## add 'developer/' to packages to be installed from github
x <- c("pbapply", "viridis", "knitr", "remotes", "maRce10/warbleR", "maRce10/ohun", "DT", "ranger")
aa <- lapply(x, function(y) {
# get pakage name
pkg <- strsplit(y, "/")[[1]]
pkg <- pkg[length(pkg)]
# check if installed, if not then install
if (!pkg %in% installed.packages()[,"Package"]) {
if (grepl("/", y)) remotes::install_github(y, force = TRUE) else
install.packages(y)
}
# load package
try(require(pkg, character.only = T), silent = T)
})cols <- viridis(10, alpha = 0.6)
warbleR_options(wav.path = "./data/raw")manual_detections_Raven_bats4testing.csv: Manual detection of all visible bat vocalizations (Raven)
manual_detections_bats4testing.csv: Manual detection of all high quality bat vocalizations (Avisoft)
automated_detections_bats4testing.csv: Automated detection with Eric’s function
reference_all <- read.csv("./data/raw/manual_detections_Raven_bats4testing.csv")
reference_best <- read.csv("./data/raw/manual_detections_bats4testing.csv")
ericks_detection <- read.csv("./data/raw/automated_detections_bats4testing.csv")Exhaustive detection (all signals)
# all
check_sels(reference_all, pb = FALSE)## all selections are OKfeature_reference(reference_all, path = "./data/raw")## min mean max
## sel.duration 0.00046883 0.003376288 0.03785942
## gap.duration 0.01268074 0.08107097 0.938257
## duty.cycle 0.03712569 0.006502513 0.1009111High quality subset
check_sels(reference_best, pb = FALSE)## all selections are OKfeature_reference(reference_best, path = "./data/raw")## min mean max
## sel.duration 0.000208 0.003602071 0.03638
## gap.duration 0.012139 0.1011788 0.937004
## duty.cycle 0.03020532 0.004873684 0.08718Detection from Erick’s algorithm
check_sels(ericks_detection, pb = FALSE)## all selections are OK but some have very few samples (less than 20, potentially problematic for some analyses)
## Check 'min.n.samples' columnfeature_reference(ericks_detection, path = "./data/raw")## min mean max
## sel.duration 2.604167e-06 0.005757727 0.2316797
## gap.duration 0.001611979 0.07480331 0.936732
## duty.cycle 0.05292565 0.006045013 0.251075# all vs best
diagnose_detection(reference = reference_all, detection = reference_best)## true.positives false.positives false.negatives split.positives
## 1 139 1 59 1
## merged.positives overlap.to.true.positives sensitivity specificity
## 1 0 0.7612143 0.7020202 0.9928571
Compare to exhaustive detection:
diagnose_detection(reference = reference_all, detection = ericks_detection)## true.positives false.positives false.negatives split.positives
## 1 98 74 100 4
## merged.positives overlap.to.true.positives sensitivity specificity
## 1 2 0.626583 0.4949495 0.5697674
Compare to high quality subset:
diagnose_detection(reference = reference_best, detection = ericks_detection)## true.positives false.positives false.negatives split.positives
## 1 100 73 41 3
## merged.positives overlap.to.true.positives sensitivity specificity
## 1 2 0.6925425 0.7092199 0.5780347
ohun_opt_detec <- optimize_energy_detector(reference = reference_best, threshold = c(0.5, 1:4), smooth = 1:4, hold.time = 1:8, min.duration = c(0.05, 0.1, 0.2, 0.3, 0.4, 1, 2), max.duration = c(30, 40, 50), parallel = 10, path = "./data/raw", bp = c(5, 100))
saveRDS(ohun_opt_detec, "./data/processed/high_quality_reference_optimization.RDS")
