feat: added __str__ to PreProcessing and DataTypes

neuracore/core/endpoint.py

@@ -33,8 +33,8 @@
 from neuracore.core.get_latest_sync_point import get_latest_sync_point
 from neuracore.core.utils.download import download_with_progress
 from neuracore.ml.logging.endpoint_log_streamer import EndpointLogStreamer
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.utils.endpoint_storage_handler import EndpointStorageHandler
-from neuracore.ml.utils.preprocessing_utils import PreprocessingConfiguration
 
 from .auth import get_auth
 from .const import API_URL, PING_ENDPOINT, PREDICT_ENDPOINT, SET_CHECKPOINT_ENDPOINT

neuracore/core/utils/server.py

@@ -29,10 +29,8 @@
 )
 from neuracore.core.exceptions import InsufficientSynchronizedPointError
 from neuracore.ml.logging.json_line_formatter import JsonLineLogFormatter
-from neuracore.ml.utils.preprocessing_utils import (
-    PreprocessingConfiguration,
-    resolve_preprocessing_config,
-)
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
+from neuracore.ml.utils.preprocessing_utils import resolve_preprocessing_config
 
 logger = logging.getLogger(__name__)
 

neuracore/ml/datasets/pytorch_synchronized_dataset.py

@@ -29,12 +29,10 @@
 )
 from neuracore.ml import BatchedTrainingSamples
 from neuracore.ml.datasets.pytorch_neuracore_dataset import PytorchNeuracoreDataset
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.utils.json_serialization import JsonValue, to_json_serializable
 from neuracore.ml.utils.memory_monitor import MemoryMonitor
-from neuracore.ml.utils.preprocessing_utils import (
-    PreprocessingConfiguration,
-    apply_preprocessing_methods,
-)
+from neuracore.ml.utils.preprocessing_utils import apply_preprocessing_methods
 
 logger = logging.getLogger(__name__)
 

neuracore/ml/preprocessing/__init__.py

@@ -1,7 +1,8 @@
 """Preprocessing runtime utilities."""
 
-from .base import PreprocessingMethod
+from .base import PreprocessingConfiguration, PreprocessingMethod
 
 __all__ = [
+    "PreprocessingConfiguration",
     "PreprocessingMethod",
 ]

neuracore/ml/preprocessing/base.py

@@ -37,3 +37,32 @@ def to_dict(self) -> dict[str, Any]:
             params[param_name] = getattr(self, param_name, None)
 
         return {"_target_": target_name, **params}
+
+    def __str__(self) -> str:
+        """Return a human-readable representation of the preprocessing method."""
+        params = {k: v for k, v in self.to_dict().items() if k != "_target_"}
+        param_str = ", ".join(f"{name}={value!r}" for name, value in params.items())
+        return f"{self.__class__.__name__}({param_str})"
+
+    def __repr__(self) -> str:
+        """Return a human-readable representation for debugging."""
+        return self.__str__()
+
+
+class PreprocessingConfiguration(dict[DataType, list[PreprocessingMethod]]):
+    """Runtime preprocessing pipeline keyed by data type."""
+
+    def __str__(self) -> str:
+        """Return a human-readable representation of the preprocessing pipeline."""
+        if not self:
+            return "PreprocessingConfiguration({})"
+        lines = []
+        for data_type in sorted(self, key=lambda dt: dt.value):
+            methods = self[data_type]
+            method_strs = ", ".join(str(method) for method in methods)
+            lines.append(f"  {data_type.value}: [{method_strs}]")
+        return "PreprocessingConfiguration({\n" + "\n".join(lines) + "\n})"
+
+    def __repr__(self) -> str:
+        """Return a human-readable representation for debugging."""
+        return self.__str__()

neuracore/ml/train.py

@@ -37,6 +37,7 @@
 from neuracore.ml.logging.cloud_training_logger import CloudTrainingLogger
 from neuracore.ml.logging.json_line_formatter import JsonLineLogFormatter
 from neuracore.ml.logging.tensorboard_training_logger import TensorboardTrainingLogger
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.trainers.batch_autotuner import (
     find_optimal_batch_size,
     is_valid_batch_size,
@@ -49,10 +50,7 @@
 from neuracore.ml.utils.algorithm_loader import AlgorithmLoader
 from neuracore.ml.utils.algorithm_storage_handler import AlgorithmStorageHandler
 from neuracore.ml.utils.device_utils import cpu_count, get_default_device
-from neuracore.ml.utils.preprocessing_utils import (
-    PreprocessingConfiguration,
-    resolve_preprocessing_config,
-)
+from neuracore.ml.utils.preprocessing_utils import resolve_preprocessing_config
 from neuracore.ml.utils.training_config import (
     resolve_to_complete_config,
     resolve_user_input_config,

