not able to get forecasting plot in tabset panel in flexdashboard using actionButton and eventReactive methods of shiny

Question

I have a dataframe (in R) containing the data which looks like this:

[https://github.com/imlemarque/Data/raw/main/Data.csv][1]

I am creating a flexdashboard in R and the code is as follows:

---
title: "Forecast"
output: 
  flexdashboard::flex_dashboard:
    orientation: rows
    vertical_layout: fill
    theme:
      bg: '#FFFFFF'
      fg: '#2c3e50'
      primary: '#18bc9c'
runtime: shiny

---

```{r global, include=FALSE, warning=FALSE, message=FALSE}
# libraries
library(flexdashboard)
library(readr)       
library(tidyverse)    
library(ggridges)      
library(shinydashboardPlus)
library(shiny)
library(dplyr)
library(lubridate)
library(tidyr)
library(padr)
library(shinythemes)
library(rio)
library(shinyjs)
library(forecast)

# pre-set the bw theme.
theme_set(theme_bw())
  
# convert to proper date
df$Date1 <- df$Date
df$Date <- as.Date(df$Date, format="%m/%d/%Y")

Var <- setdiff(names(select_if(df, is.numeric)), c("Date", "CovidPeriod"))


```

Forecasting {data-icon="fa-signal"}
=======================================================================

## Row {.sidebar data-width=280 data-padding=10}

```{r}
br()
selectInput("Series2", label = "Select Series:", choices = Var, multiple = FALSE, width = "98%")
checkboxInput("covid1", "Exclude covid period?", FALSE)
output$value1 <- renderText({ input$covid1 })
hr()


# Select option for confidence interval
selectInput(inputId = "conf_int", label = "Confidence Interval:",
            choice = c("99%" = "99", "95%" = "95", "80%" = "80"), multiple = FALSE, width = "98%")
hr()

sliderInput("Horiz", "Forecast Horizon:", min = 1, max = 52, value = 13)
sliderInput("HoldOut", "Holdout Period:", value = 52, min = 24, max = 104, step = 4 )
hr()


actionButton(inputId="goButton", "Start forecasting!", class = "btn-primary btn-lg")


df2 <- reactive({
        newdf <- df
        if(input$covid){ 
            newdf <- filter(df, CovidPeriod != 0)
        }
        newdf
})



```


Row {data-height=160}
-----------------------------------------------------------------------

### KPI1 {.value-box}

```{r}

renderValueBox({
valueBox(round(0.7457475, digits = 3), icon = "fa-check-square", caption = "Model Accuracy", color = "#95d7ae")
})

```

### KPI2  {.value-box}

```{r}

renderValueBox({
valueBox(round(0.34535435, digits = 3), icon = "fa-check-square", caption = "Misclassification rate", color = "primary")
})

```

### KPI3 {.value-box}

```{r}

renderValueBox({
valueBox(round(0.858656, digits = 3), icon = "fa-check-square", caption = "Recall", color = "primary")
})
```

Row {data-height=850}
-----------------------------------------------------------------------------

### 

```{r}

# Create reactive Date Column in format YYYY-MM-DD
sliderValues2 <- reactive({

  df0 <- df2() %>% mutate(Year = as.integer(year(Date)), 
                          Month = as.integer(month(Date)), 
                          Day = as.integer(day(Date))
                          )
  df0$Date2 <- strftime(paste(df0$Year, df0$Month, df0$Day, sep="-"), "%Y-%m-%d")
  df0 %>% select(Date = "Date2", input$Series2)

  })

# Render data reactively as Table
ARIMA <- eventReactive(input$goButton, {

  if (nrow(sliderValues2())==0)
    return()
  
  sliderValues2()

})


tabsetPanel(type = "tabs",
            tabPanel("ARIMA", icon = icon("area-chart"), renderTable({ARIMA()}), value=1),
            tabPanel("ARIMA2", icon = icon("area-chart"), plotOutput("ARIMA2", width="100%", height="100%"), value=1),
            id = "timeSeriesTabs")




```

My problem is that I am not able to generate forecast using Arima model to show in the ARIMA2, nor I am able to generate final table that should be shown in ARIMA tab panel.

I am getting Error:

argument is of length zero

Could someone please help in:

1. Run the fresh Arima model, on the data reactively gathered and when clicked on Go Button
2. Getting ARIMA model plots in ARIMA tab?
3. Model performance metrics in another tab (please create it)

Answer 1

Seems to work fine if you comment out the themes part:

---
title: "Forecast"
output: 
  flexdashboard::flex_dashboard:
    #theme:
    #  bg: '#FFFFFF'
    #  fg: '#2C3E50'
    #  primary: '#18BC9C'
    orientation: rows
    vertical_layout: fill    
runtime: shiny
---   

```{r global, include=FALSE, warning=FALSE, message=FALSE}

# ...
# assign df to your structure to start with, i.e.,
# df <- structure(list(Date = c("1/1/2010", "1/8/2010", "1/15/2010", 
#    "1/22/2010", "1/29/2010", "2/5/2010", "2/12/2010", "2/19/2010", ...
#     ...   ...   ...
#     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
#     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0))) #, class = "col_spec")

# sapply(df, length)
#   Date        Var1        Var2        Var3        Var4 CovidPeriod 
#    592         554         543         482         592         448 

df <- as.data.frame(lapply(df, function(x) x[1:448])) # make sure all the columns have same number of rows
# ...

```

with the following output:

You may try to investigate more on why themes part is not working by reinstalling bslib etc.,

remotes::install_github("rstudio/bslib")

For forecasting with ARIMA:

library(lubridate)
library(forecast)
start <- 1
end <- 105
n <- nrow(df)
npred <- 20
var <- 'Var1'
tsdata <- ts(df[,var], start=decimal_date(ymd(df$Date[start])), end=decimal_date(ymd(df$Date[end])), frequency=365/7)
fit <- auto.arima(tsdata) # automatically find best arima parameters
pred <- as.data.frame(forecast(fit, h=npred))
out1 <- df[c('Date',var)][start:end,]
out1$type <- 'train'
out2 <- df[c('Date',var)][(end+1):(end+npred),]
out2$type <- 'holdout'
out <- rbind(out1, out2)
for (col in c('Point Forecast', 'Lo 95', 'Hi 95')) {
  out3 <- out2
  out3[,var] <- pred[col]
  out3$type <- col
  out <- rbind(out, out3)
}
ggplot(out, aes(Date, Var1, col=type)) + geom_line(lwd=2)