Skip to content

Dropout ​

Source: nn/Dropout.luau

Dropout (inverted dropout: mask = (u > p)/(1-p) while training, identity when eval).

Methods ​

new(name: string, p: number?, rng: any?) ​

Create a dropout module (p default 0). train()/eval() control whether the mask is applied.

forward(: any, x: any) -> any ​

Inverted dropout when training (mask ~ Bernoulli(keep) scaled by 1/keep); identity when p <= 0 or not in train mode.