The first problem is: if there are many browser/version pairs (e.g., 30), then it's a very cluttered line graph. This implies that some filtering must happen (e.g., "Other"), e.g., graph only the top-5.

The next problem then is: what happens when a browser/version pair breaks into the top-5 (or conversely, falls out of the top-5)? Adding more lines to returns us to the first problem.

Well, I was thinking about more condensed browser/version pairs. Why on earth does Piwik differentiate Opera versions 10.62, 10.60, 10.61, 10.53, 10.50 in the first place? That's not really useful. Aren't they essentially the same, except for security patches? That clutters up my stats to a whopping 57 browsers (of which Opera as the browser with a 3% share takes up 15 versions).

But even if you count only major versions, you end up with an estimated 20 versions, that's right.

So the top-5 might be a good idea. As for your question regarding breaking into/falling out of that selected group: From a user standpoint, I'd say if a browser is within the top-5 more than half the time, show it. You could argument that cases like a 50% share for a day and a 1% share on all other days are not shown even though they ought to be. But that case is a bit far-fetched.

From a developer standpoint, I'd say forget that calculation -- it's too expensive. What we could do is look at the beginning, middle and end of the period. Or draw an arbitrary border: Every browser with a current (!) share of at least 5% is shown, and ditch the top-5.

I'm open for more suggestions.

What you are asking is I think two things:

• plot any line/data set over time Row Evolution: historical view of one or many specific rows in all Piwik reports! #534
• do not count Opera sub versions (is Opera the only one abusing user agent sub versioning ?)

I'll let you create the ticket for "do not count Opera sub versions" if this is relevant.