Author Topic: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own  (Read 19183 times)

0 Members and 1 Guest are viewing this topic.

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« on: Sun, 11 January 2015, 22:10:05 »
Some interesting statistics based on the layout analyzer I mentioned in this other thread: https://geekhack.org/index.php?topic=67430

The text used for this particular example was a book called "The Wasted Generation" which you can find at the Gutenberg project. I tested these layouts with a few other texts as well and the results are generally the same.

A few comments on how I value some statistics:
1. Same finger is the most critical statistic. Having to type two letters in a row using the same finger (double letters don't count, such as LL in heLLo) is the slowest thing you can do.
2. Alternation is the next most important statistic. It allows your other hand to get ready to type the next key while the first hand is pressing the first key.
3. Inward rolls are very comfortable, something I learned from using Colemak.
4. Outward rolls are generally bad. If possible try to convert them to alternation or inward rolls.
5. I think that the G and H keys (positions given relative to QWERTY) are not "free" homerow positions. In the Workman layout, he talks about the lateral movement from F to G and from J to H being worse than the vertical movement in reaching from S to W or D to E for example. That being said, those G and H positions are still pretty good, just not the best. I prefer WE on the left hand and IO on the right hand. So because of this, I don't care as much about the "middle row usage" as others might. To me the homerow would be ASDFWE on the left and JKL;IO on the right, so I put the most frequent letters on those positions.
6. Your punctuation keys should be on the hand opposite the thumb you use to press spacebar. That's because punctuation will always be followed by a space. If you were in the business of making your own layout or learning a new one I would suggest flipping the layout if the punctuation is on the same side as your spacebar thumb. I use my left thumb for spacebar.

I'll let the stats speak for themselves for the most part and just provide some general thoughts.

Dvorak
A very good layout overall! Same finger usage is a bit high and inward rolls could be a bit higher. Right pinky perhaps a bit too used and right hand is used notably more than left. One thing to notice here is that the percent usage of each finger in Dvorak is much closer than in any other layout. This could be considered good or bad. I personally tend to think that you should emphasize the strong fingers more but arguments could be made for a balance.


Colemak
From http://colemak.com/
This is currently what I type in. But looking at these stats I'm not too impressed! Terrible alternation. Too many outward rolls. Right hand used notably more than left.


Workman
From http://www.workmanlayout.com/blog/
Statistically it is very similar to Colemak. The big different is that Workman tries to emphasize human comfort by removing the lateral movements of the index fingers and instead using vertical movements. The problem with this is you get higher same finger frequency. And while I agree with reducing lateral movement, I think that some letters could be better positioned to increase alternation.


Carpalx
From http://mkweb.bcgsc.ca/carpalx/
A layout algorithmically generated based on heuristics. I'm not a fan of this layout at all. Here the focus was too much on "distance" which is how far your fingers have to move. It doesn't pay enough attention to other statistics which I find are more important (same finger and outward rolls tend to slow down movement) and doesn't pay attention to human comfort (what good is travelling a short distance if every movement is uncomfortable?).


MTGAP
From https://mathematicalmulticore.wordpress.com/category/keyboards/
A layout algorithmically generated based on heuristics. I like this layout a lot and as you'll see in a bit it's very similar to my own (but flipped). It has pretty stats all over but alternation could be higher and outward rolls could be lower.


OneProduct (my own. Need a name for it...)
A layout algorithmically generated based on heuristics. I used the same "simulated annealing" approach that Carpalx used, but with different heuristics. I then hand tweaked the results a bit so that it was a bit more "comfortable" as the heuristics don't account for human comfort as much as they could. I also made a few changes where the layout became more similar to some other layouts (for familiarity) where there was a very minimal decrease in heuristic viability. A few things in particular/of note:
1. Compared to Colemak, my current layout, the following are in the same places: GJEIRSTZVB. That's 10 out of 30.
2. ZCV remain on the bottom row near the left but because it is statistically good! I didn't have to force them to be there.
3. Compared to MTGAP, our algorithms came to the same conclusions a lot. In particular:
3.1 EO? in a column
3.2 A and U on the same finger
3.3 Punctuation on the same finger as A
3.4 A as the index finger key (this is important because the index fingers control 2 columns, and as such cause a lot of same finger clashes if not planned well)
3.5 RST arranged as an inward roll (colemak is the same)
3.6 IEA arranged as an inward roll
3.8 PNZ in a column


Same text analyzed by http://patorjk.com/keyboard-layout-analyzer to at least show that my statistics aren't completely wacky. I lost by a bit based on their scoring system but it's so close (0.07) that is basically a tie. ;D However I think that my layout's stats (notably higher alternation and "Workman" style top row usage) provide a bigger bonus that's not spoken for as by this scoring system as much as it could be.
« Last Edit: Mon, 12 January 2015, 17:22:32 by oneproduct »
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline mstechfreak

  • Posts: 9
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #1 on: Mon, 12 January 2015, 02:44:48 »
What about typing speed here? How much can you get?  - http://www.typingstudy.com/en-us_dvorak-3/speedtest

Offline highend

  • Posts: 13
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #2 on: Mon, 12 January 2015, 04:48:24 »
Very interesting, thanks for analyzing those layouts! Maybe you could do it for the Workman layout, too?

