Maneira eficiente de eliminar linhas com tempos de sobreposição

Eu tenho um longo conjunto de dados com colunas que representam os horários de início e de término e quero descartar uma linha se ela se sobrepuser a outra e tiver uma prioridade mais alta (por exemplo, 1 é a prioridade mais alta). Meus dados de exemplo são

library(tidyverse)
library(lubridate)
times_df <- tibble(start = as_datetime(c("2019-10-05 14:05:25", 
    "2019-10-05 17:30:20", 
    "2019-10-05 17:37:00", 
    "2019-10-06 04:43:55", 
    "2019-10-06 04:53:45")), 
    stop = as_datetime(c("2019-10-05 14:19:20",
    "2019-10-05 17:45:15", 
    "2019-10-05 17:50:45", 
    "2019-10-06 04:59:00",
    "2019-10-06 05:07:10")), priority = c(5,3,4,3,4))

A maneira como desenvolvi ataca o problema de trás para frente, encontrando as sobreposições com um valor de prioridade mais alto e usando um anti_joinpara removê-las do quadro de dados original. Esse código não funciona se houver três períodos sobrepostos ao mesmo ponto no tempo e tenho certeza de que há uma maneira mais eficiente e funcional de fazer isso.

dropOverlaps <- function(df) {
    drops <- df %>% 
        filter(stop > lead(start) | lag(stop) > start) %>% 
        mutate(group = ({seq(1, nrow(.)/2)} %>% 
        rep(each=2))) %>% 
        group_by(group) %>% 
        filter(priority == max(priority))
    anti_join(df, drops)
}

dropOverlaps(times_df)
#> Joining, by = c("start", "stop", "priority")
#> # A tibble: 3 x 3
#>   start               stop                priority
#>   <dttm>              <dttm>                 <dbl>
#> 1 2019-10-05 14:05:25 2019-10-05 14:19:20        5
#> 2 2019-10-05 17:30:20 2019-10-05 17:45:15        3
#> 3 2019-10-06 04:43:55 2019-10-06 04:59:00        3

Alguém pode me ajudar a obter a mesma saída, mas com uma função mais limpa? Bônus se ele puder manipular uma entrada com três ou mais períodos de tempo que se sobrepõem.

Aqui está uma data.tablesolução usada foverlapspara detectar os registros sobrepostos (como já mencionado por @GenesRus). Os registros sobrepostos são atribuídos a grupos para filtrar o registro com no máx. prioridade no grupo. Adicionei mais dois registros aos dados de exemplo, para mostrar que este procedimento também está funcionando para três ou mais registros sobrepostos:

Edit: Modifiquei e traduzi a solução @ pgcudahy para a data.tablequal fornece código ainda mais rápido:

library(data.table)
library(lubridate)

times_df <- data.frame(
  start = as_datetime(
    c(
      "2019-10-05 14:05:25",
      "2019-10-05 17:30:20",
      "2019-10-05 17:37:00",
      "2019-10-06 04:43:55",
      "2019-10-06 04:53:45",
      "2019-10-06 04:53:46",
      "2019-10-06 04:53:47"
    )
  ),
  stop = as_datetime(
    c(
      "2019-10-05 14:19:20",
      "2019-10-05 17:45:15",
      "2019-10-05 17:50:45",
      "2019-10-06 04:59:00",
      "2019-10-06 05:07:10",
      "2019-10-06 05:07:11",
      "2019-10-06 05:07:12"
    )
  ),
  priority = c(5, 3, 4, 3, 4, 5, 6)
)

resultDT <- setDT(times_df, key="start")[!(stop >= shift(start, type="lead", fill = TRUE) & priority > shift(priority, type="lead", fill = TRUE)) &
                                         !(start <= shift(stop, type="lag", fill = FALSE) & priority > shift(priority, type="lag", fill = TRUE))]

# old approach ------------------------------------------------------------
# times_dt <- as.data.table(times_df)
# setkey(times_dt, start, stop)[, index := .I]
# overlaps_dt <- foverlaps(times_dt, times_dt, type = "any", which = TRUE)[xid != yid][, group := fifelse(xid > yid, yes = paste0(yid, "_", xid), no = paste0(xid, "_", yid))]
# overlaps_merged <- merge(times_dt, overlaps_dt, by.x = "index", by.y = "xid")[, .(delete_index = index[priority == max(priority)]), by = "group"]
# result_dt <- times_dt[!unique(overlaps_merged$delete_index)][, index := NULL]

Para mais detalhes, consulte ?foverlaps- Existem alguns recursos mais úteis implementados para controlar o que é considerado uma sobreposição, como maxgap, minoverlapou type(qualquer, dentro, início, fim e igual).