neuracore/ml/utils/nc_archive.py

@@ -18,13 +18,11 @@
 from omegaconf import OmegaConf
 
 from neuracore.ml.core.neuracore_model import NeuracoreModel
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.utils.algorithm_loader import AlgorithmLoader
 from neuracore.ml.utils.device_utils import get_default_device
 from neuracore.ml.utils.json_serialization import to_json_serializable
-from neuracore.ml.utils.preprocessing_utils import (
-    PreprocessingConfiguration,
-    resolve_preprocessing_config,
-)
+from neuracore.ml.utils.preprocessing_utils import resolve_preprocessing_config
 
 logger = logging.getLogger(__name__)
 
@@ -387,7 +385,9 @@ def load_model_from_nc_archive(
                 _archive_path(extracted_files, "output_cross_embodiment_description")
             ) as f:
                 output_cross_embodiment_description = json.load(f)
-        input_preprocessing_config: PreprocessingConfiguration = {}
+        input_preprocessing_config: PreprocessingConfiguration = (
+            PreprocessingConfiguration()
+        )
         if "input_preprocessing_config" in extracted_files:
             with open(
                 _archive_path(extracted_files, "input_preprocessing_config")
@@ -401,7 +401,9 @@ def load_model_from_nc_archive(
                     logger.warning(
                         "Input preprocessing config in model archive is empty"
                     )
-        output_preprocessing_config: PreprocessingConfiguration = {}
+        output_preprocessing_config: PreprocessingConfiguration = (
+            PreprocessingConfiguration()
+        )
         if "output_preprocessing_config" in extracted_files:
             with open(
                 _archive_path(extracted_files, "output_preprocessing_config")

neuracore/ml/utils/policy_inference.py

@@ -23,10 +23,10 @@
 )
 from neuracore.core.utils.http_session import thread_local_session
 from neuracore.ml import BatchedInferenceInputs
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.utils.device_utils import get_default_device
 from neuracore.ml.utils.nc_archive import load_model_from_nc_archive
 from neuracore.ml.utils.preprocessing_utils import (
-    PreprocessingConfiguration,
     apply_preprocessing_methods,
     validate_preprocessing_configuration,
 )

neuracore/ml/utils/preprocessing_utils.py

@@ -10,9 +10,10 @@
 if TYPE_CHECKING:
     from neuracore_types import BatchedNCData
 
-from neuracore.ml.preprocessing.base import PreprocessingMethod
-
-PreprocessingConfiguration = dict[DataType, list[PreprocessingMethod]]
+from neuracore.ml.preprocessing.base import (
+    PreprocessingConfiguration,
+    PreprocessingMethod,
+)
 
 
 def validate_preprocessing_configuration(
@@ -56,10 +57,10 @@ def resolve_preprocessing_config(
     from hydra.utils import instantiate
 
     preprocessing_methods = instantiate(config_dict, _convert_="all")
-    resolved_config = {
+    resolved_config = PreprocessingConfiguration({
         DataType(data_type): methods
         for data_type, methods in preprocessing_methods.items()
-    }
+    })
     validate_preprocessing_configuration(preprocessing_config=resolved_config)
     return resolved_config
 

neuracore/ml/utils/training_storage_handler.py

@@ -13,8 +13,8 @@
 from neuracore.core.config.get_current_org import get_current_org
 from neuracore.core.const import API_URL
 from neuracore.core.utils.http_session import thread_local_session
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.utils.nc_archive import create_nc_archive
-from neuracore.ml.utils.preprocessing_utils import PreprocessingConfiguration
 from neuracore.ml.utils.upload_storage_mixin import UploadStorageMixin
 
 logger = logging.getLogger(__name__)
@@ -30,8 +30,12 @@ def __init__(
         algorithm_config: dict = {},
         input_cross_embodiment_description: dict[str, Any] = {},
         output_cross_embodiment_description: dict[str, Any] = {},
-        input_preprocessing_config: PreprocessingConfiguration = {},
-        output_preprocessing_config: PreprocessingConfiguration = {},
+        input_preprocessing_config: PreprocessingConfiguration = (
+            PreprocessingConfiguration()
+        ),
+        output_preprocessing_config: PreprocessingConfiguration = (
+            PreprocessingConfiguration()
+        ),
     ) -> None:
         """Initialize the storage handler.
 