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #3 on: Mon, 12 January 2015, 08:55:29 »
What about typing speed here? How much can you get?  - http://www.typingstudy.com/en-us_dvorak-3/speedtest

I haven't tried that particular typing speed test. On typeracer.com my average is 100 WPM, max 123 http://data.typeracer.com/pit/profile?user=stephenpoconnor. On 10fastfingers.com my average is 113, max 132 http://10fastfingers.com/user/107456/. That's with Colemak, which I learned about 3 years ago.

Very interesting, thanks for analyzing those layouts! Maybe you could do it for the Workman layout, too?

Sure, I'll do that tonight when I get home.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline naz

  • Posts: 53
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #4 on: Mon, 12 January 2015, 12:43:39 »
hi! is there any chance you could do a run with other lenguages??? most layouts are made thinking in english, so it would be interesting to see how well they do in spanish, french, german, etc.

Regards!

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #5 on: Mon, 12 January 2015, 13:11:07 »
I'd most likely have to add support for foreign characters depending on how they are input. From what I've seen of European keyboards, they either have more keys for special characters, have them on other layers or press combinations of keys to get them, such as /e making é or something like that.

It's certainly something I'd like to support with this layout analyzing tool though. I'll see what I can do.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline rebus

  • Posts: 14
  • Location: Italy
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #6 on: Mon, 12 January 2015, 14:54:56 »
Most analysis claim Colemak to be a better layout over Dvorak but, after trying both, I feel much more better on Dvorak. I think comfort is very difficult to predict on the base of just numbers, and that personal experience may show different feelings among different people. At the end I am beginning to think that the best-for-everybody doesn't exists

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #7 on: Mon, 12 January 2015, 16:32:07 »
I had the opposite experience somewhat. I started on Dvorak and ended up with Colemak. However, I didn't give Dvorak enough time to become fully proficient with it.

Comfort is certainly hard to predict, but there are a few things which most people will generally find comfortable based on hand physiology. There certainly won't be one layout that's best for everyone and I would guess that finger length would be the largest contributing factor to affecting what is comfortable.

The same kind of thing affects people's preferences in switches. A lot of people like heavy switches here, but I'd say that's because many of them have rather large/heavy hands. I have very small, light hands so I prefer light switches.

And as you said, numbers don't tell the story of comfort well. While trying my layout now I just realized that it's fairly uncomfortable to do 2-grams with H(vowel) which are very common. That's likely why MTGAP puts N with the vowels instead of H. I might try to fiddle around with something similar.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #8 on: Mon, 12 January 2015, 17:22:21 »
Added the Workman layout to the analysis.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline batfink

  • Posts: 57
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/my own
« Reply #9 on: Tue, 13 January 2015, 03:40:50 »
I started on Dvorak and ended up with Colemak .....
While trying my layout now I just realized that it's fairly uncomfortable to do 2-grams with H(vowel) which are very common.

There is a simple adjustment to Colemak that can be done which fixes exactly that issue (switch H with M).

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #10 on: Tue, 13 January 2015, 09:43:40 »
That's very interesting batfink. I'll have to take a look at that when I get home tonight.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #11 on: Wed, 14 January 2015, 07:02:22 »
Interesting. Maybe one extra source of inspiration: ADNW.de and the ADNW Google Group.

My top 6 interesting websites with regards to layouts  is, in no particular order:
- MTGAP - information, discussion, algorithm available (C source code is on GitHub)
- carpalx - information, algorithm available (Perl code)
- colemak - information, discussion. No algorithm - colemak is sort of 'god given'  ;). But forum members do post nice an usable tools, such as autohotkey scripts, xkb maps etc. that are also usable for other layouts.
- ADNW plus the Google ADNW Group - information, discussion (on Google Group), algorithm - (C++ code)
- Deskthority Layouts Wiki - information
- patorjk - information, visualisation of layouts, some testing

I type mainly in Dutch, plus some English and occasionally German. I have tried Colemak (not good for Dutch and German). I calculated my own layouts, using the MTGAP algorithm, and based on a Dutch/English corpus. Much better, but not enough alternation. I didn't like the Carpalx layouts, not even the one that was made for Dutch especially.

For me, ADNW is best. However, both the website and the google group are in German, which really is shame. The algorithm etc. are universal of course.
I agree with the assumptions, which as Dvorak-like:
- high alternation
- minimal adjacent fingers, esp with pinky & ring. Qwerty as or l; are bad
+ the usual goals (minimize distance, same finger, row jumps; equal hand balance; optimal finger balance).
Also the algorithm is really tweakable and edven has a nice visual presentation of the results.
 
Let's compare ADNW versions to your layout. The adnw versions were calculated by me and a fellow adnw group member. that I and an google group member calculated for use with English text.  The locations marked * are free, put characters like / ? # " on those places. We did not optimize these, although you could do so. I prefer to do that by hand.

