The Great Keyboard Layout Comparison: 2021 edition.

[iandoug]

Hi all

It's been a while since the the last time I did this exercise (2017) but I have done another comparison between the various keyboard layouts.

The results are up at https://www.keyboard-design.com/best-layouts.html

As part of the process, I also refreshed the Internet Letter Layout DB (ILLDB) at https://www.keyboard-design.com/internet-letter-layout-db.html

The layouts were evaluated for finger-based metrics using a fork of the Keyboard Layout Analyzer at https://klanext.keyboard-design.com , as well as some word and same-finger bigram tests via custom programs.

There were 355 layouts evaluated, in ANSI, Ergodox or Matrix form factors.

I included new layouts developed in recent times, as well as removing older development and minor variants, and those deemed "bad for hands"  (chiefly constant AltGr on ANSI) and "bad for heads", chiefly the Seelpy/Essie style layouts.

The analysis includes new features, while some used last time have been removed. The motivations are at https://www.keyboard-design.com/testing-2021.html

Click the links in the top 30 lists to see each layout's analysis, or via the ILLDB.

Feedback welcome.

Thanks, Ian

[Volny]

Seems very interesting. But coming at it from the perspective of someone with no prior knowledge of you or your website, or the keyboard analyzer tool, I felt there was some very basic info missing about the project's raison d'etre that would help me understand the purpose of it all.

I read (ok, skim read) a number of pages on the site and came across a lot of technical explanation and even the word acetylsalicylates. Yet I still couldn't tell what exactly you're assessing, and why: WPM, number of typos? Total distance travelled by fingers? Subjective measurement of RSI pain/fatigue? And who is the site for: people worried about RSI, coders who want to type as quickly as possible, interested consumers who want to get a sense of the market's breadth, retailers, engineers?

I'm sure I'd get answers to those questions if I drilled deeply enough into the content, but they should have been made clearly available to me earlier. I shouldn't have still been wondering those things after reading 3 or so pages. If I arrived from Google, I probably wouldn't have hung around long enough to get any actual value from the site.

I hope that feedback helped.

[iandoug]

Hi Volny

Thanks for the feedback.

It's aimed at people looking for something better than QWERTY.

If you're not in that group, then it will be irrelevant to you.

I will write an intro. :-)

Thanks, Ian

[Sylvester_Stalin]

Hello Ian, i am trully happy that you resurected this analyzer.

i have a short feedback to give you just after few minutes of usage:

In results section, 3 digits decimal is way too much, it is really hard to read them properly without getting an insane headache.
1 digit is far enough imo.
I don't get how your shakespeare corpus work. Could you give us a hint please.

The second point i really do like on others analyzers is on the heatmap. We have the possibility to quickly check the number of keystrokes per key when pointing them with mouse cursor.

Beside, i took a quick look on what you're planning to do (possibility of customization of fingers penalties), it's something that was in my mind when i saw all different forks from Patrick's original analyzer with different scoring system (and results) given for the same corpus and layout.

Keep up

Edit: typos

[iandoug]

Hi Sylvester

You are probably right, I put in three when I was trying to debug the "distance measuring" problem.

There's a write-up here (scroll down to 4 May 2021)
https://www.keyboard-design.com/index.html

Regarding the heatmap, I inherited the fork from Den / Shenafu, he made those changes. I will take a look after I find and fix the "reload on first edit" bug ... spent most of the day going around in circles trying to find why it happens...

For the Shakespeare corpus, see my breathless prose in
Keyboard-Layout-Analysis-Createng-the-Corpus-Bigram-Chains-and-Shakespeares-Monkeys-1.0.1.pdf

(embarrassing typo included at no extra cost) at
https://zenodo.org/record/4644104

The zip file is also useful). Get the 101 versions.

Cheers, Ian

[Sylvester_Stalin]

hello Doug

i have an idea to submit
We all use different languages aside of english, and some of us are trying to find the layout that fits the best to our use of english and native.
If i remember well, there were non english test from the wiki of Joan of Arc, on the Den's KLA, in yours there is only latin (which isn't used anymore).

i just read your .pdf about how you create design the differents clever monkey english bigram corpus.
I would find interesting to have the ability to analyze our layouts with texts similar to Shakespeare clever monkey, but with a mix of like 40% english /40% french or spanish  or any language /20% program to figure out how good/bad/performant they are.

i really don't know how long it would take to create such corpus for the differents germanic and latin based languages.

