read-tv is a shiny application and associated console interface designed for visualizing changepoints in irregularly and regularly spaced longitudinal data. Technical users can rapidly view data through the R console interface, and non-technical users can use it as a standard web application. Intermediate users, as well as users who want to replicate plots, will benefit from the source code generation associated with any plot.
This application is built on top of a wide array of packages, and especially wraps functionality from the changepoint package and tidyverse set of packages.
Below is a plot of new COVID-19 cases in the United States with detected
changepoints.
This software has been tested with R Version 3.6.
This package is not on CRAN. Please install from GitHub with the install_github function from the Devtools package.
# Install devtools if not yet installed
install.packages("devtools")
devtools::install_github("JDMusc/READ-TV")
There is also a Docker option.
Please note that since the docker image is a light version of Ubuntu, it will not run if the host machine (your local computer) is Windows.
There is a tutorial README.
A data frame or RDS/CSV/TSV file where each row is an observation. If the data does not have columns for Case, Event.Type, or Time, then read-tv will open a pop-up that enables the user to map Case and Event.Type to another column (or mock value), and map Time to a column.
read-tv exports 3 data sets:
* Global COVID data, from Our World in
Data
* US state-based COVID data, from NY
Times
* USGS seismic records from Japan on
March 11, 2011.
A sample CSV data file that works with read-tv can be found here.
The main function is launchReadtv.
launchReadtv()
Or it can be launched as
app = readtv::launchReadtv()
shiny::runApp(app)
Both files and in-memory objects can be passed to launchReadtv
library(dplyr)
library(magrittr)
library(readtv)
app = readtv::japan_eq_3_11 %>%
mutate(Time = time, Event.Type = place, Case = 1) %>% #bypass popup
launchReadtv(plotOpts = tvOpts(y = mag, x = Time, color = place))
#shiny::runApp(app) #uncomment to run application
Executing the above shiny::runApp(app)
code will generate the
following plot. The plotOpts
argument and the tvOpts
function tells
the plot to load with the specified axes, colors, etc.
If the code is run without the mutate
call, then a pop-up first
requests the user to map the Case
, Time
, and Event.Type
columns.
app = readtv::japan_eq_3_11 %>%
launchReadtv(plotOpts = tvOpts(y = mag, x = Time, color = place))
#shiny::runApp(app) #uncomment to run application
The Filter & Facet
tab will show the same plot after the pop-up is
submitted.
read-tv exports 5 functions:
launchReadtv
- launch the application
tvOpts
- tells the plot to load with the specified axes, colors, etc.
- used with
launchReadtv
โsplotOpts
argument
preprocessForCpa
- regularly spaces and smooths time series data
- used in the
CPA
tabs - how to use visible from
Source Code
sub-tab.
slidingWindow
- same sub-points as
preprocessForCpa
.
- same sub-points as
generatePlotDefaults
- user can run
generatePlotDefaults()
to view defaults fortvOpts
- user can run
addCpaMarkersToPlot
- user can add cpa markers to a gg-plot.
- used in the
CPA
tabs, after CPA is calculated - how to use visible from
Source Code
sub-tab.
Each of these functions has associated help documentation.
The CPA tab has features to regularly space the data (preprocess), and
then execute methods from the changepoints package. It spaces the data
with the preprocessForCpa
function and slidingWindow
function.
To avoid the column mapping pop up, we can use dplyr to specify the
columns and mock values (Case) before it is passed into launchReadtv
.
library(dplyr, warn.conflicts = FALSE)
library(magrittr)
library(readtv)
app = readtv::japan_eq_3_11 %>%
mutate(Time = time, Event.Type = place, Case = 1) %>%
launchReadtv(plotOpts = tvOpts(y = mag, x = Time, color = place))
#shiny::runApp(app) # and move to CPA tab after viewing in Basic display tab
Below is a screen shot from read-tv. Note that the time points are regularized with values interpolated through a sliding window. The timing intervals were not originally evenly spaced (irregular) since it is earthquake event data. The CPA tab has functionality to create regular spacing.
Sometimes it helps to facet and paginate the data.
app = readtv::covid_usa %>%
mutate(Time = date, Case = state, Event.Type = state) %>%
launchReadtv(plotOpts =
tvOpts(y = cases, facetOn = state, isFacetPaginated = TRUE,
facetRowsPerPage = 3))
#shiny::runApp(app) #un comment to run application
And the source code can also be viewed from the app. Please note that
it assumes tidyverse
, changepoint
, ggforce
, and readtv
packages
are attached.