ADNW English, version A (trigram optimisation)
kyu.* zlmdpv
rieao hnstcw
x**,* jqfgb


ADNW English, version B (bigram optimisation)
qyu.* zmldbp
sieao hnrtcg
j**,* fxwkv


Standard ADNW (made for English + German):
ku*.* vgcljz
hieao dtrnsf
*y*,q bpwmx
[/size][/font]

Your version (mirrored, and special characters replaced by *, for easy comparison ):
qyoxj gwdlp
hieau mtsrn
k**., bvfcz


I see similarities! To me, that means you are on the right track, yours looks like a nice layout. I will see if I can do some analysis on these layouts and compared the numbers.
« Last Edit: Wed, 14 January 2015, 07:08:19 by PieterGen »

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #12 on: Wed, 14 January 2015, 10:37:42 »
Hi PieterGen. I'll take a look at those layouts and see what my analyzer says about them, though it should be similar to what other analyzers say for the most part. I'm still working on mine and adding a few different measurements as I go along. Yesterday I added "average penalty" which is where you assign a weight to each key. I have some catching up to do compared to some of the other layout analyzing tools in some places. My tool was mostly built to be convenient for hand tweaking, as you can move keys and see updated stats in real time.

Right away what strikes me with the ADNW layouts is having EU on the same finger and AO on the same finger as opposed to EO and AU as I do which cause fewer same finger bigrams (for english in any case). Also, one thing that I constantly find myself trying to optimize is the word "you." Although using a large corpus tends to indicate that that word doesn't deserve special treatment, in reality when I send emails or instant messages things like you/your/you're are very frequent so I try to make it comfortable. In my layout I like that (in addition to reducing same finger bigrams by going with EO and AU) YOU is an inwards roll.

The only thing that I really don't like about my layout is that it is not comfortable to type the letter H followed by a vowel as inward rolls that start with the pinky aren't really that comfortable, as you stated with your "minimal adjacent fingers, esp with pinky & ring. Qwerty as or l; are bad"

The H key is really troubling for me because if you put it with the vowels, but not on the pinky, then you create a lot of outward rolls. The alternative is to put it on the "inside" of T since TH is the most common bigram, but any consonant that you put with the vowels other than H results in around 5% less alternation.

I'll do a bit more looking into the ADNW layouts you posted when I get home tonight, but yes, the similarities are certainly encouraging! Thanks for the feedback!

Edit: Oh, just realized that the ADNW layouts use two pinky columns on the right hand, so my layout analyzer can't actually look at them at the moment! Guess I have more expanding to do.
« Last Edit: Wed, 14 January 2015, 12:16:24 by oneproduct »
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #13 on: Wed, 14 January 2015, 15:58:04 »
Here are rankings. Explanation and details tomorrow. OneProduct does very well  :thumb:

HOW WELL CAN YOU TYPE ENGLISH TEXTS ON THESE LAYOUTS? SCORING BASED ON BIGRAMS.
Lower is better

521.6232 qwertyuiop*asdfghjkl**zxcvbnm,.*   QWERTY
290.8380 *,.pyfgctz*aoeiuhdrnsl**jkxbmwvq   Dvorak
261.9580 pldwgjxoyq*nrstmuaeih*zcfvb,.**k   OneProduct
259.8202 qwfpgjluy**arstdhneio*zxcvbkm,.*   Colemak
255.8730 qyoxjgwdlp*hieaumtsrn*k**.,bvfcz   OneProduct,Mirrored
232.8857 ku*.*vgcljfhieaodtrns*xy*,qbpwmz   ADNW(German/English)(bigrams)
232.3586 ku*.*vgcljzhieaodtrnsf*y*,qbpwmx   ADNW(German/English)(trigrams)
225.7789 kyu.*zlmdpvrieaohnstcwx**,*jqfgb   ADNW(English)(trigrams)
220.8365 qyu.*zmldbpsieaohnrtcgj**,*fxwkv   ADNW(English)(bigrams)
er

HOW WELL CAN YOU TYPE ENGLISH TEXTS ON THESE LAYOUTS? SCORING BASED ON TRIGRAMS.
Lower is better

576.3826 qwertyuiop*asdfghjkl**zxcvbnm,.*   QWERTY
334.4042 *,.pyfgctz*aoeiuhdrnsl**jkxbmwvq   Dvorak
308.8693 pldwgjxoyq*nrstmuaeih*zcfvb,.**k   OneProduct
307.7219 qwfpgjluy**arstdhneio*zxcvbkm,.*   Colemak
302.7142 qyoxjgwdlp*hieaumtsrn*k**.,bvfcz   OneProduct,Mirrored
281.0470 ku*.*vgcljfhieaodtrns*xy*,qbpwmz   ADNW(German/English)(bigrams)
280.8407 ku*.*vgcljzhieaodtrnsf*y*,qbpwmx   ADNW(German/English)(trigrams)
262.6170 qyu.*zmldbpsieaohnrtcgj**,*fxwkv   ADNW(English)(bigrams)
261.1716 kyu.*zlmdpvrieaohnstcwx**,*jqfgb   ADNW(English)(trigrams)


