A way to sanitize spiked/flooded download data

[samal-rasmussen]

The Svelte package recently had their download numbers flooded with a huge amount of downloads for a short amount of time. This makes the graph unreadable as the downloads spike compresses the graph all the way down.
https://npmtrends.com/svelte

I am searching for a way to clean up this spike, so I can actually view the unsquashed graph.

I did make this super hacky workaround to flatten the date, if you're really desperate to see the realistic graph:

In the .js bundle on the npmtrends site I found the success handler for fetch requests. It looks like this:

  e.fetch = function(t, e) {
      ...
      return this.retryer = new a.m4({
          fn: y.fetchFn,
          abort: null == h || null == (o = h.abort) ? void 0 : o.bind(h),
          onSuccess: function(t) {
              s.setData(t),

I put a breakpoint of the setData line and executed this in the terminal when the t value is the graph data:

// t is the fetched data. We update it here.
t = t.map((tt) => {
    tt.downloads = tt.downloads.map((d) => {
        // These values are hand picked from reading the svelte download data.
        // You'll have to hand pick some different values for a different package.
        d.downloads = d.downloads > 155911 ? 70000 : d.downloads;
        return d;
    })
    return tt;
})

You can trigger the downloads data fetch by selecting a new time interval in the dropdown on the page.

[dominikg]

note that a similar spike happened to vue, so you're either looking at manually patching historical data on a package name + timeframe basis which is an ongoing effort, or you'd need to apply some kind of heuristic to find unrealistic spikes and flatten them.