deseasonalized_trend: deseasonalized_trend

In bandit: Functions for Simple a/B Split Test and Multi-Armed Bandit Analysis

deseasonalized_trend

Description

A convenience function to analyze a timeseries and return an estimate (via gam, using day of week factors and smoothed timestamp) of whether, after accounting for day-of-week, there is a significant time-based influence and what that influence is.

Usage

deseasonalized_trend(df, w=NULL)

Arguments

df	a data frame containing timestamp and value entries
w	number of attempts (n for binomial data)

Value

a list with the following items:

pval	pval given by anova on gam, to indicate whether s(timestamp) is significant
smoothed_prediction	a smoothed prediction over time (on Wednesdays), to give a human-understandable idea of what the change over time has been

Author(s)

Thomas Lotze <[e-mail]>

Examples

timestamps = as.numeric(as.POSIXct(seq(as.Date("2012-01-01"),as.Date("2012-05-03"),by=1)))
df=data.frame(timestamp = timestamps, value = rnorm(length(timestamps)))
dt = deseasonalized_trend(df)
if (dt$pval < 0.01) {
  print("Significant time-based factor")
  plot(df$timestamp, dt$smoothed_prediction)
} else {
  print("No significant time-based factor")
}

df=data.frame(timestamp = timestamps,
              value = sapply(timestamps, function(t) {rpois(1, lambda=t-min(timestamps))}))
dt = deseasonalized_trend(df)
if (dt$pval < 0.01) {
  print("Significant time-based factor")
  plot(df$timestamp, dt$smoothed_prediction)
} else {
  print("No significant time-based factor")
}

bandit documentation built on June 29, 2022, 5:05 p.m.