What does this mean?
1. Qwerty scores lousy, as expected
2. OneProduct beats Dvorak and Colemak ! The mirrored version does best. OneProduct, this analysis says you might try using the mirrored version - of course this is based on subjective parameters and a standard staggered keyboard.... In the end, these are just numbers, nothing beats "trying out in real life"
3. ANDW is a ' Dvorak like layout/algorithm'. And yet, remarkably, according to the ADNW "judge", Colemak beats the old Dvorak. That is, for English text - see later !
4. The ADNW English versions are better for typing English than the versions that were made for mixed English and German. Duh! This is like saying: playing tennis is easier with a tennis racket than with a baseball bat. But still.... the differences are small! The mixed (German/English) ADNW is better for typing English than "pure English" Colemak...




FOR COMPARISON: HOW WELL CAN YOU TYPE GERMAN TEXTS ON THESE LAYOUTS? SCORING BASED ON TRIGRAMS.
Lower is better

676.8695 qwertyuiop*asdfghjkl**zxcvbnm,.*   QWERTY
396.2951 qwfpgjluy**arstdhneio*zxcvbkm,.*   Colemak
360.9954 *,.pyfgctz*aoeiuhdrnsl**jkxbmwvq   Dvorak
359.4779 pldwgjxoyq*nrstmuaeih*zcfvb,.**k   OneProduct
358.7381 qyoxjgwdlp*hieaumtsrn*k**.,bvfcz   OneProduct,Mirrored
325.9837 qyu.*zmldbpsieaohnrtcgj**,*fxwkv   ADNW(English)(bigrams)
324.2879 kyu.*zlmdpvrieaohnstcwx**,*jqfgb   ADNW(English)(trigrams)
297.9264 ku*.*vgcljzhieaodtrnsf*y*,qbpwmx   ADNW(German/English)(trigrams)
292.9240 ku*.*vgcljfhieaodtrns*xy*,qbpwmz   ADNW(German/English)(bigrams)

How does this work for German? (PS, corpus lenght may differ to English corpus, so don't compare numbers please).
1. For German, Dvorak is better than Colemak. Colemak is not very good for German.
2. It's very nice that OneProduct does better than Dvorak, even for typing German texts!
3. The ADNW versions, especially the versions that were partly made for German, do best.

I'll post the detail figures later.

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #14 on: Wed, 14 January 2015, 16:26:54 »
I have finished it now. But no time for better text layout, I'm sorry.

Here are the details.

Explanation:
The total score (lower is better) consists of base (how well are the keys layed out) plus other factors:
- SameFinger (% of bigrams typed with 1 finger). Should be low
- Alternation. Should be high
- In/Outward movement ratio. Should be high (1.00 means the same amount of inward rolls as outward rolls)
- adjacent: bigrams typed with pink/ring (worst), ring/middle (bad) middle/index (somewhat bad). ADNW prefers for instance ring/index or mid/pinky. This in constast to Colemak
- left/right - hands, %
- upper, home, lower - the rows
- the distribution of weight over the fingers. Optimal (says ADNW) is something like 8/10/18/14

For brevity, only trigram analises. Sorry for the bad text layout, no more time....


QWERTY           576.383 Total score  338.719 base        left right
                   6.804 SameFinger     6.299 Shift-SameFinger  upper 28.0 20.2
  qwert yuiop*    52.755 Alternation   41.474 Shift-Alternation  home 22.1  9.5
  asdfg hjkl**     1.080 In/Out Ratio  11.226 IndirSameFing     lower  6.8 13.3
  zxcvb nm,.*     21.628 adjacent      12.586 Shift-adjacent  sum 56.9 43.1
                  9.1  8.4 18.5 20.9 --.- --.- 18.4  8.9 12.1  3.6 Sh  1.1  1.7

QWERTY: what can I say? It's worse in ALL aspects.

Dvorak           334.404 Total score  200.043 base        left right
                   3.235 SameFinger    13.145 Shift-SameFinger  upper  6.0 15.7
  *,.py fgctz*    70.679 Alternation   33.265 Shift-Alternation  home 36.1 31.7
  aoeiu hdrnsl     1.445 In/Out Ratio   6.896 IndirSameFing     lower  2.9  7.6
  **jkx bmwvq      9.404 adjacent      21.079 Shift-adjacent  sum 45.0 55.0
                  9.7  8.2 13.0 14.1 --.- --.- 16.5 10.7 16.3 11.5 Sh  1.8  0.9

DVORAK HAS A GREAT alternation, nice in/out balance, bad left/right balance (45/55), it overloads the right pinky a bit(11,5%), and a too high SameFinger (3,2% is too much, even though it is only half of Qwerty's lousy 6,8%)

Colemak          307.722 Total score  186.227 base        left right
                   1.348 SameFinger    14.374 Shift-SameFinger  upper  7.8  8.2
  qwfpg jluy**    58.143 Alternation   39.607 Shift-Alternation  home 32.7 37.3
  arstd hneio*     1.041 In/Out Ratio   8.772 IndirSameFing     lower  6.8  7.3
  zxcvb km,.*     17.492 adjacent       9.009 Shift-adjacent  sum 47.2 52.8
                  9.1  7.8 11.6 18.7 --.- --.- 18.8 15.4  9.8  8.7 Sh  1.1  1.7

COLEMAK: good same finger, does worse on in/out than Dvorak (!) and scores high on adjacent fingers. Which is something that Colemak sees as GOOD.

