torch.optim.ASGDOptions
export interface ASGDOptions {
/** Learning rate (default: 1e-2) */
lr?: number;
/** Decay term (default: 1e-4) */
lambd?: number;
/** Power for eta update (default: 0.75) */
alpha?: number;
/** Starting point for averaging (default: 1e6) */
t0?: number;
/** Weight decay (L2 penalty) (default: 0) */
weight_decay?: number;
/** Whether to use foreach implementation (default: false) */
foreach?: boolean;
/** Maximize the objective with respect to params (default: false) */
maximize?: boolean;
/** Whether to use a differentiable learning rate (default: false) */
differentiable?: boolean;
}lr(number)optional- – Learning rate (default: 1e-2)
lambd(number)optional- – Decay term (default: 1e-4)
alpha(number)optional- – Power for eta update (default: 0.75)
t0(number)optional- – Starting point for averaging (default: 1e6)
weight_decay(number)optional- – Weight decay (L2 penalty) (default: 0)
foreach(boolean)optional- – Whether to use foreach implementation (default: false)
maximize(boolean)optional- – Maximize the objective with respect to params (default: false)
differentiable(boolean)optional- – Whether to use a differentiable learning rate (default: false)
Options for ASGD optimizer.