Cumulative History Analysis for Bistable Perception Time Series.

Cumulative History Analysis For Bistable Perception Time Series

A package to compute a cumulative history for time-series of perceptual dominance in bistable displays.

Estimates cumulative history, an estimate of accumulating adaptation/prediction error for the dominant percept, for time-series for continuously viewed bistable perceptual rivalry displays. Computes cumulative history via a homogeneous first order differential process. I.e., it assumes exponential growth/decay of the history as a function of time and perceptually dominant state. Supports Gamma, log normal, and normal distribution families.

If you use the toolbox in your work, please cite Pastukhov, A., (2022). bistablehistory: an R package for history-dependent analysis of perceptual time series. Journal of Open Source Software, 7(70), 3901, https://doi.org/10.21105/joss.03901

Installation

For current stable version use

install.packages("bistablehistory")

The master branch is the development version. To install it please use

library("devtools")
install_github("alexander-pastukhov/bistablehistory", dependencies = TRUE)

Note

This package uses Stan, a "state-of-the-art platform for statistical modeling and high-performance statistical computation". Therefore, it depends on the package rstantools, which in turn depends on the rstan package, which uses the V8 JavaScript library, through the V8 R package.

Therefore, you will need to install the V8 JavaScript library on your system, and it is recommended that you also install the V8 R package beforehand. For detailed instructions, please see https://github.com/jeroen/v8.

You will also need the R package curl, which depends on libcurl-* in various operating systems. Please see the documentation at https://cran.r-project.org/package=curl.

Usage

The main function is fit_cumhist that takes a data frame with time-series as the first argument. Minimally, you need to specify state --- string with the column name that encodes perceptually dominant state --- and either duration (column name with duration of individual dominance phases) or onset (column name with onset times of individual dominance phases). Thus, for a simplest case of a single subject and single run/block measurement with all defaults (gamma distribution, fitted cumulative history time constant but fixed mixed state value and history mixing proportion) the call would be

library(bistablehistory)
data(br_singleblock)
gamma_fit <- fit_cumhist(br_singleblock,
                         state = "State",
                         duration = "Duration")

or, equivalently

library(bistablehistory)
data(br_singleblock)
gamma_fit <- fit_cumhist(br_singleblock,
                         state = "State",
                         onset = "Time")

Now you can look at the fitted value for history time constant via

history_tau(gamma_fit)

and main effect of history for both parameters of gamma distribution

coef(gamma_fit)

For further details please see vignettes on package usage (Usage examples and Cumulative history) and on an example of writing Stan code directly (Writing Stan code).