I'm one of the "Apple Loyalists" seriously considering Linux. It's fairly easy to experiment with software (my main difficulty is Lightroom, currently looking at Corel AfterShot Pro). I've heard System76 isn't great hardware despite the convenience of shipping with Linux, but I'm not sure what the best options are.

Looking for a 15" laptop that can do RAW photo processing, I've come up with the Asus ZenBook Pro and the Dell XPS line. I've found articles on installing Ubuntu on both; they sound like they have some issues but it can be done.

Are there clear winners for Linux laptops outside of System76?

My strengths are definitely on the electronics / software side; I did fine in stats in high school, but haven't done anything significant since. Thanks for the notes on tests of randomness.

If you'd like to chip in, all the data is available on github. I closed my post with a call for statisticians to come along and do a better analysis; maybe you're it! (: Glad to help you extract the roll sequence numbers if using summarize.py to pull it out of each die's summary.json is not your thing.

And yeah, that first chart was frustrating. The easiest tool to hand was Google Spreadsheets, but the choices are side-by-side bar graph and line + bar graph. As you said, clealry the wrong tool for the job, but at least it gets the point across. But that too is in github along with its source CSV if you'd like to make a better one!

I actually get frustrated with clickbait titles and summaries, so I wanted to pack as much information into the summary as possible. My goal was to intrigue people by telling them what they could hear more about, rather than luring them by leaving out interesting detail.

So yes, I was trying to write the abstract of a paper, more or less. :)

I haven't tried a floating test; that'd be interesting to compare.

I did measure the diameters of a handful of the dice, though, and found some correlation with die behavior. markfickett.com/dice#geometry for more on that.

Do you think the dice float with one side up because of bubbles inside them, or something else? I wonder how a die that's egg-shaped or otherwise irregular in shape (but not mass distribution) would behave in a float test.

Woah, that's great!

It looks like it takes more infrastructure than my setup, though. :)

If you want to join the thread above about number of rolls, I'd be curious to hear more of your thoughts.

I think it depends on what you have on hand. I had a microcontroller and a servo motor, but not a woodshop or a reliable DC motor. I like the simplicity of mechanical systems though, avoiding computer control.

My guess is it would also be hard to make the motor and the timer coordinate precisely enough, over 3000 rolls. You could add some interconnect so the camera is tripped when the cam gets to the top, though.

One advantage of computer control though is in chaining the whole system together, roll + photograph + analyze. Timothy Weber actually has a video of something very similar to what I did, but with realtime analysis; unfortunately no detailed writeup yet. timothyweber.org/dieroller

Waiting for the dice to roll definitely was the longest step. But I also ran out of dice (even though now I have 5 seven-die sets, and three assortments of d20s!), and was making graphs or improving the software (or sleeping / going to work) while the dice ran. So even though it's the right target for optimizing the overall process (as the longest part), there wasn't a lot of pressure to optimize.

Also, I got away with a pretty simple heuristic for finding the die in the picture; just looking for an area that changed and using a simple flood-fill to identify the area. (That part was just using PIL, not any OpenCV fanciness. More at markfickett.com/dice/#cropping.) Extracting multiple separate dice would make that step much harder, especially since you'd need to not only extract them but identify which die is which.

What do you think of my comparison of different numbers of rolls out of the sequence? I actually started with 8k rolls. But comparing (computer-selected pseudo-)random subsamples of the roll sequence, using 2k, 3k, or 8k all looked about the same, whereas using 100 or 1000 rolls was quite different. markfickett.com/dice#numrolls

I actually did some investigation of p-values along the way. For example the Koplow dice I tested, which were slightly fairer than average among the 19 d20s I ran, all have p 0.001 for their 3001 roll datasets. But what I want to say isn't "with X% confidence, this die is not fair". I basically know that to start with. I want to know "how does this die behave?"

Me too! (: Unfortunatley neither my camera's mechanical shutter nor my budget for recreational dice purchases supports that whim. But I think the handful I tested still proved interesting.

Now, if someone wants to donate some larger bags of dice, I might try to port this to a Pi and its camera module (with an electronic shutter). And maybe add a mechanism for swapping out one die for the next. That's starting to sound like another serious project!

For better or worse, all the dice I tested seemed to have fairly symmetricl distributions. That is, if they favor 20, they favor 1 too. Even the very "unfair" dice of the lot tended just to favor or avoid some random middle numbers. So, no good tidbit for the summary on how to roll crits.

Same here. I scored it as 'something more complicated'.

Not necessarily. I hope a good spam filter takes the recipient into account as context. E-mail from Verizon to me? Spam. E-mail from Verizon to my co-worker with FiOS? Maybe not. E-mail addressing Mr. Smith delivered to neo@? More spam.