## ----gw-knit-opts, include = FALSE--------------------------------------------
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 6, fig.height = 4,
  fig.align = "center"
)

## ----setup--------------------------------------------------------------------
library(MetaHunt)
set.seed(1)

## ----sim-data-----------------------------------------------------------------
m <- 40
G <- 30
x <- seq(0, 1, length.out = G)

# True bases on the grid.
basis <- rbind(sin(pi * x), cos(pi * x), x)

# Mixing weights driven by two latent covariates.
W <- data.frame(w1 = rnorm(m), w2 = rnorm(m))
beta <- cbind(c(1, -0.8), c(-0.5, 1.2), c(0, 0))
eta  <- as.matrix(W) %*% beta
pi_true <- exp(eta) / rowSums(exp(eta))

F_hat <- pi_true %*% basis + matrix(rnorm(m * G, sd = 0.05), m, G)
dim(F_hat)

## ----two-weights--------------------------------------------------------------
gw_uniform <- rep(1 / G, G)

# Up-weight x > 0.5 by ~5x. Renormalisation is optional but tidy.
gw_right   <- ifelse(x > 0.5, 5, 1)
gw_right   <- gw_right / sum(gw_right) * G   # comparable scale, optional

fit_unif  <- dfspa(F_hat, K = 3, grid_weights = gw_uniform, denoise = FALSE)
fit_right <- dfspa(F_hat, K = 3, grid_weights = gw_right,   denoise = FALSE)

# Index sets selected can differ.
fit_unif$original_indices
fit_right$original_indices

## ----overlay-plot-------------------------------------------------------------
matplot(
  x,
  cbind(fit_unif$bases[1, ], fit_right$bases[1, ]),
  type = "l", lty = 1, lwd = 2,
  col  = c("#0072B2", "#D55E00"),
  xlab = "x",
  ylab = "first recovered basis",
  main = expression(
    "Recovered first-vertex basis under uniform vs. target-weighted " ~ L^2 * (mu)
  )
)
abline(v = 0.5, lty = 2, col = "grey60")
legend("topright",
       legend = c("uniform", "up-weight x > 0.5", "x = 0.5 boundary"),
       col    = c("#0072B2", "#D55E00", "grey60"),
       lty = c(1, 1, 2), lwd = c(2, 2, 1), bty = "n")