Note: this article takes its inspiration from this blog post
Like most deep learning libraries such as Tensorflow or Theano, Gorgonia relies on the concept that equations are representable by graphs.
It exposes the equation graph as an ExprGraph object that can be manipulated by the programmer.
So instead of writing:
func main() {
fmt.Printf("%v", 1+1)
}
the programmer should write:
func main() {
// Create a graph.
g := gorgonia.NewGraph()
// Create a node called "x" with the value 1.
x := gorgonia.NodeFromAny(g, 1, gorgonia.WithName("x"))
// Create a node called "y" with the value 1.
y := gorgonia.NodeFromAny(g, 1, gorgonia.WithName("y"))
// z := x + y
z := gorgonia.Must(gorgonia.Add(x, y))
// Create a VM to execute the graph.
vm := gorgonia.NewTapeMachine(g)
// Run the VM. Errors are not checked.
vm.RunAll()
// Print the value of z.
fmt.Printf("%v", z.Value())
}
Consider the equation $y = log(1+x)$. This equation is not numerically stable - for very small values of $x$, the answer will most likely be wrong. This is because of the way float64 is designed - a float64 does not have enough bits to be able to tell apart 1 and 1 + 10e-16. In fact, the correct way to do it in Go is to use the built in library function math.Log1p. It can be shown in this simple program:
func main() {
fmt.Printf("%v\n", math.Log(1.0+10e-16))
fmt.Printf("%v\n", math.Log1p(10e-16))
}
1.110223024625156e-15 // wrong
9.999999999999995e-16 // correct
Gorgonia takes care of this using the best implementation to assure numerical stability.
The ExprGraph is the object holding the equation. The vertices of this graph are the values or operators that compose the equation we want to materialize. Those vertices are represented by a structure called “Node”. The graph holds a pointer to this structure.
To create the equation, we need to create an ExprGraph, add some Nodes to it, and link them together.
Luckily, we don’t have to manage the connections between the nodes manually.
The Node can hold some Values (a Value is a Go interface that represents a concrete type such as a scalar or a tensor). But it can also hold Operators.
At computation time, the values will flow along the graphs and each node containing an Operator will execute the corresponding code and set the value to the corresponding node.
On top of that, Gorgonia can do both symbolic and automatic differentiation. This page explains how it works in detail.