In JMP 10 for Windows, the Line in Graph Builder would fit through every point (every row). I'm testing JMP 11 on Mac, and I get this result:
Is this a Mac thing, a JMP 11 thing, or am I missing a setting somewhere?
If you sort your X variable column and then replot you'll get a line through every point. If there are multiple Ys for a particular X the line will go through the mean value. You can also click the Row order checkbox for the line.
The Sort/Row Order fix works for the first part of the data, but not all of it. With the data sorted by occurred_at (the x variable), I get this when Row Order is unchecked:
The line does not fit through each point, instead smoothing them as I showed previously.
If I select "Row Order" on the same sorted data, I get this:
The line fits each point for part of the data on the left (the desired effect) , but then it just stops! Very mysterious indeed. I'll reproduce the graph in JMP 10 on Windows to show the difference if that helps.
I've noticed that with smaller datasets, the graphs perform the way I'd expect (i.e. more like the JMP10 on Windows). Is it possible that the line averaging is part of a data compression process used to display large data sets?
I typically graph between 100,000 and 1,000,000 rows. 3 days of 1 second resolution data for 3 objects is a common data set, for example.
I've been running JMP10 for the last few months due to the issue above, but I just installed JMP 11 to see if the update had fixed the issue. I have both 10 and 11 installed side by side, and you can clearly see the differences. This is the same file with the same scripts run in 10 (left) and 11 (right) with the Graph Builder (top) and Overlay Plot (bottom) compared. Only JMP11 Graph Builder shows this odd averaging behavior, and it appears to be a regression from JMP10's behavior.
Has anyone found a setting to fix this? I'd love to stick with JMP11 if I can find a solution.
Sorry about that, afterword. We (JMP development) found that Windows would hang trying to draw a line with tens of thousands of line segments, and figured you can't see that many unique X values anyway and down-sampling is better than hanging. Looks like we didn't compare Mac OS drawing speed and used the same limit for both (10,000 X values). And given that your Overlay Plot came out OK, maybe the limit should be higher for WinOS, too. Will investigate.
I didn't make it very clear in my last post, but that four image comparison is of Windows JMP 10 and 11 running on a Windows 7 VM inside Mac OS (Mavericks) with Parallels. How's that for a test environment?!
The original image was Mac JMP11 on native MacOS.
Thanks for the explanation, though. I was perplexed.
That probably counts as WinOS as far as the graphics libraries go. After a little exploration, we're still seeing sluggishness above 10K on both Mac and Windows (one second and up pause). How is the speed for you? I wanted to keep it fast because I'm still operating on the assumption that no sane person would draw lines with 10K unique X values , and it's more likely a mistake. Maybe we can find a smarter way to down-sample or provide an override.
The bulk of the time on my canned reports comes from pulling the data via our API, so I probably don't notice the rendering time as much. Is there a sensible way for me to actually measure that time?
I like the idea of an option like the one that exists for the "Fast Marker Threshold." The first thing I do when I install JMP is up that number to 1e6 or so. It's possible that I don't notice the "sluggishness" because I'm not building these by drag and drop each time. If you were doing a new analysis, then the speed matters more.