Atualização - nova referência

Unit: microseconds
          expr       min         lq      mean    median        uq        max neval
          Paul 25572.550 26105.2710 30183.930 26514.342 29614.272 153810.600   100
           MKa  5100.447  5276.8350  6508.333  5401.275  5832.270  23137.879   100
      pgcudahy  3330.243  3474.4345  4284.640  3556.802  3748.203  21241.260   100
 ismirsehregal   711.084   913.3475  1144.829  1013.096  1433.427   2316.159   100

Código de referência:

#### library ----

library(dplyr)
library(lubridate)
library(igraph)
library(data.table)
library(microbenchmark)

#### data ----

times_df <- data.frame(
  start = as_datetime(
    c(
      "2019-10-05 14:05:25",
      "2019-10-05 17:30:20",
      "2019-10-05 17:37:00",
      "2019-10-06 04:43:55",
      "2019-10-06 04:53:45",
      "2019-10-06 04:53:46",
      "2019-10-06 04:53:47"
    )
  ),
  stop = as_datetime(
    c(
      "2019-10-05 14:19:20",
      "2019-10-05 17:45:15",
      "2019-10-05 17:50:45",
      "2019-10-06 04:59:00",
      "2019-10-06 05:07:10",
      "2019-10-06 05:07:11",
      "2019-10-06 05:07:12"
    )
  ),
  priority = c(5, 3, 4, 3, 4, 5, 6)
)

times_tib <- as_tibble(times_df)
times_dt <- as.data.table(times_df)

#### group_interval function ----

# buffer to take a form similar to: days(1), weeks(2), etc.
group_interval <- function(start, end, buffer = 0) {

  dat <- tibble(rid = 1:length(start),
                start = start,
                end = end,
                intervals = case_when(!is.na(start) & !is.na(end) ~ interval(start, end),
                                      is.na(start) ~ interval(end, end),
                                      is.na(end) ~ interval(start, start),
                                      TRUE ~ interval(NA, NA)))

  # apply buffer period to intervals
  int_start(dat$intervals) <- int_start(dat$intervals) - buffer + seconds(0.01)
  int_end(dat$intervals) <- int_end(dat$intervals) + buffer - seconds(0.01)

  df_overlap <- bind_cols(
    expand.grid(dat$rid, dat$rid), # make a 2 col table with every combination of id numbers
    expand.grid(dat$intervals, dat$intervals)) %>% # make a combination of every interval
    mutate(overlap = int_overlaps(.data$Var11, .data$Var21)) %>% # determine if intervals overlap
    rename("row" = "Var1", "col" = "Var2")

  # Find groups via graph theory See igraph package
  dat_graph <- graph_from_data_frame(filter(df_overlap, overlap) %>% select(row, col))
  groups <- components(dat_graph)$membership[df_overlap$row]

  # create a 2 column df with row (index) and group number, arrange on row number and return distinct values
  df_groups <- tibble(row = as.integer(names(groups)), group = groups) %>%
    unique()

  # returns
  left_join(select(dat, rid), df_groups, by = c("rid" = "row"))$group

}

#### benchmark ----

library(igraph)
library(data.table)
library(dplyr)
library(lubridate)
library(microbenchmark)

df_Paul <- df_MKa <- df_pgcudahy <- df_ismirsehregal <- times_df <- data.frame(
  start = as_datetime(
    c(
      "2019-10-05 14:05:25",
      "2019-10-05 17:30:20",
      "2019-10-05 17:37:00",
      "2019-10-06 04:43:55",
      "2019-10-06 04:53:45",
      "2019-10-06 04:53:46",
      "2019-10-07 06:00:00",
      "2019-10-07 06:10:00",
      "2019-10-07 06:20:00",
      "2019-10-08 06:00:00",
      "2019-10-08 06:10:00",
      "2019-10-08 06:20:00",
      "2019-10-09 03:00:00",
      "2019-10-09 03:10:00",
      "2019-10-10 03:00:00",
      "2019-10-10 03:10:00",
      "2019-10-11 05:00:00",
      "2019-10-11 05:00:00")
  ),
  stop = as_datetime(
    c(
      "2019-10-05 14:19:20",
      "2019-10-05 17:45:15",
      "2019-10-05 17:50:45",
      "2019-10-06 04:59:00",
      "2019-10-06 05:07:10",
      "2019-10-06 05:07:11",
      "2019-10-07 06:18:00",
      "2019-10-07 06:28:00",
      "2019-10-07 06:38:00",
      "2019-10-08 06:18:00",
      "2019-10-08 06:28:00",
      "2019-10-08 06:38:00",
      "2019-10-09 03:30:00",
      "2019-10-09 03:20:00",
      "2019-10-10 03:30:00",
      "2019-10-10 03:20:00",
      "2019-10-11 05:40:00",
      "2019-10-11 05:40:00")
  ),
  priority = c(5, 3, 4, 3, 4, 5, 4, 3, 4, 3, 4, 3, 1, 2, 2, 1, 3, 4)
)