OneProduct       308.869 Total score  198.465 base        left right
                   1.598 SameFinger    20.786 Shift-SameFinger  upper 13.3  9.1
  pldwg jxoyq*    71.206 Alternation   23.896 Shift-Alternation  home 30.3 33.9
  nrstm uaeih*     3.474 In/Out Ratio   7.473 IndirSameFing     lower  8.5  4.9
  zcfvb ,.**k     10.651 adjacent       9.409 Shift-adjacent  sum 52.1 47.9
                  9.8 12.9 12.4 17.1 --.- --.- 13.1 18.7  8.7  7.4 Sh  1.0  1.8

OneProduct,mirror302.714 Total score  192.927 base        left right
                   1.598 SameFinger    20.786 Shift-SameFinger  upper  9.1 13.3
  qyoxj gwdlp*    71.206 Alternation   23.896 Shift-Alternation  home 33.9 30.3
  hieau mtsrn*     3.474 In/Out Rati    7.473 IndirSameFing     lower  4.9  8.5
  k**., bvfcz     10.651 adjacent       9.409 Shift-adjacent      sum 47.9 52.1
                   7.4  8.7 18.7 13.1 --.- --.- 17.1 12.4 12.9  9.8 Sh  1.8  1.0

ONEPRODUCT(normal and mirrored). Good alternation, very good in/out ration, nice left/right balance, base is somewhat worse, meaning that the letters are not optimal under the fingers. But keyboards are always compromises. All in all very good. says ADNW ;-) 


ADNW(Eng, bigr)  262.617 Total score  182.722 base        left right
                   0.737 SameFinger     2.154 Shift-SameFinger  upper  5.5 13.6
  qyu.* zmldbp    68.832 Alternation   34.746 Shift-Alternation  home 40.0 31.1
  sieao hnrtcg     1.134 In/Out Ratio   6.990 IndirSameFing     lower  3.1  6.7
  j**,* fxwkv      9.123 adjacent      10.742 Shift-adjacent  sum 48.7 51.3
                  8.5  8.7 14.2 17.3 --.- --.- 16.4 11.5 13.1 10.3 Sh  1.8  1.0

ADNW(Eng, tri    261.172 Total score  183.941 base        left right
                   0.882 SameFinger      7.423 Shift-SameFinger  upper  6.1 13.1
  kyu.* zlmdpv    65.718 Alternation     28.070 Shift-Alternation  home 39.5 31.4
  rieao hnstcw     1.016 In/Out Ratio    6.177 IndirSameFing     lower  3.3  6.5
  x**,* jqfgb      8.665 adjacent       9.226 Shift-adjacent  sum 49.0 51.0
                  8.8  8.7 14.2 17.3 --.- --.- 15.8 11.0 14.3  9.9 Sh  2.0  0.8

ADNW(Ger/Eng, bi 281.047 Total score  184.270 base        left right
                   1.020 SameFinger     17.224 Shift-SameFinger  upper  4.5 12.0
  ku*.* vgcljf    71.389 Alternation     21.757 Shift-Alternation  home 38.3 31.6
  hieao dtrns*     2.136 In/Out Ratio    7.795 IndirSameFing     lower  5.0  8.6
  xy*,q bpwmz      8.122 adjacent      20.009 Shift-adjacent  sum 47.8 52.2
                  7.6 11.2 11.7 17.4 --.- --.- 18.7 10.7 13.1  9.7 Sh  1.9  0.9

ADNW(Ger/En,tri )280.841 Total score  184.072 base        left right
                   1.019 SameFinger     17.543 Shift-SameFinger  upper  4.5 10.0
  ku*.* vgcljz    71.575 Alternation     21.418 Shift-Alternation  home 38.3 33.7
  hieao dtrnsf     2.208 In/Out Ratio    7.831 IndirSameFing     lower  4.8  8.7
  *y*,q bpwmx      8.077 adjacent      20.027 Shift-adjacent  sum 47.6 52.4
                  7.4 11.2 11.7 17.4 --.- --.- 18.7 10.7 13.1  9.9 Sh  1.9  0.9

ADNW (all versions). The pure English versions have a very good, very low same finger %. The mixed language versions have more alternation, and more inward movements. Both come mainly from the placement of the H. The very common digram TH will now give alternation. The balance (L/R) of rieao is better than the hieao versions though. Compromises.... ! What I like less is that the index does more than the middle finger, I prefer that the other way around.

