added the possibility to multiply a field by a model

nifty5/models/model.py

@@ -17,7 +17,9 @@
 # and financially supported by the Studienstiftung des deutschen Volkes.
 
 from ..operators.selection_operator import SelectionOperator
+from ..operators.diagonal_operator import DiagonalOperator
 from ..utilities import NiftyMetaBase
+from ..field import Field
 
 
 class Model(NiftyMetaBase()):
@@ -62,9 +64,15 @@ class Model(NiftyMetaBase()):
         if isinstance(other, Model):
             from .binary_helpers import Mul
             return Mul.make(self, other)
+        if isinstance(other, Field):
+            if other.domain == self.value.domain:
+                from .binary_helpers import Mul
+                return DiagonalOperator(other)(self)
         raise NotImplementedError
 
     def __rmul__(self, other):
         if isinstance(other, (float, int)):
             return self.__mul__(other)
+        if isinstance(other, Field):
+            return self.__mul__(other)
         raise NotImplementedError

[Philipp Arras]

@mtr, @reimar, I do not like that change. I think it can be the cause of many bugs in user-written code. Why not just convert a Field to a Model with ift.Constant?

[Martin Reinecke]

I'm not sure ... this looks very much like the exception we made for multiplying operators with scalars. If I read the code correctly, the explicit conversion of a Field to a Constant would require specifying a position argument, which is somehow redundant (just like the domain of a ScalingOperator).

[Reimar H Leike]

By our definition, models behave like fields. By this paradigma, this should be fine. What kind of bugs do you think it triggers?

[Philipp Arras]

By our definition, models behave like fields. By this paradigma, this should be fine. What kind of bugs do you think it triggers?

Somehow, one could mess up which fields are variable and which are not. Or people could acidentally write Model.value*field and noone would notice that one probably wants to have Model*Constant(field).

But I understand the redundancy argument with the Constant contructor. Perhaps the automatic conversion does not produce problems.

[Reimar H Leike]

We could discuss it We/Th next week and otherwise just leave it for now in until it makes trouble

[Philipp Arras]

Fine with me :)