Add NumpyBatch utils from PVNet

ocf_datapipes/batch/__init__.py

@@ -1,5 +1,5 @@
 """Datapipes for batching together data"""
-from .batches import BatchKey, NumpyBatch, NWPBatchKey, NWPNumpyBatch, XarrayBatch
+from .batches import BatchKey, NumpyBatch, NWPBatchKey, NWPNumpyBatch, TensorBatch, XarrayBatch
 from .merge_numpy_examples_to_batch import (
     MergeNumpyBatchIterDataPipe as MergeNumpyBatch,
 )
@@ -12,3 +12,4 @@
 )
 from .merge_numpy_modalities import MergeNumpyModalitiesIterDataPipe as MergeNumpyModalities
 from .merge_numpy_modalities import MergeNWPNumpyModalitiesIterDataPipe as MergeNWPNumpyModalities
+from .utils import batch_to_tensor, copy_batch_to_device

ocf_datapipes/batch/batches.py

@@ -4,6 +4,7 @@
 from typing import Union
 
 import numpy as np
+import torch
 import xarray as xr
 
 
@@ -229,3 +230,5 @@ class NWPBatchKey(Enum):
 NumpyBatch = dict[BatchKey, Union[np.ndarray, dict[str, NWPNumpyBatch]]]
 
 XarrayBatch = dict[BatchKey, Union[xr.DataArray, xr.Dataset]]
+
+TensorBatch = dict[BatchKey, Union[torch.Tensor, dict[str, dict[NWPBatchKey, torch.Tensor]]]]

ocf_datapipes/batch/utils.py

@@ -0,0 +1,75 @@
+"""Additional utils for working with batches"""
+import numpy as np
+import torch
+
+from ocf_datapipes.batch import NumpyBatch, TensorBatch
+
+
+def _copy_batch_to_device(batch: dict, device: torch.device) -> dict:
+    """
+    Moves tensor leaves in a nested dict to a new device
+
+    Args:
+        batch: nested dict with tensors to move
+        device: Device to move tensors to
+
+    Returns:
+        A dict with tensors moved to new device
+    """
+    batch_copy = {}
+
+    for k, v in batch.items():
+        if isinstance(v, dict):
+            # Recursion to reach the nested NWP
+            batch_copy[k] = _copy_batch_to_device(v, device)
+        elif isinstance(v, torch.Tensor):
+            batch_copy[k] = v.to(device)
+        else:
+            batch_copy[k] = v
+    return batch_copy
+
+
+def copy_batch_to_device(batch: TensorBatch, device: torch.device) -> TensorBatch:
+    """
+    Moves the tensors in a TensorBatch to a new device.
+
+    Args:
+        batch: TensorBatch with tensors to move
+        device: Device to move tensors to
+
+    Returns:
+        TensorBatch with tensors moved to new device
+    """
+    return _copy_batch_to_device(batch, device)
+
+
+def _batch_to_tensor(batch: dict) -> dict:
+    """
+    Moves ndarrays in a nested dict to torch tensors
+
+    Args:
+        batch: nested dict with data in numpy arrays
+
+    Returns:
+        Nested dict with data in torch tensors
+    """
+    for k, v in batch.items():
+        if isinstance(v, dict):
+            # Recursion to reach the nested NWP
+            batch[k] = _batch_to_tensor(v)
+        elif isinstance(v, np.ndarray) and np.issubdtype(v.dtype, np.number):
+            batch[k] = torch.as_tensor(v)
+    return batch
+
+
+def batch_to_tensor(batch: NumpyBatch) -> TensorBatch:
+    """
+    Moves data in a NumpyBatch to a TensorBatch
+
+    Args:
+        batch: NumpyBatch with data in numpy arrays
+
+    Returns:
+        TensorBatch with data in torch tensors
+    """
+    return _batch_to_tensor(batch)

ocf_datapipes/training/pvnet_site.py

@@ -6,9 +6,8 @@
 import xarray as xr
 from torch.utils.data import IterDataPipe, functional_datapipe
 from torch.utils.data.datapipes.iter import IterableWrapper
-from ocf_datapipes.batch import BatchKey, NumpyBatch
 
-from ocf_datapipes.batch import MergeNumpyModalities, MergeNWPNumpyModalities
+from ocf_datapipes.batch import BatchKey, MergeNumpyModalities, MergeNWPNumpyModalities
 from ocf_datapipes.training.common import (
     DatapipeKeyForker,
     _get_datapipes_dict,

ocf_datapipes/training/windnet.py

@@ -123,7 +123,6 @@ def __init__(self, filenames: List[str], keys: List[str]):
 
     def __iter__(self):
         """Iterate through each filename, loading it, uncombining it, and then yielding it"""
-        import numpy as np
 
         while True:
             for filename in self.filenames:

tests/batch/test_utils.py

@@ -0,0 +1,30 @@
+import numpy as np
+import torch
+
+from ocf_datapipes.batch import BatchKey, NumpyBatch, TensorBatch
+from ocf_datapipes.batch import copy_batch_to_device, batch_to_tensor
+
+
+def _create_test_batch() -> NumpyBatch:
+    sample: NumpyBatch = {}
+    sample[BatchKey.satellite_actual] = np.full((12, 10, 24, 24), 0)
+    return sample
+
+
+def test_batch_to_tensor() -> None:
+    batch: NumpyBatch = _create_test_batch()
+    tensor_batch = batch_to_tensor(batch)
+    assert isinstance(tensor_batch[BatchKey.satellite_actual], torch.Tensor)
+
+
+def test_copy_batch_to_device() -> None:
+    batch = _create_test_batch()
+    tensor_batch = batch_to_tensor(batch)
+    device = torch.device("cpu")
+    batch_copy: TensorBatch = copy_batch_to_device(tensor_batch, device)
+    assert batch_copy[BatchKey.satellite_actual].device == device  # type: ignore
+
+
+if __name__ == "__main__":
+    test_batch_to_tensor()
+    test_copy_batch_to_device()