« Last Edit: Wed, 14 January 2015, 16:34:10 by PieterGen »

Offline jacobolus

  • Posts: 3634
  • Location: San Francisco, CA
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #15 on: Wed, 14 January 2015, 16:55:35 »
- adjacent: bigrams typed with pink/ring (worst), ring/middle (bad) middle/index (somewhat bad). ADNW prefers for instance ring/index or mid/pinky. This in constast to Colemak
middle/pinky (3–5) is just about as bad as ring/pinky (4–5) IMO, with middle/ring (3–4) being worse than either one, and middle/index (2–3) being not very bad at all. Three-grams using the last three fingers are especially bad, unless it’s a straight roll (especially an inward roll is alright).

Just measuring bigrams with these fingers is not ideal, because if it’s an isolated bigram followed by forefinger (2), thumb (1), or a finger on the other hand, then the two keys in a row can usually be struck in a single motion. But sequences like 5–3–4 or 3–4–3 basically need to be executed in two discrete chunks, because they require flexing and then relaxing the relevant muscles before flexing them again.

Quote
- the distribution of weight over the fingers. Optimal (says ADNW) is something like 8/10/18/14
Can you explain what these numbers mean? I’m interpreting it as pinky/ring/middle/index finger ratios. As in, among non-space characters, each pinky should get 8% of keypresses, ring finger should get 10%, etc.

If that’s the case, I think that breakdown seems nutty. The forefinger is much stronger and more agile than any of the other fingers, and controls more keys. Why in the world should it have a lower percentage of keypresses than the middle finger?

In any event, I don’t think this weight distribution number is relevant per se. Any useful information it adds to a prediction of layout quality is already present in other better measurements, and most trade-offs that have to be made with other measures to improve this 'distribution' score are IMO likely to be net-negative.

* * *

You should explain what weighting you apply to these criteria to decide on a final score, and mention whether you’re trying to score layouts on a 1930s typewriter, an ANSI-ish arrangement laptop scissor keyboard, a Model F, an Ergodox, or what. In particular, Dvorak tries to eliminate bottom row keypresses because they’re really bad on old typewriters. But IMO bottom row keys (especially for index/pinky, though B is a nasty stretch) are great on a flat keyboard.

If you have a particular keyboard in mind, the first step toward making an optimal layout is IMO to figure out the most comfortable placement of the hands. For example, if trying to make an optimal layout for an Apple laptop keyboard (since these can’t be easily changed in physical arrangement) and assuming we need to use the shift keys in their existing positions, I would start with a 'home' position of something like F–E–W-A and K–O–P–' (using QWERTY for the names of these keys), and then from that starting point figure out which keys on the physical keyboard as it exists today are easiest to reach with each finger, and then start assigning letters to the keys from there.

Alternately, if you have full control over the keyboard, there’s no reason not to use a column-staggered layout (with better stagger than the Ergodox) on split tented keyboard halves, in which case the bottom row is definitely easier to reach than the top row.
« Last Edit: Wed, 14 January 2015, 17:16:19 by jacobolus »

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #16 on: Wed, 14 January 2015, 21:09:39 »
Quote
OneProduct,mirror302.714 Total score  192.927 base        left right
                   1.598 SameFinger    20.786 Shift-SameFinger  upper  9.1 13.3
  qyoxj gwdlp*    71.206 Alternation   23.896 Shift-Alternation  home 33.9 30.3
  hieau mtsrn*     3.474 In/Out Rati    7.473 IndirSameFing     lower  4.9  8.5
  k**., bvfcz     10.651 adjacent       9.409 Shift-adjacent      sum 47.9 52.1
                   7.4  8.7 18.7 13.1 --.- --.- 17.1 12.4 12.9  9.8 Sh  1.8  1.0

ADNW(Eng, bigr)  262.617 Total score  182.722 base        left right
                   0.737 SameFinger     2.154 Shift-SameFinger  upper  5.5 13.6
  qyu.* zmldbp    68.832 Alternation   34.746 Shift-Alternation  home 40.0 31.1
  sieao hnrtcg     1.134 In/Out Ratio   6.990 IndirSameFing     lower  3.1  6.7
  j**,* fxwkv      9.123 adjacent      10.742 Shift-adjacent  sum 48.7 51.3
                  8.5  8.7 14.2 17.3 --.- --.- 16.4 11.5 13.1 10.3 Sh  1.8  1.0

I'll use ADNW bigram for comparison since it scored the best (edit: whoops, it actually scored worse but has a better base than ADNW trigram, my mistake).

I have a few qualms with this, though of course everything depends on how you do your measurements, so I'll explain where I disagree about the outcome here. Keep in mind that this is merely my subjective opinion and is just to promote thought on your part as it's interesting to see other people's point of views.

1. My layout has approximately 3.4 times the in/out ratio with only a minor amount more adjacency penalty relative to the in/out boost yet this doesn't seem to affect the score very much. Even compared to the eng/ger version my layout has 1.5 times the in/out ratio. What is the weight given to each stat?

