bin_data {mpathsenser}

Create bins in variable time series

Description

In time series with variable measurements, an often recurring task is calculating the total time spent (i.e. the duration) in fixed bins, for example per hour or day. However, this may be difficult when two subsequent measurements are in different bins or span over multiple bins.

Usage

bin_data(
  data,
  start_time,
  end_time,
  by = c("sec", "min", "hour", "day"),
  fixed = TRUE
)

Arguments

data

A data frame or tibble containing the time series.

start_time

The column name of the start time of the interval, a POSIXt.

end_time

The column name of the end time of the interval, a POSIXt.

by

A binning specification.

fixed

Whether to create fixed bins. If TRUE, bins will be rounded to, for example, whole hours or days (depending on by). If FALSE, bins will be created based on the first timestamp.

Value

A tibble containing the group columns (if any), date, hour (if by = "hour"), and the duration in seconds.

See Also

link_gaps() for linking gaps to data.

Examples

library(dplyr)

data <- tibble(
  participant_id = 1,
  datetime = c(
    "2022-06-21 15:00:00", "2022-06-21 15:55:00",
    "2022-06-21 17:05:00", "2022-06-21 17:10:00"
  ),
  confidence = 100,
  type = "WALKING"
)

# get bins per hour, even if the interval is longer than one hour
data |>
  mutate(datetime = as.POSIXct(datetime)) |>
  mutate(lead = lead(datetime)) |>
  bin_data(
    start_time = datetime,
    end_time = lead,
    by = "hour"
  )

# Alternatively, you can give an integer value to by to create custom-sized
# bins, but only if fixed = FALSE. Not that these bins are not rounded to,
# as in this example 30 minutes, but rather depends on the earliest time
# in the group.
data |>
  mutate(datetime = as.POSIXct(datetime)) |>
  mutate(lead = lead(datetime)) |>
  bin_data(
    start_time = datetime,
    end_time = lead,
    by = 1800L,
    fixed = FALSE
  )

# More complicated data for showcasing grouping:
data <- tibble(
  participant_id = 1,
  datetime = c(
    "2022-06-21 15:00:00", "2022-06-21 15:55:00",
    "2022-06-21 17:05:00", "2022-06-21 17:10:00"
  ),
  confidence = 100,
  type = c("STILL", "WALKING", "STILL", "WALKING")
)

# binned_intervals also takes into account the prior grouping structure
out <- data |>
  mutate(datetime = as.POSIXct(datetime)) |>
  group_by(participant_id) |>
  mutate(lead = lead(datetime)) |>
  group_by(participant_id, type) |>
  bin_data(
    start_time = datetime,
    end_time = lead,
    by = "hour"
  )
print(out)

# To get the duration for each bin (note to change the variable names in sum):
purrr::map_dbl(
  out$bin_data,
  ~ sum(as.double(.x$lead) - as.double(.x$datetime),
    na.rm = TRUE
  )
)

# Or:
out |>
  tidyr::unnest(bin_data, keep_empty = TRUE) |>
  mutate(duration = .data$lead - .data$datetime) |>
  group_by(bin, .add = TRUE) |>
  summarise(duration = sum(.data$duration, na.rm = TRUE), .groups = "drop")