up doc

Author: RafaelSdeSouza

NAMESPACE

@@ -12,11 +12,7 @@ export(plot_cluster_spectra)
 export(reconstruct_cluster_cube)
 export(reconstruct_flux_preserving_cube)
 export(segment)
-export(segment_approx)
 export(segment_big_cube)
-export(segment_blockward)
-export(segment_masked)
-export(segment_robust)
 export(segment_starlet)
 export(summarize_cluster_spectra)
 export(torch_dist)

NEWS.md

@@ -4,8 +4,7 @@ Released: 2026-03-18
 
 ## Added
 
-- `segment_masked()` for missing-data-safe segmentation with the same clustering logic as `segment()`.
-- `segment_big_cube()` as a clearer user-facing name for scalable large-cube segmentation when exact all-pairs distances would exceed available RAM.
+- `segment_big_cube()` as the user-facing scalable segmentation path for cubes where exact all-pairs distances would exceed available RAM.
 - `build_starlet_mask()` and `segment_starlet()` for Sagui-style white-light starlet masking before clustering.
 - `summarize_cluster_spectra()` for median, summed, and inverse-variance-weighted cluster spectra.
 - `choose_ncomp_by_snr()` for variance-aware component selection from an SNR threshold.
@@ -14,9 +13,11 @@ Released: 2026-03-18
 
 ## Changed
 
-- `segment()` and `segment_masked()` now share the same internal segmentation core.
+- `segment()` now handles missing spectral channels directly in the exact workflow.
+- `segment_big_cube()` now uses block medoids rather than block averages, improving compact structures in large cubes.
+- `segment_starlet()` can now apply the white-light mask before either the exact or big-cube backend.
 - `torch` is now optional; Capivara falls back to base R distance calculations when `torch` is unavailable.
-- `segment_robust()` and `cube_cluster_with_snr()` are preserved as compatibility wrappers around the new API.
+- The public segmentation API is now limited to `segment()`, `segment_big_cube()`, and `segment_starlet()` to avoid duplicate legacy entry points.
 - GitHub and website documentation now describe the missing-data, starlet-mask, and variance-aware workflows.
 
 ## Fixed

R/cube_cluster_with_snr.R

@@ -26,7 +26,7 @@
 #' @importFrom stats cutree
 #'
 #' @seealso \code{\link{choose_ncomp_by_snr}}, \code{\link{segment}},
-#'   \code{\link{segment_masked}}
+#'   \code{\link{segment}}
 #'
 #' @examples
 #' if (torch::torch_is_installed()) {

R/segment.R

@@ -7,6 +7,11 @@
 #' with the original spatial dimensions. The function also retains the original data cube
 #' for reference and post-processing.
 #'
+#' Missing spectral channels are handled automatically: non-finite values produced
+#' during row-wise scaling are replaced with zero before distances are computed.
+#' This keeps the standard exact workflow usable for masked cubes without a
+#' separate public entry point.
+#'
 #' @param input A FITS object representing the input data cube. Typically, this is an IFU data cube.
 #' @param Ncomp Integer, the number of clusters to form.
 #' @param redshift Numeric, the redshift to apply for wavelength correction. Defaults to 0 (no correction).
@@ -35,7 +40,9 @@
 #' This process is often used in IFU data analysis, where clustering is applied to grouped
 #' spectral profiles of spatial pixels to identify regions with similar characteristics.
 #'
-#' @seealso \code{\link{torch_dist}}, \code{\link[fastcluster]{hclust}}, \code{\link[stats]{cutree}}
+#' @seealso \code{\link{segment_big_cube}}, \code{\link{segment_starlet}},
+#'   \code{\link{torch_dist}}, \code{\link[fastcluster]{hclust}},
+#'   \code{\link[stats]{cutree}}
 #'
 #' @examples
 #' if (torch::torch_is_installed()) {
@@ -56,6 +63,6 @@ segment <- function(input, Ncomp = 5, redshift = 0, scale_fn = median_scale) {
     Ncomp = Ncomp,
     redshift = redshift,
     scale_fn = scale_fn,
-    na_to_zero = FALSE
+    na_to_zero = TRUE
   )
 }