add memory usage recoding & figs

Snakefile

@@ -818,6 +818,10 @@ plts = {
 	"memory": expand(
 		"plts/mbs_{reftyp}.rds",
 		reftyp = REFTYPS),
+	"scalability": expand(
+		"plts/{which}_{reftyp}.rds",
+		which = ["rts", "mbs"],
+		reftyp = REFTYPS),
 	"scatters": expand(
 		"plts/stat_{dim}d-scatters.rds",
 		dim = ["1", "2"]),

code/07-plot_memory.R

@@ -36,11 +36,20 @@ df <- data.frame(mbs, reftyp, refset, vars) %>%
     group_by(method, dim, n) %>%
     summarise_at("mbs", mean)
 
+# order methods by average across subsets
+ms <- df %>% 
+    group_by(method) %>% 
+    summarise_at("mbs", mean) %>% 
+    arrange(mbs) %>% pull("method")
+df$method <- factor(df$method, levels = ms)
+
 pal <- .methods_pal[levels(df$method)]
+
 lab <- parse(text = paste(
     sep = "~",
     sprintf("bold(%s)", LETTERS), 
     gsub("\\s", "~", names(pal))))
+
 anno <- mutate(df, letter = LETTERS[
     match(method, levels(method))])    
 

code/07-plot_runtimes.R

@@ -46,6 +46,14 @@ df <- res %>%
     group_by(method, step, dim, n) %>%
     summarise_at("t", mean)
 
+# order methods by overall average across subsets
+ms <- df %>% 
+    filter(step == "overall") %>% 
+    group_by(method) %>% 
+    summarise_at("t", mean) %>% 
+    arrange(t) %>% pull("method")
+df$method <- factor(df$method, levels = ms)
+
 pal <- .methods_pal[levels(df$method)]
 
 lab <- parse(text = paste(
@@ -65,13 +73,13 @@ plt <- ggplot(df, aes(factor(n), t,
         stat = "identity",
         position = "dodge") +
     geom_text(data = anno, 
-        aes(label = letter, y = 0.1),
-        size = 1.5, color = "black",
+        aes(label = letter, y = 0),
+        size = 1.5, vjust = 2, color = "black",
         position = position_dodge(0.9)) + 
     scale_fill_manual(values = pal, labels = lab) +
     scale_color_manual(values = pal, labels = lab) +
-    xlab(NULL) + scale_y_log10("runtime (s)", 
-        expand = expansion(mult = 0.1))
+    xlab(NULL) + scale_y_sqrt("runtime (s)", 
+        expand = expansion(mult = c(0.2, 0.1)))
 
 thm <- theme(
     axis.ticks.x = element_blank(),

code/08-fig_scalability.R

@@ -0,0 +1,59 @@
+# args <- list(
+#     rds = list.files("plts", "(rts|mbs).*\\.rds", full.names = TRUE),
+#     pdf = "figs/rts_vs_mbs.pdf",
+#     uts = "code/utils-plotting.R")
+
+source(args$uts)
+ps <- lapply(args$rds, readRDS)
+
+pat <- ".*_([a-z])\\.rds"
+reftyp <- gsub(pat, "\\1", basename(args$rds))
+is <- split(seq_along(ps), reftyp)[c("n", "b", "k")]
+
+ps <- lapply(is, \(i) {
+
+    nm <- c("rts", "mbs")
+    for (j in seq_along(nm)) {
+        k <- grep(nm[[j]], args$rds[i])
+        df <- paste0("df_", nm[[j]])
+        assign(df, ps[i][[k]]$data)
+    }
+
+    i <- c("method", "dim", "n", j <- c("t", "mbs"))
+    #f <- \(.) factor(., sort(as.numeric(as.character(unique(.)))))
+    f <- \(.) as.numeric(as.character(.))
+    df_rts <- df_rts %>% 
+        filter(step == "overall") %>% 
+        mutate(n = f(n)) %>% ungroup() %>% select(any_of(i)) 
+
+    df_mbs <- df_mbs %>% 
+        mutate(n = f(n)) %>% ungroup() %>% select(any_of(i)) 
+
+    df <- full_join(
+        df_rts, df_mbs, 
+        by = setdiff(i, j)) %>% 
+        filter(!is.na(t), !is.na(mbs))
+    pal <- .methods_pal[levels(df$method)]
+
+    p <- ggplot(df, aes(t, mbs, col = method, fill = method)) +
+        scale_color_manual(values = pal) +
+        scale_fill_manual(values = pal) +
+        facet_grid(~ dim) +
+        geom_point(alpha = 0.8) + 
+        geom_path(size = 0.2) +
+        scale_x_continuous("runtime (s)", limits = c(0, NA), trans = "sqrt") + 
+        scale_y_continuous("memory usage (MBs)", limits = c(0, NA)) 
+
+    .prettify(p) + theme(
+        panel.grid.major = element_line(color = "grey", size = 0.2))
+})
+
+fig <- wrap_plots(ps, ncol = 1) +
+    plot_annotation(tag_levels = "a") &
+    theme(
+        plot.margin = margin(),
+        panel.spacing = unit(1, "mm"),
+        legend.title = element_blank(),
+        plot.tag = element_text(size = 9, face = "bold"))
+
+ggsave(args$pdf, fig, width = 16, height = 16, units = "cm")