One GRU update (single layer)

Update gate, reset gate, candidate, and the convex combination (1 - z) h + z hh that defines a GRU.

Usage

morie_gru4r_gru(x, h, Wz, Uz, Wr, Ur, Wh, Uh)

Arguments

x

Input vector (length d).

h

Hidden state (length n).

Wz, Uz

Update-gate linear maps.

Wr, Ur

Reset-gate linear maps.

Wh, Uh

Candidate-h linear maps.

Value

New hidden state.

Examples

set.seed(5)
d <- 3; nh <- 2
mk <- function(r, c) matrix(rnorm(r * c, sd = 0.5), r, c)
morie_gru4r_gru(x = rnorm(d), h = rep(0, nh), Wz = mk(nh, d), Uz = mk(nh, nh),
                Wr = mk(nh, d), Ur = mk(nh, nh), Wh = mk(nh, d), Uh = mk(nh, nh))
#> [1] -0.4362470  0.2313327