finishing up #302

DESCRIPTION

@@ -1,5 +1,5 @@
 Package: aqp
-Version: 2.0.2
+Version: 2.0.3
 Title: Algorithms for Quantitative Pedology
 Authors@R: c(person(given="Dylan", family="Beaudette", role = c("aut", "cre"), email = "dylan.beaudette@usda.gov"), person(given="Pierre", family="Roudier", email="roudierp@landcareresearch.co.nz", role = c("aut", "ctb")), person(given="Andrew", family="Brown", email="andrew.g.brown@usda.gov", role = c("aut", "ctb")))
 Author: Dylan Beaudette [aut, cre], Pierre Roudier [aut, ctb], Andrew Brown [aut, ctb]

NEWS.md

@@ -1,3 +1,6 @@
+# aqp 2.0.3 (2023-12-05)
+ * performance improvements in `profileInformationIndex()`
+
 # aqp 2.0.2 (2023-11-18)
  * CRAN release
  * bug fix / enhancements in `evalMissingData()`

R/profileInformationIndex.R

@@ -323,31 +323,54 @@ profileInformationIndex <- function(x, vars, method = c('joint', 'individual'),
   }
 
 
-  ## TODO: this will error / drop profiles in the presence of bad horizonation
+  ## note: this will error / drop profiles in the presence of bad horizonation
 
-  # dice() to 1cm intervals for common baseline
+  ## dice() to 1cm intervals for common baseline
   #  -> 10 horizons of the same data NOT more informative than 1 horizon 
   #  -> causes data corruption when bad hz depths present (lots of messages)
   x <- dice(x, fill = padNA)
 
-  ## TODO: convert to data.table
-  # SPC = FALSE
-  # data.table()
-  # dt[, ]
-  # as.vector()
-
-  # iterate over profiles
+  ## iterate over profiles
   # result is a vector suitable for site-level attribute
-  res <- profileApply(
-    x, 
-    simplify = TRUE, 
-    FUN = .PII_by_profile, 
-    vars = vars, 
-    baseline = baseline, 
-    method = method,
-    numericDigits = numericDigits, 
-    compression = compression
-  )
+
+  ## slow / but simple to understand and test
+  # res <- profileApply(
+  #   x,
+  #   simplify = TRUE,
+  #   FUN = .PII_by_profile,
+  #   vars = vars,
+  #   baseline = baseline,
+  #   method = method,
+  #   numericDigits = numericDigits,
+  #   compression = compression
+  # )
+  # return(res)
+
+  ## fast via data.table
+  .idn <- idname(x)
+  .pIDs <- profile_id(x)
+
+  # result is data.table: id + PII
+  # work on horizon data as data.frame
+  x <- data.table(horizons(x))
+  res <- x[ , 
+            .PII_by_profile(
+              x = .SD, 
+              vars = vars, 
+              baseline = baseline, 
+              method = method, 
+              numericDigits = numericDigits, 
+              compression = compression
+            ), 
+            by = .idn
+  ]
+
+  # double check order is preserved
+  stopifnot(all(res$id == .pIDs))
+
+  # down-grade data.table to named numeric vector
+  res <- as.vector(res$V1)
+  names(res) <- .pIDs
 
   # done
   return(res)

misc/sandbox/profile-information-index.R

@@ -86,7 +86,7 @@ y <- cbind(
   runif(n, min = min(x), max = max(x)), 
   runif(n, min = min(x), max = max(x)),
   runif(n, min = min(x), max = max(x))
-  )
+)
 cor(y)
 
 .cols <- hcl.colors(n = 100, palette = 'zissou 1')
@@ -425,12 +425,15 @@ x <- combine(x)
 
 vars <- c('hzname', 'hue', 'value', 'chroma', 'texture_class')
 
-
-z <- data.frame(
-  baseline.joint = profileInformationIndex(x, vars = vars, baseline = TRUE, method = 'j', padNA = FALSE),
-  baseline.individual = profileInformationIndex(x, vars = vars, baseline = TRUE, method = 'i', padNA = FALSE),
-  joint = profileInformationIndex(x, vars = vars, baseline = FALSE, method = 'j', padNA = FALSE),
-  individual = profileInformationIndex(x, vars = vars, baseline = FALSE, method = 'i', padNA = FALSE)
+# data.table: 44 seconds
+# profileApply:
+system.time(
+  z <- data.frame(
+    baseline.joint = profileInformationIndex(x, vars = vars, baseline = TRUE, method = 'j', padNA = FALSE),
+    baseline.individual = profileInformationIndex(x, vars = vars, baseline = TRUE, method = 'i', padNA = FALSE),
+    joint = profileInformationIndex(x, vars = vars, baseline = FALSE, method = 'j', padNA = FALSE),
+    individual = profileInformationIndex(x, vars = vars, baseline = FALSE, method = 'i', padNA = FALSE)
+  )
 )
 
 cor(z)
@@ -502,7 +505,7 @@ ggplot(site(x), aes(x = pi, y = greatgroup)) +
   stat_summary(geom = 'point', fun = median, shape = 21, fill = 'black', col = 'white', cex = 3) +
   scale_color_brewer() + 
   scale_x_continuous(n.breaks = 16) +
-  xlab('Profile Information Index') + ylab('') +
+  xlab('Profile Information Index (bytes)') + ylab('') +
   labs(title = 'Profile Information Index for Select Greatgroup Taxa', color = 'Interval')
 
 
@@ -579,3 +582,37 @@ par(mar = c(3, 0, 0, 2))
 plotSPC(z1, color = 'p1', plot.order = order(z1$pi), print.id = FALSE, width = 0.35, divide.hz = FALSE)
 axis(side = 1, at = 1:length(z1), labels = format(z1$pi[order(z1$pi)], digits = 3), cex.axis = 0.66)
 
+
+
+z1 <- lapply(
+  1:10000, 
+  random_profile, 
+  n = 5, 
+  exact = TRUE, 
+  n_prop = 3, 
+  SPC = TRUE, 
+  method = 'LPP', 
+  lpp.a = 5, 
+  lpp.b = 10, 
+  lpp.d = 5, 
+  lpp.e = 5, 
+  lpp.u = 25, 
+  min_thick = 2, 
+  max_thick = 50
+)
+
+z1 <- combine(z1)
+
+# 1k  : 4.6 seconds
+# 10k : 55 seconds
+system.time(
+  old <- profileInformationIndex(z1, vars = c('p1', 'p2', 'p3'), method = 'j', scale = TRUE, baseline = TRUE, padNA = FALSE)
+)
+
+# 1k  : 1.56 seconds
+# 10k : 18 seconds
+system.time(
+  new <- profileInformationIndex(z1, vars = c('p1', 'p2', 'p3'), method = 'j', scale = TRUE, baseline = TRUE, padNA = FALSE)
+)
+
+