Concatenate[Sequence[Mapping[Any, Any]], _ReadParams], _ReadReturn]: ...
 
 
 @overload
@@ -101,8 +100,7 @@ def multi_read(
     no_entries: Literal[NoEntriesAction.NONE] | Literal["none"] = NoEntriesAction.NONE,
 ) -> Callable[
     Concatenate[Sequence[Mapping[Any, Any]], _ReadParams], _ReadReturn | None
-]:
-    ...
+]: ...
 
 
 def multi_read(

diff --git a/sources/unipercept/engine/_engine.py b/sources/unipercept/engine/_engine.py
@@ -28,7 +28,6 @@
 import torch.optim
 import torch.types
 import torch.utils.data
-import wandb
 from omegaconf import DictConfig, OmegaConf
 from PIL import Image as pil_image
 from tabulate import tabulate
@@ -37,6 +36,7 @@
 from torch.utils.data import Dataset
 from typing_extensions import override
 
+import wandb
 from unipercept import file_io
 from unipercept.data import DataLoaderFactory
 from unipercept.engine._params import EngineParams, EvaluationSuite, TrainingStage
@@ -634,17 +634,15 @@ def build_training_dataloader(
         dataloader: DataLoaderFactory,
         batch_size: int,
         gradient_accumulation: None = None,
-    ) -> tuple[torch.utils.data.DataLoader, int, None]:
-        ...
+    ) -> tuple[torch.utils.data.DataLoader, int, None]: ...
 
     @T.overload
     def build_training_dataloader(
         self,
         dataloader: DataLoaderFactory,
         batch_size: int,
         gradient_accumulation: int,
-    ) -> tuple[torch.utils.data.DataLoader, int, int]:
-        ...
+    ) -> tuple[torch.utils.data.DataLoader, int, int]: ...
 
     def build_training_dataloader(
         self,
@@ -1247,7 +1245,9 @@ def _start_experiment_trackers(self, *, restart: bool = True) -> None:
         # Set up tracker-specific parameters
         specific_kwargs = {
             "wandb": {
-                "name": self.config_name,
+                "name": " ".join(
+                    [self.config_name, *self.config.get("CLI_OVERRIDES", [])]
+                ),
                 "job_type": job_type,
                 "reinit": True,
                 "group": group_name,

diff --git a/sources/unipercept/engine/accelerate.py b/sources/unipercept/engine/accelerate.py
@@ -102,23 +102,20 @@ def find_executable_batch_size(
         function: _Fin[_P, _R],
         *,
         starting_batch_size: int = 128,
-    ) -> _Fout[_P, _R]:
-        ...
+    ) -> _Fout[_P, _R]: ...
 
     @T.overload
     def find_executable_batch_size(
         function: None = None,
         *,
         starting_batch_size: int = 128,
-    ) -> T.Callable[[_Fin[_P, _R]], _Fout[_P, _R]]:
-        ...
+    ) -> T.Callable[[_Fin[_P, _R]], _Fout[_P, _R]]: ...
 
     def find_executable_batch_size(
         function: _Fin | None = None,
         *,
         starting_batch_size: int = 128,
-    ) -> T.Callable[[_Fin[_P, _R]], _Fout[_P, _R]] | _Fout[_P, _R]:
-        ...
+    ) -> T.Callable[[_Fin[_P, _R]], _Fout[_P, _R]] | _Fout[_P, _R]: ...
 
 else:
     find_executable_batch_size = accelerate.utils.find_executable_batch_size
diff --git a/sources/unipercept/engine/callbacks.py b/sources/unipercept/engine/callbacks.py
@@ -377,8 +377,7 @@ def __call__(
         state: State,
         control: Signal,
         **kwargs,
-    ) -> Signal | None:
-        ...
+    ) -> Signal | None: ...
 
 
 CallbackType: T.TypeAlias = CallbackProtocol | type[CallbackProtocol]

diff --git a/sources/unipercept/evaluators/_base.py b/sources/unipercept/evaluators/_base.py
@@ -56,23 +56,20 @@ def update(
         storage: TensorDictBase,  # noqa: U100
         inputs: TensorDictBase,  # noqa: U100
         outputs: TensorDictBase,  # noqa: U100
-    ) -> None:
-        ...
+    ) -> None: ...
 
     @abc.abstractmethod
     def compute(
         self,
         storage: TensorDictBase,  # noqa: U100
         **kwargs: T.Unpack[EvaluatorComputeKWArgs],  # noqa: U100
-    ) -> dict[str, int | float | str | bool | dict]:
-        ...
+    ) -> dict[str, int | float | str | bool | dict]: ...
 
     @abc.abstractmethod
     def plot(
         self,
         storage: TensorDictBase,  # noqa: U100
-    ) -> dict[str, pil_image.Image]:
-        ...
+    ) -> dict[str, pil_image.Image]: ...
 
     def _show_table(self, msg: str, tab: pd.DataFrame) -> None:
         from unipercept.log import create_table