Appearance
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.