train_ppgrid: Train prior period information.

Description Usage Arguments Value Examples

Description

Produces a grid of weighted mean and total weight for the preceding period of each group-time combination. All parameters are taken as vectors that are assumed to be associated and in the same order. train_ppgrid will compute ~ granularity * (max(time_var) - min(time_var) - window_size) distinct aggregations for each distinct id_var.

Usage

1
2
train_ppgrid(id_var, time_var, response_var, wt_var = rep(1,
  length(response_var)), lag, window_size, granularity)

Arguments

id_var

Vector identifying the groups.

time_var

Numeric vector of time, there is no support for character time vectors or other datetime datatypes yet.

response_var

Numeric vector of responses to summarise.

wt_var

Optional numeric weight vector. Defaults to 1.0.

lag

Number of units of time to lag the variable accumulation before scoring. Useful if there is a period after data acquisition before the response is known; this parameter can mimic model implementation.

window_size

Number of units of time to aggregate the response and weight variables over.

granularity

How severely to round the input time variable. Times will be rounded to the nearest granularity. Improves runtime at nominal cost to accuracy.

Value

ppgrid object, containing a data.table containing prior period information as well as some input parameters.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
set.seed(1234)
ppdf <- make_longitudinal_data(1000)

ppgrid <- train_ppgrid(ppdf$id,
                       ppdf$date,
                       ppdf$resp,
                       lag = 25,
                       window_size = 25,
                       granularity = 25)

print(ppgrid)

DexGroves/ppR documentation built on May 6, 2019, 2:13 p.m.