tv {tv}

Create a time-varying dataset

Description

Create a time-varying dataset

Usage

time_varying(
  x,
  specs,
  exposure,
  ...,
  grid.only = FALSE,
  time_units = c("days", "seconds"),
  id = "pat_id",
  sort = NA,
  n_cores = as.numeric(Sys.getenv("SLURM_CPUS_PER_TASK", 1))
)

check_tv_data(x, time_units, id, sort)

check_tv_exposure(x, expected_ids, time_units, id, ..., check_overlap = TRUE)

check_tv_specs(specs, expected_features = NULL)

Arguments

x

A data.frame with four columns: <id>, "feature", "datetime", "value"

specs

a data.frame with four columns: "feature", "use_for_grid", "lookback_start", "lookback_end", "aggregation". See details below.

exposure

a data.frame with (at least) three columns: <id>, "exposure_start", "exposure_stop"

...

Other arguments. Currently just passes check_overlap.

grid.only

Should just the grid be computed and returned? Useful only for debugging

time_units

What time units should be used? Seconds or days

id

The id to use. Default is "pat_id"

sort

Logical, indicating whether to sort the data before performing the analysis. By default (NA), sorting is only done when useful (that is: x$datetime is a POSIXct and time_units == "days"). A warning is issued when x$datetime is a Date to make the user aware that the input ought to be sorted to get the right answer.

n_cores

Number of cores to use. If slurm is being used, it checks the SLURM_CPUS_PER_TASK variable. Else it defaults to 1, for no parallelization.

expected_ids

A vector of expected ids based on the data.

check_overlap

Should overlap be checked among exposure rows? A potentially costly operation, so you can opt out of it if you're really sure.

expected_features

A vector of expected features based on the data.

Details

The defaults for specs are to use everything for the grid creation, and to set lookback_start=0, with a message in both cases. Currently supported aggregation functions include counting ("count" or "n"), last-value-carried forward ("last value" or "lvcf"), any/none ("any" or "binary"), time since ("time since" or "ts"), min/max/mean, and the special "event" (for which look backs are ignored).

The look back window begins at row_start - lookback_end and ends at row_start - lookback_start. Passing NA to either look back changes the corresponding window boundary to exposure_start.

Value

A data.frame, with one row per grid value and one column per feature specification (plus grid columns).

Examples

  data(tv_example)
  time_varying(tv_example$data, tv_example$specs, tv_example$exposure,
               time_units = "days", id = "mcn")