Improve sits_apply() function

R/sits_classification.R

@@ -262,11 +262,12 @@ sits_classify_raster <- function (file = NULL, raster.tb, samples.tb, ml_method
 
  # ensure metadata tibble exists
  .sits_test_tibble (raster.tb)
+
  # ensure patterns tibble exits
  .sits_test_tibble (samples.tb)
 
  # ensure that file name and prediction model are provided
- ensurer::ensure_that(file,  !purrr::is_null(.), err_desc = "sits-classify-raster: please provide name of output file")
+ ensurer::ensure_that(file, !purrr::is_null(.), err_desc = "sits-classify-raster: please provide name of output file")
 
  # set up the ML model
  distances.tb <- sits_distances (samples.tb)

R/sits_tibble_functions.R

@@ -66,29 +66,61 @@ sits_apply <- function(data.tb, fun, fun_index = function(index){ return(index)
  # verify if data.tb has values
  .sits_test_tibble (data.tb)
 
- # computes fun and fun_index for all time series and substitutes the original time series data
- data.tb$time_series <- data.tb$time_series %>%
- purrr::map(function(ts.tb) {
- ts_computed.lst <- dplyr::select(ts.tb, -Index) %>%
- purrr::map(fun)
-
- # append bands names' suffixes
- if (nchar(bands_suffix) != 0)
- names(ts_computed.lst) <- paste0(names(ts_computed.lst), ".", bands_suffix)
-
- # unlist if there are more than one result from `fun`
- if (is.recursive(ts_computed.lst[[1]]))
- ts_computed.lst <- unlist(ts_computed.lst, recursive = FALSE)
-
- # convert to tibble
- ts_computed.tb <- tibble::as_tibble(ts_computed.lst)
-
- # compute Index column
- ts_computed.tb <- dplyr::mutate(ts_computed.tb, Index = fun_index(ts.tb$Index))
+ # save columns name from the first time series tibble
+ ts_cols <- names(data.tb$time_series[[1]])
+
+ # unnest tibble
+ data.tb <-
+ data.tb %>%
+ tidyr::unnest()
+
+ # computes fun for all time series fields and substitutes the original time series data
+ data.tb[,ts_cols] <-
+ data.tb[,ts_cols[-1:0]] %>%
+ purrr::map(fun) %>%
+ dplyr::bind_cols()
+
+ # computes fun_index for all time series fields and substitutes the original time series data
+ data.tb[,ts_cols[1]] <-
+ data.tb[,ts_cols[1]] %>%
+ purrr::map(fun_index) %>%
+ dplyr::bind_cols()
+
+ # verifies if there is a suffix and updates the columns names
+ if (nchar(bands_suffix) != 0)
+ ts_cols[-1:0] <- paste0(ts_cols[-1:0], ".", bands_suffix)
 
- # reorganizes time series tibble
- return(dplyr::select(ts_computed.tb, Index, dplyr::everything()))
- })
+ # append bands names' suffixes
+ names(data.tb) <- c(names(data.tb)[1:(length(names(data.tb))-length(ts_cols))], ts_cols)
+
+ # nest result and return
+ data.tb <-
+ data.tb %>%
+ tidyr::nest(ts_cols, .key = "time_series")
+ #
+ # # computes fun and fun_index for all time series and substitutes the original time series data
+ # data.tb$time_series <- data.tb$time_series %>%
+ # purrr::map(function(ts.tb) {
+ # ts_computed.lst <- dplyr::select(ts.tb, -Index) %>%
+ # purrr::map(fun)
+ #
+ # # append bands names' suffixes
+ # if (nchar(bands_suffix) != 0)
+ # names(ts_computed.lst) <- paste0(names(ts_computed.lst), ".", bands_suffix)
+ #
+ # # unlist if there are more than one result from `fun`
+ # if (is.recursive(ts_computed.lst[[1]]))
+ # ts_computed.lst <- unlist(ts_computed.lst, recursive = FALSE)
+ #
+ # # convert to tibble
+ # ts_computed.tb <- tibble::as_tibble(ts_computed.lst)
+ #
+ # # compute Index column
+ # ts_computed.tb <- dplyr::mutate(ts_computed.tb, Index = fun_index(ts.tb$Index))
+ #
+ # # reorganizes time series tibble
+ # return(dplyr::select(ts_computed.tb, Index, dplyr::everything()))
+ # })
  return(data.tb)
 }
 #' @title Apply a function on each SITS tibble column element

inst/extdata/markdown/sits_description.Rmd

@@ -14,9 +14,9 @@ author:
 - name: Gilberto Camara
  affiliation: National Institute for Space Reseach, INPE, Brazil
 - name: Alexandre Iwata
- affiliation: Institute for Applied Economics Research (IPEA), Brazil
+ affiliation: Institute for Applied Economics Research, IPEA, Brazil
 - name: Victor Maus
- affiliation: International Institute for Applied System Analysis (IIASA)
+ affiliation: International Institute for Applied System Analysis, IIASA, 
 abstract: "Using time series derived from big Earth Observation data sets is one of the leading research trends in Land Use Science and Remote Sensing. One of the more promising uses of satellite time series is its application for classification of land use and land cover, since our growing demand for natural resources has caused major environmental impacts. The SITS package provides support on how to use statistical learning techniques with image time series. These methods include linear and quadratic discrimination analysis, support vector machines, random forests and neural networks."
 date: "`r format(Sys.time(), '%B %d, %Y')`"
 fontfamily: mathdesign