[iandoug]

Hi

I still have a mirror of KLAtest up at https://klatest.keyboard-design.com/ but note that apart from the "reload" bug ( which I have not tracked down yet) there is also a problem with the distance measurement.
See item dated 4 May 2021 on https://www.keyboard-design.com/index.html . I fixed that on KLANext (still has "reload" bug).

I have been thinking about other languages and actually got corpus files for a variety of (western) European languages. So it's on the ToDo list but don't know when.

At present trying to find the best layout programmatically...

Curiously, layouts that do well at English tend to do well ta Latin. It's just there for a bit of entertainment.

I need a better Dropdown system, they are getting too long.

Cheers, Ian

[Sylvester_Stalin]

old message deleted: misread

programatically best layout, is the symbol layout?

[iandoug]

No, using computer program to find best layout.

Now I have millions of layouts, need a way to evaluate them that is faster than KLE and that will give same results...

All attempts so far are failing... 

[metalkeyboards]

madness!

[iandoug]

madness!

Maybe. It will take about a month just to generate the layouts, so I did a more limited ZXCV version,,, you are welcome to test it on Patrick's KLA or on https://klanext.keyboard-design.com/#/main

Cheers, Ian

[Sylvester_Stalin]

I have some questions about your analyzer.

What i discovered is for ergo keyboards only (not tested with matrix boards).  If you take a layout and make a mirrored version of that layout, and then compare them to each other. results given favors a layout more. But how is that possible for 2 exact layouts to have 2 different results?
Is there any penalty applied for a hand more than the other?



I figured out that the analyzer had some preferences if you set mods on both thumbs. it will always pick up the left hand one.

There is also something i would like you explain to me, i did create a variation of CHIA and copare it to the original, the scores given to me, give CHIA winners against my CHIA 2, i says distance travel is better on CHIA, but when a took a look at distances, my version had a better distance scoring. is there a bigger penalty for pinkies?



   

[iandoug]

Hi

Yes, left hand has penalty of 1.1 times right hand.

We can debate if this is fair, it assumes a right-handed typist, and the 10% is a general scientific consensus for the difference between dominant and other hand.

The fingers have different penalties (SteveP's version has different values, as do Den's versions. Mine are taken from a published paper.)

KB.dScoring[KB.finger.RIGHT_THUMB] = 1;
KB.dScoring[KB.finger.RIGHT_INDEX] = 1.85;
KB.dScoring[KB.finger.RIGHT_MIDDLE] = 2.0;
KB.dScoring[KB.finger.RIGHT_RING] = 2.9;
KB.dScoring[KB.finger.RIGHT_PINKY] = 3.67;

and same for left hand, but * 1.1. So mirroring a layout will give different score.

The figures in the results tables include those weightings ... you can find the unweighted values in the other result tabs.

Not sure what mods you are referring to, but KLA scans the keyboard and uses the first one it finds. So you should only have one of each... same with characters.

Can you post json?

Hope that helps :-)

Cheers, Ian

[Sylvester_Stalin]

It's not my job to tell what's fair and what is not, you are the judge, you code this analyzer, but who knows what kind of typist is using your analyzer?
What about ambidextreous people? I am a right handed main but there are things i can do well with my left hand.
It biaised results, neutrality has to be a priority, as long as users can't customize penalties per hands and per fingers, but this is my opinion.
 CHIA.json (15.35 kB - downloaded 88 times.)
 CHIA mirror.json (15.35 kB - downloaded 88 times.)

Edit: typo

[iandoug]

I didn't code it, it's a fork of a fork.

I did make some changes to make it "fairer" as best possible.

We have discussed letting users set parameters, but then people will adjust them so that their own layouts score well and comparisons become meaningless.

KLA makes various assumptions. Apart from assuming you are right-handed, it also assumes  that you are a perfect "touch typist" who always uses the correct finger for each key, returns each finger to home position when not used, and never makes mistakes.

And also that ====== is typed with 6 key presses and not auto-repeated.

These assumptions are not true for most people, so the layout evaluation is a 'best guess' in some ways.

Cheers, Ian

[Sylvester_Stalin]

I know that you are not the original designer of it and that it just acts like in the perfect world.
I just make suggestions. If i had some knowledge about JS i would be glad to help you if you wanted to. but it is not the case  (JS is on my list, but not right now)
You are right about the customization of parameters, it would be amazing but the original idea wouldn't last long, as you say people would make perfect layouts based on their preferencies.
I still think that the "hand neutrality" should be.

Anyway, thank you for lighting my candle about those hidden parameters that are probably ignored by many of us.