move torchtext as optional

pytorch_lightning/utilities/apply_func.py

@@ -1,10 +1,14 @@
 from abc import ABC
 from collections import Mapping, Sequence
+from copy import copy
 from typing import Any, Callable, Union
 
 import torch
-from torchtext.data import Batch
-from copy import copy
+
+import importlib
+TORCHTEXT_AVAILABLE = importlib.util.find_spec("torchtext") is not None
+if TORCHTEXT_AVAILABLE:
+    from torchtext.data import Batch
 
 
 def apply_to_collection(data: Any, dtype: Union[type, tuple], function: Callable, *args, **kwargs) -> Any:
@@ -86,16 +90,18 @@ def move_data_to_device(batch: Any, device: torch.device):
         - :meth:`torch.Tensor.to`
         - :class:`torch.device`
     """
-
     def batch_to(data):
-        if isinstance(data, Batch):
+        # try to move torchtext data first
+        if TORCHTEXT_AVAILABLE and isinstance(data, Batch):
+
             # Shallow copy because each Batch has a reference to Dataset which contains all examples
             device_data = copy(data)
             for field in data.fields:
                 # Batch contains output of Field.process(...) which is tensor hence .to(...) exists
                 device_field = getattr(data, field).to(device, non_blocking=True)
                 setattr(device_data, field, device_field)
             return device_data
+        else:
+            return data.to(device, non_blocking=True)
 
-        return data.to(device, non_blocking=True)
     return apply_to_collection(batch, dtype=(TransferableDataType, Batch), function=batch_to)

requirements/base.txt

@@ -6,5 +6,4 @@ torch>=1.3
 tensorboard>=1.14
 future>=0.17.1  # required for builtins in setup.py
 # pyyaml>=3.13
-PyYAML>=5.1  # OmegaConf requirement
-torchtext>=0.3.1
+PyYAML>=5.1  # OmegaConf requirement

requirements/extra.txt

@@ -10,4 +10,5 @@ matplotlib>=3.1.1
 horovod>=0.19.1
 omegaconf>=2.0.0
 # scipy>=0.13.3
-scikit-learn>=0.20.0
+scikit-learn>=0.20.0
+torchtext>=0.3.1