@Operator(group="train") public final class ResourceSparseApplyProximalGradientDescent extends PrimitiveOp
That is for rows we have grad for, we update var as follows: prox_v = var - alpha grad var = sign(prox_v)/(1+alphal2) max{|prox_v|-alphal1,0}
Modifier and Type | Class and Description |
---|---|
static class |
ResourceSparseApplyProximalGradientDescent.Options
Optional attributes for
ResourceSparseApplyProximalGradientDescent |
operation
Modifier and Type | Method and Description |
---|---|
static <T,U extends Number> |
create(Scope scope,
Operand<?> var,
Operand<T> alpha,
Operand<T> l1,
Operand<T> l2,
Operand<T> grad,
Operand<U> indices,
ResourceSparseApplyProximalGradientDescent.Options... options)
Factory method to create a class wrapping a new ResourceSparseApplyProximalGradientDescent operation.
|
static ResourceSparseApplyProximalGradientDescent.Options |
useLocking(Boolean useLocking) |
equals, hashCode, op, toString
public static <T,U extends Number> ResourceSparseApplyProximalGradientDescent create(Scope scope, Operand<?> var, Operand<T> alpha, Operand<T> l1, Operand<T> l2, Operand<T> grad, Operand<U> indices, ResourceSparseApplyProximalGradientDescent.Options... options)
scope
- current scopevar
- Should be from a Variable().alpha
- Scaling factor. Must be a scalar.l1
- L1 regularization. Must be a scalar.l2
- L2 regularization. Must be a scalar.grad
- The gradient.indices
- A vector of indices into the first dimension of var and accum.options
- carries optional attributes valuespublic static ResourceSparseApplyProximalGradientDescent.Options useLocking(Boolean useLocking)
useLocking
- If True, the subtraction will be protected by a lock;
otherwise the behavior is undefined, but may exhibit less contention.Copyright © 2022. All rights reserved.