NEWS.md
getRDS() updated with new URL.clusterPlot() accepts character vectors and factors as arguments to label.spatialPreprocess() uses exact rather than approximate PCA by default.spatialCluster() and spatialEnhance() now use a faster implementation of the multivariate normal density that reduces runtime by approximately 40%.qTune(), the min_rep and max_rep parameters have been replaced with burn.in and nrep, respectively, to be consistent with spatialCluster().getRDS() gains a cache parameter. When TRUE, the RDS is cached locally using BiocFileCache.spatialCluster() and spatialEnhance() handle the edge case where only one iteration is kept after excluding burn-in.coda::mcmc object returned by mcmcChain() now specifies the thinning interval used in enhanced objects.spatialCluster() and spatialEnhance() now include platform-specific defaults for the gamma parameter.spatialCluster() and spatialEnhance(), setting burn.in equal to nrep now raises an error.enhanceFeatures() now takes an nrounds parameter that corresponds to the same parameter in xgboost. If nrounds is set to 0, we automatically tune the parameter using a train/test split for improved feature prediction.spatialCluster() and spatialEnhance() both gain a burn.in parameter specifying the number of MCMC iterations to exclude when aggregating cluster labels and enhanced PCs.clusterPlot(), label now accepts factors and vectors of strings, in addition to numeric vectors or a column name in colData.spatialEnhance(), PCs are now averaged over the MCMC iterations (excluding the burn-in period).enhanceFeatures(), negative expression is now clipped to 0.spatialPreprocess() now adds a boolean is.HVG column to rowData.featurePlot(), additional arguments to geom_polygon() are correctly passed through.spatialEnhance() incorrectly added row offset to spot column coordinate when generating subspot colData, and vice versa. This resulted in subspots being reflected over y=x in spatial plots, and has been fixed.