tests/testthat/test-mage.R

In iglu: Interpreting Glucose Data from Continuous Glucose Monitors

# 1. The "manual_calculations.csv" file contains the original manual calculations from "https://journals.sagepub.com/doi/10.1177/19322968211061165"
#
# 2. The "mage_ground_truth.csv" file contains the iglu::mage() results (for iglu v.4.0.0) after significant modifications to the original algorithm.
#     We have extensively vetted the current version of iglu::mage for accuracy and are confident the current outputs on the input data are correct.
#
# 3. The mage_plots/ folder is a helper folder - run "make_pdfs" to visualize the output of the MAGE algorithm



### 0. load in data (assumes cwd is iglu's project directory)
# filepath = paste0(getwd(), '/tests/testthat/data/')
filepath = paste0(getwd(), '/data/')
manual_calc <- read.csv(paste0(filepath, "manual_calculations.csv"))
JHU <- iglu::example_data_5_subject

# A. exclude the samples where the comment is "exclude"
cgm_all_data <- lapply(1:length(manual_calc$dataset), function(x) {
  as.list(manual_calc[x, ])
})

cgm_all_data <- Filter(function(x) is.na(x$comment) || x$comment != "exclude", cgm_all_data)

# B. Subset the complete data sets by the row numbers found in the manual calculations
cgm_dataset_df <- lapply(cgm_all_data, function(x) {
  dataset <- x$dataset

  if (dataset != "Dubosson2018" && dataset != "Tsalikian2005" && dataset != "Hall2018") {
    eval(parse(text=dataset))[x$start:x$end, ] # evaluate the text
  }
})

cgm_manual_calc <- sapply(cgm_all_data, function(x) x$manual) # get manual calculations

idx_to_remove = which(lengths(cgm_dataset_df) == 0)

cgm_dataset_df <- cgm_dataset_df[-idx_to_remove]
cgm_manual_calc <- cgm_manual_calc[-idx_to_remove]

### 0.25 Helper F(x)
# make_pdfs <- function(filepath, filename_prefix) {
#   prefix = paste0(filepath, "/", filename_prefix)
#
#   make_plot <- function(direction) {
#     pdf(file = paste0(prefix, "_mage_", direction, ".pdf"), width=10, height=5, onefile=TRUE)
#     for (i in 1:length(cgm_manual_calc)) {
#       manual = cgm_manual_calc[i]
#       val = round(iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction = direction)$MAGE, 2)
#       p = iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, direction = direction, plot = TRUE, static_or_gui = "ggplot", title=paste0("[i=", i, "] MAGE", direction, " = ", val, " (% diff from manual calc: ", round((val - manual)/manual*100, 2), "%)"))
#       print(p)
#     }
#     dev.off()
#   }
#
#   make_plot("plus")
#   make_plot("minus")
#   make_plot("service")
#   make_plot("avg")
#   # make_plot("max") # TODO: plotting functionality for MAGEmax hasn't been implemented yet
# }
#
# ### 0.5 Params
short_ma = 5
long_ma = 32
inter_gap = 45
max_gap = 180
return_type = "num"

# ### 1. Generate Ground Truth Labels
# ### If you are sure that the current iglu::mage has been thoroughly validated and is accurate, you can
# ### generate "ground truth labels" (i.e., what the output of the current iglu::mage f(x) is on some dataframes).
# ### Then, future tests will check for compliance/identicalness w/ the current MAGE outputs
#
# df = setNames(data.frame(matrix(ncol = 6, nrow = 0)), c("idx", "mage_plus", "mage_minus", "mage_avg", "mage_service", "mage_max"))
#
# for (i in 1:length(cgm_dataset_df)) {
#   mage_plus = iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction = "plus")$MAGE
#   mage_minus = iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction = "minus")$MAGE
#   mage_avg = iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction = "avg")$MAGE
#   mage_service = iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction = "service")$MAGE
#   mage_max = iglu::mage(cgm_dataset_df[[i]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction = "max")$MAGE
#
#   df <- base::rbind(df, data.frame(idx=i, mage_plus=mage_plus, mage_minus=mage_minus, mage_avg=mage_avg, mage_service=mage_service, mage_max=mage_max))
# }
#
# write.csv(df, file=base::paste0(filepath, 'mage_ground_truth.csv'))
# make_pdfs(paste0(filepath, '/mage_plots'), 'ground_truth') # see [here](https://github.com/Nathaniel-Fernandes/iglu/tree/df78ce4d1cde06a9b744afd71b8e1605971c5c54/tests/testthat/data/mage_plots) for ground truth plots

### 2. TESTS: Check for compliance (i.e., identicalness) to previous validated version of MAGE

ground_truth <- read.csv(paste0(filepath, 'mage_ground_truth.csv'))

# test MAGE+
test_that("iglu::MAGE+ == old_iglu::MAGE+", {
  for (i in 1:nrow(ground_truth)) {
    row = ground_truth[i, ]
    expect_equal(iglu::mage(cgm_dataset_df[[row$idx]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction="plus")$MAGE, row$mage_plus, tolerance = 0.1)
  }
})

# test MAGE-
test_that("iglu::MAGE- == old_iglu::MAGE-", {
  for (i in 1:nrow(ground_truth)) {
    row = ground_truth[i, ]
    expect_equal(iglu::mage(cgm_dataset_df[[row$idx]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction="minus")$MAGE, row$mage_minus, tolerance = 0.1)
  }
})

# test MAGE Avg
test_that("iglu::MAGEavg == old_iglu::MAGEavg", {
  for (i in 1:nrow(ground_truth)) {
    row = ground_truth[i, ]
    expect_equal(iglu::mage(cgm_dataset_df[[row$idx]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction="avg")$MAGE, row$mage_avg, tolerance = 0.1)
  }
})

# test MAGE Service
test_that("iglu::MAGEservice == old_iglu::MAGEservice", {
  for (i in 1:nrow(ground_truth)) {
    row = ground_truth[i, ]
    expect_equal(iglu::mage(cgm_dataset_df[[row$idx]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction="service")$MAGE, row$mage_service, tolerance = 0.1)
  }
})

# test MAGE Max
test_that("iglu::MAGEmax == old_iglu::MAGEmax", {
  for (i in 1:nrow(ground_truth)) {
    row = ground_truth[i, ]
    expect_equal(iglu::mage(cgm_dataset_df[[row$idx]], short_ma = short_ma, long_ma = long_ma, inter_gap = inter_gap, max_gap = max_gap, return_type = "num", direction="max")$MAGE, row$mage_max, tolerance = 0.1)
  }
})

### 3. Debugging Help
# make_pdfs(paste0(getwd(), '/tests/testthat/data/mage_plots'), 'debug') # if running from command line, you need to change 1st param to be 'iglu' directory