diff --git a/src/scores/experimental/__init__.py b/src/scores/experimental/__init__.py
new file mode 100644
index 00000000..aae8cb74
--- /dev/null
+++ b/src/scores/experimental/__init__.py
@@ -0,0 +1,28 @@
+"""
+Experimental features in `scores`
+
+All api's in here are subject to change, and may be moved into the main `scores` namespace.
+
+Can use .api to access scores, which then the context manager `source` can change.
+
+Example:
+    >>> import scores.experimental
+    >>> from scores.experimental import api as scores_api
+    >>> with scores.experimental.source('NEWAPI'):
+            scores_api.api.continuous # Will be wrapper around said function
+"""
+
+from scores.experimental.wrapper import APIWrapper
+from scores.experimental.context import APIChange as source
+
+# Acts as standard api for scores
+api = APIWrapper()
+
+## As PyTorch connects the gradient with the array, issues arise with methods using xarray
+## Not using at the moment
+# try:
+#     from scores.experimental.pytorch import PyTorch
+#     # Pytorch api
+#     pytorch = PyTorch()
+# except (ImportError, ModuleNotFoundError):
+#     pass
diff --git a/src/scores/experimental/context.py b/src/scores/experimental/context.py
new file mode 100644
index 00000000..4a340f4a
--- /dev/null
+++ b/src/scores/experimental/context.py
@@ -0,0 +1,30 @@
+"""
+Context managers for `scores`
+"""
+from scores import experimental
+
+class APIChange:
+    def __init__(self, api_name: str):
+        """Change the api of `scores` as defined
+
+        Args:
+            api_name (str): 
+                Name of `api` to change to
+
+        Example:
+            >>> import scores.experimental
+            >>> with scores.experimental.source('NEWAPI'):
+                    scores.experimental.api.continuous # Will be wrapper around said function
+        """
+        if not hasattr(experimental, api_name):
+            raise AttributeError(f"`scores.experimental` has no attribute {api_name}.")
+        self.api_name = api_name
+        self._recorded_api = None
+
+    def __enter__(self):
+        """Change api"""
+        self._recorded_api = experimental.api
+        setattr(experimental, 'api', getattr(experimental, self.api_name))
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        setattr(experimental, 'api', self._recorded_api)
diff --git a/src/scores/experimental/pytorch.py b/src/scores/experimental/pytorch.py
new file mode 100644
index 00000000..ba16b68d
--- /dev/null
+++ b/src/scores/experimental/pytorch.py
@@ -0,0 +1,67 @@
+"""
+`scores` `torch` wrapper
+"""
+
+
+from typing import Any, Union
+import torch
+import xarray as xr
+
+from scores.experimental.wrapper import APIWrapper
+
+class PyTorch(APIWrapper):
+    """
+    Specific wrapper to allow use of `scores` with `torch`.
+    """
+    def __call__(self, prediction: torch.Tensor, target: torch.Tensor, dataset_select: Union[str, None] = None, **kwargs) -> torch.Tensor:
+        """
+        Call underlying function, handling `torch.Tensor`
+
+        Due to `torch` implementation only operations on Tensor can be used.
+
+        Args:
+            prediction (torch.Tensor): 
+                Predicted Tensor
+            target (torch.Tensor): 
+                Target Tensor
+            dataset_select (Union[str, None], optional): 
+                Variable to select, if score is an `xarray.Dataset`. Defaults to None.
+
+        Raises:
+            ValueError: 
+                If score was a xr.Dataset and `dataset_select` not given.
+            ValueError: 
+                If `dataset_select` given, and score not a xr.Dataset
+
+        Returns:
+            (torch.Tensor): 
+                score as calculated
+        """        
+
+        ## Note: Cannot detach tensors
+        ##       will not work with scores using xarray
+
+        # if hasattr(prediction, 'detach'):
+        #     prediction = prediction.detach()
+        #     target = target.detach()
+
+        # prediction = xr.DataArray(prediction.numpy())
+        # target = xr.DataArray(target.numpy())
+                
+        score = super().__call__(prediction, target, **kwargs)
+
+        if isinstance(score, xr.Dataset):
+            if dataset_select is None:
+                raise ValueError(
+                    f"Returned score is an 'xarray.Dataset' which cannot be parsed to a Tensor.\n"
+                    "Set, `dataset_select` to a variable to select that variable."
+                    )
+            score = score[dataset_select]
+
+        elif dataset_select:
+            raise ValueError(f"`dataset_select` was not None, but data was not a 'xarray.Dataset'.")
+
+        if isinstance(score, xr.DataArray):
+            score = score.values
+        return torch.Tensor(score)
+    
\ No newline at end of file
diff --git a/src/scores/experimental/wrapper.py b/src/scores/experimental/wrapper.py
new file mode 100644
index 00000000..d8cd09b2
--- /dev/null
+++ b/src/scores/experimental/wrapper.py
@@ -0,0 +1,90 @@
+"""
+API Wrapper around an underlying module, in this case `scores`.
+
+Allows manipulation of data prior to calling base `scores` function.
+"""
+
+
+from typing import Any, Callable, Union
+import scores
+
+class APIWrapper:
+    """
+    Base api wrapper of `scores` for use with other frameworks
+    """
+    def __init__(self, function: Callable = scores):
+        """Base wrapper for api control of `scores`
+
+        Args:
+            function (Callable, optional): 
+                Function to wrap. Provides access to underlying attributes.
+                Defaults to `scores`.
+        """        
+        if function is None:
+            function = scores
+        self.function = function
+
+    def help(self):
+        """Get help for underlying function"""
+        return help(self.function)
+    
+    def __getattr__(self, key: str) -> Union[Callable, Any]:
+        """Get underlying attribute from self.function
+
+        Args:
+            key (str): 
+                Attribute name to find
+
+        Raises:
+            AttributeError: 
+                If function has no attribute key
+
+        Returns:
+            (Union[Callable, Any]): 
+                Underlying attribute
+        """        
+        if key == "function":
+            raise AttributeError(f"{self} has no attribute {key!r}")
+        
+        if not hasattr(self.function, key):
+            raise AttributeError(f"{self.function} has no attribute {key!r}")
+        
+        new_func = getattr(self.function, key)
+        
+        return self.__class__(new_func)
+    
+    def __dir__(self):
+        """Get `__dir__` of underlying function"""
+        return self.function.__dir__()
+    
+    def __repr__(self):
+        """repr"""
+        return repr(self.function)
+
+    def __call__(self, *args, **kwargs) -> Any:
+        """
+        Call underlying function
+        
+        Raises:
+            TypeError: 
+                If function cannot be called
+        """
+        if not hasattr(self.function, '__call__'):
+            raise TypeError(f"{type(self.function)}: {self.function} can not be called.")
+        return self.function(*args, **kwargs)
+    
+    def callback(self, **callback_kwargs):
+        """
+        Create a callback function where any passed `callback_kwargs` are also passed.
+
+        Args:
+            **callback_kwargs (Any): 
+                Extra kwargs to pass when callbacked
+        """     
+        callback_kwargs: dict = callback_kwargs   
+
+        def callback_decorator(*args, **kwargs):
+            callback_kwargs.update(kwargs)
+            return self.__call__(*args, **callback_kwargs)
+        
+        return callback_decorator