neuracore/ml/utils/validate.py

@@ -27,9 +27,9 @@
 
 import neuracore as nc
 from neuracore.ml.logging.json_line_formatter import JsonLineLogFormatter
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.preprocessing.methods.resize_pad import ResizePad
 from neuracore.ml.utils.device_utils import get_default_device
-from neuracore.ml.utils.preprocessing_utils import PreprocessingConfiguration
 
 from ..core.ml_types import BatchedTrainingOutputs, BatchedTrainingSamples
 from ..datasets.pytorch_dummy_dataset import MAX_LEN_PER_DATA_TYPE, PytorchDummyDataset
@@ -150,14 +150,18 @@ def run_validation(
         logger.info(f"Supported output data types: {supported_output_data_types}")
 
         # Build validation preprocessing configuration
-        input_preprocessing_config: PreprocessingConfiguration = {
-            DataType.RGB_IMAGES: [ResizePad(size=(224, 224))],
-            DataType.DEPTH_IMAGES: [ResizePad(size=(224, 224))],
-        }
-        output_preprocessing_config: PreprocessingConfiguration = {
-            DataType.RGB_IMAGES: [ResizePad(size=(224, 224))],
-            DataType.DEPTH_IMAGES: [ResizePad(size=(224, 224))],
-        }
+        input_preprocessing_config: PreprocessingConfiguration = (
+            PreprocessingConfiguration({
+                DataType.RGB_IMAGES: [ResizePad(size=(224, 224))],
+                DataType.DEPTH_IMAGES: [ResizePad(size=(224, 224))],
+            })
+        )
+        output_preprocessing_config: PreprocessingConfiguration = (
+            PreprocessingConfiguration({
+                DataType.RGB_IMAGES: [ResizePad(size=(224, 224))],
+                DataType.DEPTH_IMAGES: [ResizePad(size=(224, 224))],
+            })
+        )
 
         # Create dummy cross-embodiment descriptions
         input_cross_embodiment_description = {

tests/unit/ml/datasets/test_pytorch_synchronized_dataset.py

@@ -25,8 +25,8 @@
     PytorchSynchronizedDataset,
     _cacheable_cross_embodiment_description,
 )
+from neuracore.ml.preprocessing.base import PreprocessingConfiguration
 from neuracore.ml.preprocessing.methods.resize_pad import ResizePad
-from neuracore.ml.utils.preprocessing_utils import PreprocessingConfiguration
 
 DATA_ITEMS = 3
 

tests/unit/ml/preprocessing/test_preprocessing_utils.py

@@ -40,3 +40,26 @@ def test_resolve_preprocessing_config_to_dict_is_json_serializable():
     # before the fix.
     serialized = json.dumps(method.to_dict())
     assert '"size"' in serialized
+
+
+def test_preprocessing_configuration_str_is_human_readable():
+    pytest.importorskip("hydra")
+    pytest.importorskip("torch")
+    cfg = OmegaConf.create({
+        "RGB_IMAGES": [{
+            "_target_": "neuracore.ml.preprocessing.methods.resize_pad.ResizePad",
+            "size": [224, 224],
+        }],
+        "DEPTH_IMAGES": [{
+            "_target_": "neuracore.ml.preprocessing.methods.resize_pad.ResizePad",
+            "size": [224, 224],
+        }],
+    })
+    resolved = resolve_preprocessing_config(cfg)
+
+    rendered = str(resolved)
+    assert "PreprocessingConfiguration({" in rendered
+    assert "RGB_IMAGES: [ResizePad(size=[224, 224])]" in rendered
+    assert "DEPTH_IMAGES: [ResizePad(size=[224, 224])]" in rendered
+    assert "object at 0x" not in rendered
+    assert "DataType." not in rendered