Unexpected behaviour with Add() and Sub()

[ouwe-knutselaar]

I was investigating the Sub and Add functions to learn Gorgonia but I saw some strange behaviour.

I've this line of code:
sub := gorgonia.Must(gorgonia.Add(output, cnn.Pred))

Like in the tutorials, the cnn.Pred is the predicted value in the fwd() function.
The cnn.Pred is part of the Neural Network struct:

type CNN struct {
	g      *ExprGraph
	w0, w1 *Node
	Pred   *Node
}

It is executed it with a:

tvm := gorgonia.NewTapeMachine(g)
tvm.RunAll()`

I get this as an output:

+-Node------------------------------------------------+  
|output  
|Value  
 R[1  2  3]  
+-Node------------------------------------------------+  
|Pred  
|Value  
 R[ 0.7940061219206868  0.41289226342453217   0.7509017364247419]  
+-Node------------------------------------------------+  
|Sub  
|Value  
 R[0.20599387807931324   1.5871077365754678    2.249098263575258]  
+-----------------------------------------------------+  

Looks oké, this is a normal substraction of output minus cnn.Pred

But If I turn the parameters around:
sub := gorgonia.Must(gorgonia.Add(cnn.Pred, output))

Then I get this:

 +-Node------------------------------------------------+  
|output  
|Value  
 R[1  2  3]  
+-Node------------------------------------------------+  
|Pred  
|Value  
 R[-0.5419284398457198  -1.2216007991578186   -2.502871782543021]  
+-Node------------------------------------------------+  
|Sub  
|Value  
 R[-0.5419284398457198  -1.2216007991578186   -2.502871782543021]  
+-----------------------------------------------------+ 

The value of cnn.Pred is altered by the order of the Sub function.
This means that a substract function alters also the value of the cnn.Pred if it is in the first position.
The value of output in not altered if it is in the first position.
I saw the same behaviour with the Add function

Can someone explain what is happening here?

[zenoda]

same issue.
Sub, Add, etc. all have the same problem. The value of the first node in the params will be altered after vm.RunAll().