2. Same finger usage is reduced primarily by removing P and G to pinky keys. If you had to place P and G optimally into the * slots ADNW's same finger usage becomes higher than that in my layout. Of course if I imposed this constraint on you you would clearly generate a different layout rather than forcing them into the current * slots. My main problem is that P and G are not infrequent keys (compared to ZXJQ for example, and even worse with the V and W of ADNW trigram!) and I find that having to press them with the pinky is orders of magnitudes worse than what is gained by putting them off to the side. Also, even with them off to the side we see very little in/out ratio gains from having them being so far outside (keys in outside positions have more potential to contribute to inward rolls).

3. As Jacobulus stated, I'm in disagreement with some of the rules for adjacency you use, as some of the motions which you deem uncomfortable I would deem comfortable and vice versa. However this is highly subjective so I wouldn't make it a huge point of contention.

4. As Jacobulus stated, and as championed by the author of the Workman layout, I believe that there are some keys on the bottom row which are more comfortable than those on the top row, yet it is evident to see by the way that certain letters are placed that ADNW does not consider that to be so. E.g. X being beneath N. I believe that the X position is a fairly good position. The picture here explains best by highlighting in yellow what could be considered to be favorable positions: http://www.workmanlayout.com/blog/. This impacts my base score negatively compared to ADNW. Furthermore if row usage is taken into account, me enjoying these bottom row positions causes a double negative impact.

5. My layout has one hand with the middle finger doing more work than the index finger, which we both favor. I'm not sure if that taken into account in scoring.

A few other random comments:

Do the extra pinky keys not cause a significant increase in penalty somewhere for the mere fact that they are not only pinky keys, but must be stretched towards laterally?

Do you have an ADNW variation that would work on a grid layout keyboard that does not have those extra pinky keys? For example on a Kinesis, Ergodox or Truly Ergonomic there aren't 1-unit keys in the ' and [ position of a QWERTY staggered keyboard. I would be interested to see what you would come up with if constrained to the same 30 positions I used.

As you stated, the distances are based on a staggered layout but I do my analysis based on a grid layout. I find it amusing to hear that flipping my layout changes its score. :)

I'm not sure I understand what the shift-X stats mean. Could you describe them? I think I understand shift-same finger being (relative QWERTY) holding left shift to shift a right hand key then needing to press A since left shift and A are both pinky keys. If shift-adjacent is based on adjacency rules like shift-->S being uncomfortable then I would argue against that stat as moving the pinky one row down to press the shift key has a considerable impact on which rolls are comfortable. For example, though I consider most rolls on the middle row beginning with A to be uncomfortable, I would consider all rolls on the middle row starting with left shift to be comfortable. Shift-alternation I don't have a clue about. :)
« Last Edit: Wed, 14 January 2015, 23:34:59 by oneproduct »
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #17 on: Thu, 15 January 2015, 07:05:32 »
@jacobolus and @oneproduct   - nice to have a technical discussion here.

I'll start with some general remarks, then get to your remarks.
•   I discovered ADNW as "the layout calculator that sucks less".  By the way, ADNW is a fork of the Neo layout (which is also on patorjk), the name comes from Aus Der Neo Welt, which is German for "From the Neo World".
•   I don't know exactly how totals are calculated, must look it up in the source code. I share your crititism about totals. The ADNW "crew" does too, that is why also the detail stats are given
•   Some also think that trigram analysis brings nothing over bigram analysis. (I DO think it is slightly better, though) 

•   Shift adjacent is a penalty for standard keyboards,  for a word such as (qwerty)  Pair  The left pinky holds the shift for the P, and then that same pinky types the a.  A combo like [Ps] would also be bad. 
•   Same finger is in a word as (qwerty) free F and R are type with the index. IndirectSameFinger is only in trigram analysis.  In a word like for the F and R are typed with the same finger, be it that the other had types an O in between. If you find that FOR (Lindex-Rring-Lindex) is harder to type than FOW (Lindex-Rring-Lring), then you agree with the concept of IndirectSameFinger.  In " Adnw circles" this is a debatable topic, just like the shift adjacent concept.
•   All these things are tweakable in the Adnw code. If you don't believe in it, you set the penalty to zero. I haven't worked with the most recent version of Adnw, which is easier since it allows changing settings at runtime. Meaning: no hacking in the source code needed.  It still is a " technical" program though, which you operate from the command line. On Linux and OSX this is straightforward, on Windows you'll need CygWin.
•   There are also versions for other boards, such as matrix boards, TEK or ErgoDox. See patorjk !

•   Most important: these are just numbers. Or, to put it diffently: the map is not the territory. These analyses are just tools for us to find a good keyboard, and nothing beats trying out in real life.


Now for some remarks.

The adjacent finger thing. This turns out to be very personal. Some people like certain " rolls", others are indifferent or hate them.  I myself like the standard ADNW view on this. I prefer pinky/middle over pinky/ring. I agree with jacobolus that middle/index is not bad at all.


Quote
"sequences like 5–3–4 or 3–4–3 basically need to be executed in two discrete chunks, because they require flexing and then relaxing the relevant muscles before flexing them again"
  True, I hate most in/out motions.