ohun_opt_detec_all <- optimize_energy_detector(reference = reference_all, threshold = c(0.5, 1:4), smooth = 1:4, hold.time = 1:8, min.duration = c(0.05, 0.1, 0.2, 0.3, 0.4, 1, 2), max.duration = c(30, 40, 50), parallel = 10, path = "./data/raw", bp = c(5, 100))
saveRDS(ohun_opt_detec_all, "./data/processed/all_calls_reference_optimization.RDS")
ohun_opt_detec_ericks <- optimize_energy_detector(reference = ericks_detection, threshold = c(0.5, 1:4), smooth = 1:4, hold.time = 1:8, min.duration = c(0.05, 0.1, 0.2, 0.3, 0.4, 1, 2), max.duration = c(30, 40, 50), parallel = 10, path = "./data/raw", bp = c(5, 100))
saveRDS(ohun_opt_detec_ericks, "./data/processed/ericks_reference_optimization.RDS")Compare to exhaustive reference (only the highest sensitivity iteration is shown)
ohun_opt_detec_all <- readRDS("./data/processed/all_calls_reference_optimization.RDS")
ohun_opt_detec_all <- ohun_opt_detec_all[which.max(ohun_opt_detec_all$sensitivity), c(1:11, 17:18)]
ohun_opt_detec_all## threshold smooth hold.time min.duration max.duration thinning
## 1025 1 1 1 0.1 50 1
## true.positives false.positives false.negatives split.positives
## 1025 138 43 60 1
## merged.positives sensitivity specificity
## 1025 3 0.6969697 0.7624309# oa_DT <- datatable(ohun_opt_detec_all, editable = list(
# target = 'row'
# ), rownames = FALSE, style = "bootstrap", filter = 'top', options = list(
# pageLength = 100, autoWidth = TRUE, dom = 'ft'
# ), autoHideNavigation = TRUE, escape = FALSE)
#
# formatRound(table = oa_DT, columns = sapply(ohun_opt_detec_all, is.numeric), 3)
Compare to high quality reference (only the highest sensitivity iteration is shown)
ohun_opt_detec <- readRDS("./data/processed/high_quality_reference_optimization.RDS")
ohun_opt_detec <- ohun_opt_detec[which.max(ohun_opt_detec$sensitivity), c(1:11, 17:18)]
ohun_opt_detec## threshold smooth hold.time min.duration max.duration thinning true.positives
## 1 2 1 1 0.1 40 1 105
## false.positives false.negatives split.positives merged.positives sensitivity
## 1 48 36 0 0 0.7446809
## specificity
## 1 0.6862745# oa_DT <- datatable(ohun_opt_detec, editable = list(
# target = 'row'
# ), rownames = FALSE, style = "bootstrap", filter = 'top', options = list(
# pageLength = 100, autoWidth = TRUE, dom = 'ft'
# ), autoHideNavigation = TRUE, escape = FALSE)
#
# formatRound(table = oa_DT, columns = sapply(ohun_opt_detec, is.numeric), 3)
Compare to Erick’s detection (only the highest sensitivity iteration is shown)
ohun_opt_detec_ericks <- readRDS("./data/processed/ericks_reference_optimization.RDS")
ohun_opt_detec_ericks <- ohun_opt_detec_ericks[which.max(ohun_opt_detec_ericks$sensitivity), c(1:11, 17:18)]
ohun_opt_detec_ericks## threshold smooth hold.time min.duration max.duration thinning
## 514 2 1 1 0.1 40 1
## true.positives false.positives false.negatives split.positives
## 514 115 51 45 1
## merged.positives sensitivity specificity
## 514 10 0.71875 0.6927711# oa_DT <- datatable(ohun_opt_detec_ericks, editable = list(
# target = 'row'
# ), rownames = FALSE, style = "bootstrap", filter = 'top', options = list(
# pageLength = 100, autoWidth = TRUE, dom = 'ft'
# ), autoHideNavigation = TRUE, escape = FALSE)
#
# formatRound(table = oa_DT, columns = sapply(ohun_opt_detec_ericks, is.numeric), 3)
Before random forest filtering