benchmarks <- microbenchmark(Paul = {
  group_interval <- function(start, end, buffer = 0) {

    dat <- tibble(rid = 1:length(start),
                  start = start,
                  end = end,
                  intervals = case_when(!is.na(start) & !is.na(end) ~ interval(start, end),
                                        is.na(start) ~ interval(end, end),
                                        is.na(end) ~ interval(start, start),
                                        TRUE ~ interval(NA, NA)))

    int_start(dat$intervals) <- int_start(dat$intervals) - buffer + seconds(0.01)
    int_end(dat$intervals) <- int_end(dat$intervals) + buffer - seconds(0.01)

    df_overlap <- bind_cols(
      expand.grid(dat$rid, dat$rid), # make a 2 col table with every combination of id numbers
      expand.grid(dat$intervals, dat$intervals)) %>% # make a combination of every interval
      mutate(overlap = int_overlaps(.data$Var11, .data$Var21)) %>% # determine if intervals overlap
      rename("row" = "Var1", "col" = "Var2")

    dat_graph <- graph_from_data_frame(filter(df_overlap, overlap) %>% select(row, col))
    groups <- components(dat_graph)$membership[df_overlap$row]

    df_groups <- tibble(row = as.integer(names(groups)), group = groups) %>%
      unique()

    left_join(select(dat, rid), df_groups, by = c("rid" = "row"))$group
  }

  times_tib <- as_tibble(df_Paul)

  mutate(times_tib, group = group_interval(start, stop)) %>%
    group_by(group) %>%
    top_n(1, desc(priority)) %>%
    ungroup() %>%
    select(-group)
},
MKa = {
  df_MKa$id <- 1:nrow(df_MKa)

  # Create consolidated df which we will use to check if stop date is in between start and stop
  my_df <- bind_rows(replicate(n = nrow(df_MKa), expr = df_MKa, simplify = FALSE))
  my_df$stop_chk <- rep(df_MKa$stop, each = nrow(df_MKa))

  # Flag if stop date sits in between start and stop
  my_df$chk <- my_df$stop_chk >= my_df$start & my_df$stop_chk <= my_df$stop
  my_df$chk_id <- df_MKa[match(my_df$stop_chk, df_MKa$stop), "id"]

  # Using igrpah to cluster ids to create unique groups
  # this will identify any overlapping groups
  library(igraph)
  g <- graph.data.frame(my_df[my_df$chk == TRUE, c("id", "chk_id")])
  df_g <- data.frame(clusters(g)$membership)
  df_g$chk_id <- row.names(df_g)

  # copy the unique groups to the df
  my_df$new_id <- df_g[match(my_df$chk_id, df_g$chk_id), "clusters.g..membership"]
  my_df %>% 
    filter(chk == TRUE) %>%
    arrange(priority) %>%
    filter(!duplicated(new_id)) %>%
    select(start, stop, priority) %>%
    arrange(start)
}, pgcudahy = {
  df_pgcudahy %>%
    arrange(start) %>%
    mutate(remove1 = ifelse((stop >= lead(start, default=FALSE)) & 
                              (priority > lead(priority, default=(max(priority) + 1))), TRUE, FALSE)) %>%
    mutate(remove2 = ifelse((start <= lag(stop, default=FALSE)) & 
                              (priority > lag(priority, default=(max(priority) + 1))), TRUE, FALSE)) %>%
    filter(remove1 == FALSE & remove2 == FALSE) %>%
    select(1:3)
}, ismirsehregal = {
  setDT(df_ismirsehregal, key="start")[!(stop >= shift(start, type="lead", fill = TRUE) & priority > shift(priority, type="lead", fill = TRUE)) &
                                       !(start <= shift(stop, type="lag", fill = FALSE) & priority > shift(priority, type="lag", fill = TRUE))]
})

benchmarks

Eu tenho uma função auxiliar que agrupa dados sobrepostos / hora usando o pacote igraph (pode incluir um buffer de sobreposição, ou seja, o terminal fica dentro de 1 minuto ...)

