R Package for Measuring and Comparing Time-series Data with Bayesian Prediction Model
This R package implements an approach to integrate multivariate testing and a Bayesian time series prediction model to measure casual inference based on CausalImpact package . The benefit offering processes have suffered from a lack of effective methods due to insufficient customer data, a wide variety of benefits, and most of all, requiring predictions of customer reactions.
Given control time series (e.g. user visits in each group) and a response time series (e.g., total revenue), the package constructs Bayesian structural time-series models for each control time series. And using given event time as intevention, these models show predictions and differences between time series. i.e. what group is the activist for a promotion.
Main Function: CompareImpact() Sub-Functions: CreateCompImpPlot()
To install CompareImpact, type the following commands into an R session:
## install from github
devtools::install_github("google/Causalmpact")
devtools::install_github("cojette/CompareImpact")
CausalImpact package install is necessary from Github.
Once installed, the package can be loaded in a given R session using:
library(CompareImpact)
- Multiple user segments To estimate a causal effect of each user segment, we begin by specifying which period in the data should be used for training the model (pre-intervention period) and which period for computing a counterfactual prediction (post-intervention period).
data(promo_data)
pre.period <- c(1, 57)
post.period <- c(58, 90)
This says that time points 1 ... 57 will be used for training, and time points 58 ... 100 will be used for computing predictions. There is a big promotion at time points 57
To perform inference, we run the analysis using:
ci1 <- CompareImpact(promo_data[,-1], c(1,57), c(58,90))
## check impact data
ci1
The return value is a list of CausalImpact objects. You can see the detail of each CausalImpact objects with CausalImpact package functions and you may use summary, plot functions in CausalImpact package individualy.
The easiest way of visualizing the results is to use the CreateCompImpPlot() function that is part of the package:
CreateCompImpPlot(ci1)
You can control details with ggplot2 package functions. You can refer plot() of CausalImpact package for the detailed interpretation of the plot.
CompareImpact package will be updated. And I would appreciate to discuss any aspects of this work (cojette@gmail.com).