优化闪亮仪表板图重绘

Question

I have built a shiny dashboard which has a set of data loaded in as a data frame. It uses dplyr to then select columns, mutate new columns if needed, apply a set of filters and then plot using a variety of high-level ggplot2-based packages.

We try to do the data load, select and mutate just the once. The filtering is sat in a reactive variable, accessed by the plot, and is based upon different input values.

As far as I see this is a pretty standard and typical use case.

I'm wondering whether anyone could advise on workflow patterns to make the output more responsive.

There are two scenarios I encounter with this writing pattern which appear to cause significant user interface delay which I'd like to avoid:

• Firstly, when the dashboard first loads it tries to plot the charts using NULL data. I've found I can get around this by using if(is.null(my_data_frame)) and returning geom_blank() rather than our plot. Is there a neater / faster way to do this?

• Secondly (and more challengingly): to the right of my plot are a (potentially large) set of filter options to allow the user to analyse subsets. If the user clicks several of these options in rapid succession, Shiny will repeatedly recalculate our reactive() value and replot the chart for each click event: where the user actually just needed to set or clear 5 options. Is there a way to prevent this happening - so if the recalculation isn't complete we don't continue with the plot in progress as we've just made the data stale? Or do you think about grouping options with an update button to prevent this?

Answer 1

In response to your second bullet. Check out shiny's debounce/throttle capabilities. These should slow down the reactive response so your user can finish with the UI control before the chart or presentation element recalculates.

Answer 2

For your first question, try using the req function inside your reactive block. If your plot is waiting on the data frame to load, you can put the code for loading the data frame in a named reactive block (it seems like you may have already done this), then pass that to req . This will prevent the plot from attempting to render until it receives a valid value for whatever you passed to req .

For the second, I'd recommend wrapping your plot render in an observeEvent and having an 'Update' button, if you typically expect users to change multiple options between desired plot updates.

Finally, for added performance, I just saw today that Shiny v1.2.0 has either just been released or is about to be, the major feature of which is plot caching. You can find more details HERE .