Usei-o para agrupar seus dados com base em intervalos no lubridato e, em seguida, fazer algumas disputas de dados para obter apenas a entrada de prioridade mais alta em tempos de sobreposição.

Não tenho certeza de quão bem será dimensionado.

#### library ----

library(dplyr)
library(lubridate)
library(igraph)

#### data ----

times_df <- tibble(start = as_datetime(c("2019-10-05 14:05:25", 
                                         "2019-10-05 17:30:20", 
                                         "2019-10-05 17:37:00", 
                                         "2019-10-06 04:43:55", 
                                         "2019-10-06 04:53:45")), 
                   stop = as_datetime(c("2019-10-05 14:19:20",
                                        "2019-10-05 17:45:15", 
                                        "2019-10-05 17:50:45", 
                                        "2019-10-06 04:59:00",
                                        "2019-10-06 05:07:10")), priority = c(5,3,4,3,4))

#### group_interval function ----

# buffer to take a form similar to: days(1), weeks(2), etc.
group_interval <- function(start, end, buffer = 0) {

  dat <- tibble(rid = 1:length(start),
                start = start,
                end = end,
                intervals = case_when(!is.na(start) & !is.na(end) ~ interval(start, end),
                                      is.na(start) ~ interval(end, end),
                                      is.na(end) ~ interval(start, start),
                                      TRUE ~ interval(NA, NA)))

  # apply buffer period to intervals
  int_start(dat$intervals) <- int_start(dat$intervals) - buffer + seconds(0.01)
  int_end(dat$intervals) <- int_end(dat$intervals) + buffer - seconds(0.01)

  df_overlap <- bind_cols(
    expand.grid(dat$rid, dat$rid), # make a 2 col table with every combination of id numbers
    expand.grid(dat$intervals, dat$intervals)) %>% # make a combination of every interval
    mutate(overlap = int_overlaps(.data$Var11, .data$Var21)) %>% # determine if intervals overlap
    rename("row" = "Var1", "col" = "Var2")

  # Find groups via graph theory See igraph package
  dat_graph <- graph_from_data_frame(filter(df_overlap, overlap) %>% select(row, col))
  groups <- components(dat_graph)$membership[df_overlap$row]

  # create a 2 column df with row (index) and group number, arrange on row number and return distinct values
  df_groups <- tibble(row = as.integer(names(groups)), group = groups) %>%
    unique()

  # returns
  left_join(select(dat, rid), df_groups, by = c("rid" = "row"))$group

}

#### data munging ----

mutate(times_df, group = group_interval(start, stop)) %>%
  group_by(group) %>%
  top_n(1, desc(priority)) %>% # not sure why desc is needed, but top_n was giving the lower 
  ungroup() %>%
  select(-group)

Que dá:

    # A tibble: 3 x 3
      start               stop                priority
      <dttm>              <dttm>                 <dbl>
    1 2019-10-05 14:05:25 2019-10-05 14:19:20        5
    2 2019-10-05 17:30:20 2019-10-05 17:45:15        3
    3 2019-10-06 04:43:55 2019-10-06 04:59:00        3

Desci uma toca de coelho olhando para árvores de intervalo (e implementações R como IRanges / plyranges), mas acho que esse problema não precisa de uma estrutura de dados tão envolvida, pois os horários de início podem ser facilmente classificados. Também ampliei o conjunto de testes como @ismirsehregal para abranger mais relações de intervalo em potencial , como um intervalo que começa antes e termina após o vizinho ou quando três intervalos se sobrepõem, mas o primeiro e o último não se sobrepõem, ou dois intervalos que começam e pare exatamente nos mesmos horários.

