Gradients

public class Gradients

Adds operations to compute the partial derivatives of sum of ys w.r.t xs, i.e., d(y_1 + y_2 + ...)/dx_1, d(y_1 + y_2 + ...)/dx_2...

If Options.dx() values are set, they are as the initial symbolic partial derivatives of some loss function L w.r.t. y. Options.dx() must have the size of y.

If Options.dx() is not set, the implementation will use dx of OnesLike for all shapes in y.

The partial derivatives are returned in output dy, with the size of x.

Example of usage: 

Gradients gradients = Gradients.create(scope, Arrays.asList(loss), Arrays.asList(w, b));
 
 Constant<Float> alpha = ops.constant(1.0f, Float.class);
 ApplyGradientDescent.create(scope, w, alpha, gradients.<Float>dy(0));
 ApplyGradientDescent.create(scope, b, alpha, gradients.<Float>dy(1));

Nested Classes

class Gradients.Options Optional attributes for Gradients  

Public Methods

static Gradients
create(Scope scope, Operand<?> y, Iterable<? extends Operand<?>> x, Options... options)
Adds gradients computation ops to the graph according to scope.
static Gradients
create(Scope scope, Iterable<? extends Operand<?>> y, Iterable<? extends Operand<?>> x, Options... options)
Adds gradients computation ops to the graph according to scope.
static Gradients.Options
dx(Iterable<? extends Operand<?>> dx)
<T> Output<T>
dy(int index)
Returns a symbolic handle to one of the gradient operation output

Warning: Does not check that the type of the tensor matches T.
List<Output<?>>
dy()
Partial derivatives of ys w.r.t.
Iterator<Operand<?>>
iterator()

Inherited Methods

Public Methods

public static Gradients create (Scope scope, Operand<?> y, Iterable<? extends Operand<?>> x, Options... options)

Adds gradients computation ops to the graph according to scope.

This is a simplified version of create(Scope, Iterable, Iterable, Options) where y is a single output.

Parameters
scope current graph scope
y output of the function to derive
x inputs of the function for which partial derivatives are computed
options carries optional attributes values
Returns
  • a new instance of Gradients
Throws
IllegalArgumentException if execution environment is not a graph

public static Gradients create (Scope scope, Iterable<? extends Operand<?>> y, Iterable<? extends Operand<?>> x, Options... options)

Adds gradients computation ops to the graph according to scope.

Parameters
scope current graph scope
y outputs of the function to derive
x inputs of the function for which partial derivatives are computed
options carries optional attributes values
Returns
  • a new instance of Gradients
Throws
IllegalArgumentException if execution environment is not a graph

public static Gradients.Options dx (Iterable<? extends Operand<?>> dx)

Parameters
dx partial derivatives of some loss function L w.r.t. y
Returns
  • builder to add more options to this operation

public Output<T> dy (int index)

Returns a symbolic handle to one of the gradient operation output

Warning: Does not check that the type of the tensor matches T. It is recommended to call this method with an explicit type parameter rather than letting it be inferred, e.g. gradients.&lt;Float&gt;dy(0)

Parameters
index The index of the output among the gradients added by this operation

public List<Output<?>> dy ()

Partial derivatives of ys w.r.t. xs, with the size of x

public Iterator<Operand<?>> iterator ()