# best parameters from all signal optimization
ohun_detec <- energy_detector(threshold = 1, smooth = 1, hold.time = 1, min.duration = 0.1, max.duration = 50, parallel = 10, path = "./data/raw", bp = c(5, 100), pb = FALSE)
diagnose_detection(reference_all, ohun_detec)## true.positives false.positives false.negatives split.positives
## 1 138 43 60 1
## merged.positives overlap.to.true.positives sensitivity specificity
## 1 3 0.9351629 0.6969697 0.7624309After random forest filtering
ohun_detec <- label_detection(reference = reference_all, detection = ohun_detec, pb = FALSE)
# measure spectrographic parameters
spectral_parameters <- spectro_analysis(ohun_detec, bp = c(5, 100), fast = TRUE, ovlp = 70, parallel = 10, pb = FALSE)
spectral_parameters$SNR <- sig2noise(ohun_detec, mar = 0.01, pb = FALSE)$SNR
spectral_parameters$class <- ohun_detec$detection.class
spectral_parameters$class[spectral_parameters$class != "false.positive"] <- "true.positive"
# make it a factor for ranger to work
spectral_parameters$class <- as.factor(spectral_parameters$class)
# run RF model on spectral parameters
rfm <-
ranger(
class ~ .,
data = spectral_parameters[, !names(spectral_parameters) %in% c("sound.files", "selec")],
num.trees = 10000,
importance = "impurity",
seed = 10
)
# table(lab_detec$detection.class)
ohun_detec$pred.class <- rfm$predictions
rf_ohun_detec <- ohun_detec[ohun_detec$pred.class == "true.positive", ]
diagnose_detection(reference = reference_all, detection = rf_ohun_detec, pb = FALSE)## true.positives false.positives false.negatives split.positives
## 1 134 6 64 1
## merged.positives overlap.to.true.positives sensitivity specificity
## 1 2 0.9347908 0.6767677 0.9571429
R session information
## R version 4.1.0 (2021-05-18)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.2 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/atlas/libblas.so.3.10.3
## LAPACK: /usr/lib/x86_64-linux-gnu/atlas/liblapack.so.3.10.3
##
## locale:
## [1] LC_CTYPE=pt_BR.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=es_CR.UTF-8 LC_COLLATE=pt_BR.UTF-8
## [5] LC_MONETARY=es_CR.UTF-8 LC_MESSAGES=pt_BR.UTF-8
## [7] LC_PAPER=es_CR.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=es_CR.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] ranger_0.13.1 DT_0.20 ohun_0.1.0 warbleR_1.1.27
## [5] NatureSounds_1.0.4 seewave_2.1.8 tuneR_1.3.3.1 remotes_2.4.2
## [9] knitr_1.36 viridis_0.6.2 viridisLite_0.4.0 pbapply_1.5-0
##
## loaded via a namespace (and not attached):
## [1] tidyselect_1.1.1 xfun_0.28 bslib_0.2.5.1 purrr_0.3.4
## [5] lattice_0.20-44 colorspace_2.0-2 vctrs_0.3.8 generics_0.1.0
## [9] htmltools_0.5.2 yaml_2.2.1 utf8_1.2.2 rlang_0.4.12
## [13] jquerylib_0.1.4 pillar_1.6.4 glue_1.5.1 DBI_1.1.1
## [17] lifecycle_1.0.1 stringr_1.4.0 munsell_0.5.0 gtable_0.3.0
## [21] htmlwidgets_1.5.4 evaluate_0.14 fastmap_1.1.0 fftw_1.0-6.1
## [25] parallel_4.1.0 fansi_0.5.0 Rcpp_1.0.7 scales_1.1.1
## [29] jsonlite_1.7.2 soundgen_2.1.0 gridExtra_2.3 rjson_0.2.20
## [33] ggplot2_3.3.5 digest_0.6.29 stringi_1.7.6 dplyr_1.0.7
## [37] dtw_1.22-3 grid_4.1.0 tools_4.1.0 bitops_1.0-7
## [41] magrittr_2.0.1 sass_0.4.0 proxy_0.4-26 RCurl_1.98-1.5
## [45] tibble_3.1.6 crayon_1.4.2 pkgconfig_2.0.3 Matrix_1.3-4
## [49] ellipsis_0.3.2 MASS_7.3-54 shinyBS_0.61 assertthat_0.2.1
## [53] rmarkdown_2.9 R6_2.5.1 signal_0.7-7 compiler_4.1.0