plotly linear trend line not updating in shiny - r

Question

I am trying to add a linear trend line to a plotly plot in a shiny app. When I change the selection parameters, I see that the coefficients of the linear model change (using observe(print(summary(l)). However, the actual line on the plot seems to stay in the same place.
Here is one plot, where the trend line at least seems close to intersecting the two points:
In another plot, the trend line is nowhere near the first point:
Here is a minimal working example:
library(dplyr)
library(shiny)
library(plotly)
df <- as.data.frame(list("UserID"=c(1,1,1,1,2,2,2,2),
"QuestionID"=c(4,4,5,5,4,4,6,6),
"KeystrokeRate"=c(8,4,6,15,8,6,7,8),
"cumul.ans.keystroke"=c(1,7,1,5,1,14,1,9),
"Relative.Time.Progress"=c(0.1,1.0,0.4,1.0,0.8,1.0,0.8,1.0)
))
ui <- (fluidPage(
sidebarLayout(
sidebarPanel(
selectInput("userInput","Select User", sort(unique(df$UserID)),
selected = sort(unique(df$UserID))[1]),
uiOutput("answerOutput")#,
),
mainPanel(
plotlyOutput("mainPlot")#,
)
)
))
server <- function(input, output, session) {
# filter only based on selected user
filteredForUser <- reactive({
try(
df %>%
filter(
UserID == input$userInput
), silent=T)
})
# filter for both user and answer
filteredFull <- reactive({
try (
df %>%
filter(
UserID == input$userInput,
QuestionID == input$answerInput
), silent=T)
})
# filter answer choices based on user
output$answerOutput <- renderUI({
df.u <- filteredForUser()
if(!is.null(df)) {
selectInput("answerInput", "Select A Typing Session",
sort(unique(df.u$QuestionID)))
}
})
output$mainPlot <- renderPlotly({
if (class(filteredForUser()) == "try-error" ||
class(filteredFull()) == "try-error") {
return(geom_blank())
} else {
# plot scatter points and add trend lines
p <- plot_ly(filteredFull(), x=Relative.Time.Progress, y=cumul.ans.keystroke,
mode='markers', color=KeystrokeRate, size=KeystrokeRate,
marker=list(sizeref=100), type='scatter')
l <- lm(cumul.ans.keystroke ~ Relative.Time.Progress,
data=filteredFull())
observe(print(summary(l)))
p <- add_trace(p, y= fitted(l))
p
}
})
}
shinyApp(ui, server)

Answer 1

The problem is with the add_trace functionality. You need to provide it the x-axis to be able to correctly plot the lm result.
p <- add_trace(p, y = fitted(l), x = Relative.Time.Progress)
To see the problem more clearly, visualize the results with the entire dataset.
p <- plot_ly(df, x=Relative.Time.Progress, y=cumul.ans.keystroke,
mode='markers', color=KeystrokeRate, size=KeystrokeRate,
marker=list(sizeref=100), type='scatter')
p
l <- lm(cumul.ans.keystroke ~ Relative.Time.Progress,
data=df)
p <- add_trace(p, y = fitted(l))
p
p <- plot_ly(df, x=Relative.Time.Progress, y=cumul.ans.keystroke,
mode='markers', color=KeystrokeRate, size=KeystrokeRate,
marker=list(sizeref=100), type='scatter')
l <- lm(cumul.ans.keystroke ~ Relative.Time.Progress,
data=df)
p <- add_trace(p, y = fitted(l), x = Relative.Time.Progress)
p
So as you would see, add_trace was plotting the fitted(y) correctly but using x-axis to be c(0:7). I am guessing it is a default passed to add_trace, but I haven't looked deeply of the 'why' of this. The dataset df has eight points. You instead needed to give the actual Relative.Time.Progress values on the x-axis to correctly plot the fitted(y) w.r.t. actual x values. Hope this clarifies.