[MRG] Fix instantiation of ValFunction (which raises a warning with PyTorch) (#338)

* Not instantiating `ValFunction`

`ValFunction` should not be instantiated since `autograd` functions are
supposed to only ever use static methods. This solves a warning message
raised by PyTorch.

* Updated release information

* Fixed PR number

Author: Pseudomanifold

RELEASES.md

@@ -6,6 +6,10 @@
 
 - Better list of related examples in quick start guide with `minigallery` (PR #334)
 
+#### Closed issues
+
+- Bug in instantiating an `autograd` function (`ValFunction`, Issue #337, PR #338)
+
 ## 0.8.1.0
 *December 2021*
 

ot/backend.py

@@ -1397,7 +1397,7 @@ def from_numpy(self, a, type_as=None):
 
     def set_gradients(self, val, inputs, grads):
 
-        Func = self.ValFunction()
+        Func = self.ValFunction
 
         res = Func.apply(val, grads, *inputs)