library(lubridate)
times_df <- data.frame(
  start = as_datetime(
    c(
      "2019-10-05 14:05:25",
      "2019-10-05 17:30:20",
      "2019-10-05 17:37:00",
      "2019-10-06 04:43:55",
      "2019-10-06 04:53:45",
      "2019-10-06 04:53:46",
      "2019-10-07 06:00:00",
      "2019-10-07 06:10:00",
      "2019-10-07 06:20:00",
      "2019-10-08 06:00:00",
      "2019-10-08 06:10:00",
      "2019-10-08 06:20:00",
      "2019-10-09 03:00:00",
      "2019-10-09 03:10:00",
      "2019-10-10 03:00:00",
      "2019-10-10 03:10:00",
      "2019-10-11 05:00:00",
      "2019-10-11 05:00:00")
  ),
  stop = as_datetime(
    c(
      "2019-10-05 14:19:20",
      "2019-10-05 17:45:15",
      "2019-10-05 17:50:45",
      "2019-10-06 04:59:00",
      "2019-10-06 05:07:10",
      "2019-10-06 05:07:11",
      "2019-10-07 06:18:00",
      "2019-10-07 06:28:00",
      "2019-10-07 06:38:00",
      "2019-10-08 06:18:00",
      "2019-10-08 06:28:00",
      "2019-10-08 06:38:00",
      "2019-10-09 03:30:00",
      "2019-10-09 03:20:00",
      "2019-10-10 03:30:00",
      "2019-10-10 03:20:00",
      "2019-10-11 05:40:00",
      "2019-10-11 05:40:00")
  ),
  priority = c(5, 3, 4, 3, 4, 5, 4, 3, 4, 3, 4, 3, 1, 2, 2, 1, 3, 4)
)

Em seguida, faço duas passagens por cada intervalo para ver se ele se sobrepõe ao seu antecessor ou sucessor

stop >= lead(start, default=FALSE) e start <= lag(stop, default=FALSE))

Durante cada passagem, há uma segunda verificação para ver se a prioridade do intervalo tem um valor numérico mais alto que o predecessor ou sucessor priority > lead(priority, default=(max(priority) + 1)). Durante cada passagem, se as duas condições forem verdadeiras, um sinalizador "remover" será definido como verdadeiro em uma nova coluna usando mutate. Todas as linhas com um sinalizador de remoção são filtradas.

library(tidyverse)
times_df %>%
    arrange(start) %>%
    mutate(remove1 = ifelse((stop >= lead(start, default=FALSE)) & 
                            (priority > lead(priority, default=(max(priority) + 1))), 
                            TRUE, FALSE)) %>%
    mutate(remove2 = ifelse((start <= lag(stop, default=FALSE)) & 
                            (priority > lag(priority, default=(max(priority) + 1))), 
                            TRUE, FALSE)) %>%
    filter(remove1 == FALSE & remove2 == FALSE) %>%
    select(1:3)

Isso evita a verificação de todas as combinações potenciais de intervalos, como a resposta de @ Paul (comparações 2n versus n!), Além de acomodar minha ignorância sobre a teoria dos grafos :)

Da mesma forma, a resposta de @ ismirsehregal tem magia data.table que está além da minha compreensão.

A solução do @ MKa parece não funcionar com> 2 períodos sobrepostos

Testar as soluções fornece

#>          expr       min        lq      mean    median        uq       max
#> 1 dplyr_igraph 36.568842 41.510950 46.692147 43.362724 47.065277 241.92073
#> 2  data.table  9.126385  9.935049 11.395977 10.521032 11.446257  34.26953
#> 3       dplyr  5.031397  5.500363  6.224059  5.902589  6.373197  15.09273
#>   neval
#> 1   100
#> 2   100
#> 3   100

Deste código

library(igraph)
library(data.table)
library(microbenchmark)
benchmarks <- microbenchmark(dplyr_igraph = {
  group_interval <- function(start, end, buffer = 0) {

  dat <- tibble(rid = 1:length(start),
                start = start,
                end = end,
                intervals = case_when(!is.na(start) & !is.na(end) ~ interval(start, end),
                                      is.na(start) ~ interval(end, end),
                                      is.na(end) ~ interval(start, start),
                                      TRUE ~ interval(NA, NA)))

  int_start(dat$intervals) <- int_start(dat$intervals) - buffer + seconds(0.01)
  int_end(dat$intervals) <- int_end(dat$intervals) + buffer - seconds(0.01)

  df_overlap <- bind_cols(
    expand.grid(dat$rid, dat$rid), # make a 2 col table with every combination of id numbers
    expand.grid(dat$intervals, dat$intervals)) %>% # make a combination of every interval
    mutate(overlap = int_overlaps(.data$Var11, .data$Var21)) %>% # determine if intervals overlap
    rename("row" = "Var1", "col" = "Var2")

  dat_graph <- graph_from_data_frame(filter(df_overlap, overlap) %>% select(row, col))
  groups <- components(dat_graph)$membership[df_overlap$row]

  df_groups <- tibble(row = as.integer(names(groups)), group = groups) %>%
    unique()

  left_join(select(dat, rid), df_groups, by = c("rid" = "row"))$group
  }

  times_tib <- as_tibble(times_df)

  mutate(times_tib, group = group_interval(start, stop)) %>%
    group_by(group) %>%
    top_n(1, desc(priority)) %>%
    ungroup() %>%
    select(-group)
}, data.table = {
  times_dt <- as.data.table(times_df)
  setkey(times_dt, start, stop)[, index := .I]
  overlaps_dt <- foverlaps(times_dt, times_dt, type = "any", which = TRUE)[xid != yid][, group := fifelse(xid > yid, yes = paste0(yid, "_", xid), no = paste0(xid, "_", yid))]
  overlaps_merged <- merge(times_dt, overlaps_dt, by.x = "index", by.y = "xid")[, .(delete_index = index[priority == max(priority)]), by = "group"]
  result_dt <- times_dt[!unique(overlaps_merged$delete_index)][, index := NULL]
}, dplyr = {
times_df %>%
    arrange(start) %>%
    mutate(remove1 = ifelse((stop >= lead(start, default=FALSE)) & 
                            (priority > lead(priority, default=(max(priority) + 1))), TRUE, FALSE)) %>%
    mutate(remove2 = ifelse((start <= lag(stop, default=FALSE)) & 
                            (priority > lag(priority, default=(max(priority) + 1))), TRUE, FALSE)) %>%
    filter(remove1 == FALSE & remove2 == FALSE) %>%
    select(1:3)
})
summary(benchmarks)

Também usando igraphpara identificar grupos sobrepostos, você pode tentar:

library(tidyverse)
library(lubridate)
times_df <- data.frame(
  start = as_datetime(
    c(
      "2019-10-05 14:05:25",
      "2019-10-05 17:30:20",
      "2019-10-05 17:37:00",
      "2019-10-06 04:43:55",
      "2019-10-06 04:53:45",
      "2019-10-06 04:53:46",
      "2019-10-07 06:00:00",
      "2019-10-07 06:10:00",
      "2019-10-07 06:20:00",
      "2019-10-08 06:00:00",
      "2019-10-08 06:10:00",
      "2019-10-08 06:20:00",
      "2019-10-09 03:00:00",
      "2019-10-09 03:10:00",
      "2019-10-10 03:00:00",
      "2019-10-10 03:10:00",
      "2019-10-11 05:00:00",
      "2019-10-11 05:00:00")
  ),
  stop = as_datetime(
    c(
      "2019-10-05 14:19:20",
      "2019-10-05 17:45:15",
      "2019-10-05 17:50:45",
      "2019-10-06 04:59:00",
      "2019-10-06 05:07:10",
      "2019-10-06 05:07:11",
      "2019-10-07 06:18:00",
      "2019-10-07 06:28:00",
      "2019-10-07 06:38:00",
      "2019-10-08 06:18:00",
      "2019-10-08 06:28:00",
      "2019-10-08 06:38:00",
      "2019-10-09 03:30:00",
      "2019-10-09 03:20:00",
      "2019-10-10 03:30:00",
      "2019-10-10 03:20:00",
      "2019-10-11 05:40:00",
      "2019-10-11 05:40:00")
  ),
  priority = c(5, 3, 4, 3, 4, 5, 4, 3, 4, 3, 4, 3, 1, 2, 2, 1, 3, 4)
)
times_df$id <- 1:nrow(times_df)


# Create consolidated df which we will use to check if stop date is in between start and stop
my_df <- bind_rows(replicate(n = nrow(times_df), expr = times_df, simplify = FALSE))
my_df$stop_chk <- rep(times_df$stop, each = nrow(times_df))

# Flag if stop date sits in between start and stop
my_df$chk <- my_df$stop_chk >= my_df$start & my_df$stop_chk <= my_df$stop
my_df$chk_id <- times_df[match(my_df$stop_chk, times_df$stop), "id"]

# Using igrpah to cluster ids to create unique groups
# this will identify any overlapping groups
library(igraph)
g <- graph.data.frame(my_df[my_df$chk == TRUE, c("id", "chk_id")])
df_g <- data.frame(clusters(g)$membership)
df_g$chk_id <- row.names(df_g)

# copy the unique groups to the df
my_df$new_id <- df_g[match(my_df$chk_id, df_g$chk_id), "clusters.g..membership"]
my_df %>% 
  filter(chk == TRUE) %>%
  arrange(priority) %>%
  filter(!duplicated(new_id)) %>%
  select(start, stop, priority) %>%
  arrange(start)