Quote
" the distribution of weight over the fingers. Optimal (says ADNW) is something like 8/10/18/14. Can you explain what these numbers mean? I’m interpreting it as pinky/ring/middle/index finger ratios. As in, among non-space characters, each pinky should get 8% of keypresses, ring finger should get 10%, etc."
  - Sorry, I was too short here. You interpret this correctly. If you don't count the thumbs, the four remaining fingers of a hand should do 50% of all key presses, the other hand the other 50%.  50% equally distributed over 4 fingers would be 12.5% per finger. But the pinky is weaker and index, middle are stronger. 


Quote
"The forefinger is much stronger and more agile than any of the other fingers, and controls more keys. Why in the world should it have a lower percentage of keypresses than the middle finger?"

Maybe I have given my own preference. My thumb is the strongest finger, and after that, in order:  middle- index - ring - pinky. Again, I must look up the standard " ideal"  values of ADNW - which are tweakable as well, BTW.


Quote
"In any event, I don’t think this weight distribution number is relevant per se. Any useful information it adds to a prediction of layout quality is already present in other better measurements, and most trade-offs that have to be made with other measures to improve this 'distribution' score are IMO likely to be net-negative."
- Agreed. And I found out that it doesn' t weigh heaviliy, the algorithm does give many layouts with " skewed" load distributions. Lower same finger etc. is much more important. 


Quote
"You should explain what weighting you apply to these criteria to decide on a final score, and mention whether you’re trying to score layouts on a 1930s typewriter, an ANSI-ish arrangement laptop scissor keyboard, a Model F, an Ergodox, or what. In particular, Dvorak tries to eliminate bottom row keypresses because they’re really bad on old typewriters. But IMO bottom row keys (especially for index/pinky, though B is a nasty stretch) are great on a flat keyboard."
Agreed. Must look this up!


Quote
"If you have a particular keyboard in mind, the first step toward making an optimal layout is IMO to figure out the most comfortable placement of the hands. " Agreed again! Although... one factor might be: do you want a layout that you can also use ona regular keyboard? If you want to bring along you own layout (on a USB stick: autohotkey script, karabiner, XKB maps etc) for use on other peoples'  PCs, then maybe a layout optimized for regular keyboard is better?


Quote
"Keep in mind that this is merely my subjective opinion and is just to promote thought on your part as it's interesting to see other people's point of views."
- Thanks for the courtesy, OneProduct  :-)  and no problem, this is a crazy hobby we share, right?  :-)


Quote
"1. My layout has approximately 3.4 times the in/out ratio with only a minor amount more adjacency penalty relative to the in/out boost yet this doesn't seem to affect the score very much. Even compared to the eng/ger version my layout has 1.5 times the in/out ratio. What is the weight given to each stat?"
  -  I must look this up! Your layout does extremely well on inrolls, which is (to me) important.


Quote
"2. Same finger usage is reduced primarily by removing P and G to pinky keys. If you had to place P and G optimally into the * slots ADNW's same finger usage becomes higher than that in my layout. Of course if I imposed this constraint on you you would clearly generate a different layout rather than forcing them into the current * slots. My main problem is that P and G are not infrequent keys (compared to ZXJQ for example, and even worse with the V and W of ADNW trigram!) and I find that having to press them with the pinky is orders of magnitudes worse than what is gained by putting them off to the side. Also, even with them off to the side we see very little in/out ratio gains from having them being so far outside (keys in outside positions have more potential to contribute to inward rolls)."
- I don't think I get the question... do you mean that for a fair comparison we should make layouts for equal physical layouts? ADNW uses more keys. This comes from the German background, the Germans use 4 extra letters (äöüß). The French Azerty layout uses the number row for é è â ç  which is a different solution.
In the ADWN layout, I don't like V being in such a lousy spot (qwerty Y). In your layout, the G is in that spot. The P is indeed not well placed either.  Typing PR doesn't seem comfortable either.... Perhaps this is the price for "too much" inrolls? 


Quote
"4. As Jacobulus stated, and as championed by the author of the Workman layout, I believe that there are some keys on the bottom row which are more comfortable than those on the top row, yet it is evident to see by the way that certain letters are placed that ADNW does not consider that to be so.... " - Agreed! I did experiment with other scores for the positions, the Adnw software allows for changing the "matrix" in which each key gets a base value. This leads to different layouts. However, as everything impacts everything, this sometimes mean worse scores regarding same finger, alternation, in/out etc.


Quote
"5. My layout has one hand with the middle finger doing more work than the index finger, which we both favor. I'm not sure if that taken into account in scoring."
- I don't think the finger load distribution counts heavily. I agree on the middle finger thing (strongest finger). jacobolus prefers the index. So this is personal.
I found that I value equal hand load (close to 50/50) more then ADNW does standard (can be tweaked of course). To me, a 47/53 keyboard is not nice.


Quote
"Do the extra pinky keys not cause a significant increase in penalty somewhere for the mere fact that they are not only pinky keys, but must be stretched towards laterally?"
- penalty I must look up. Like you, I feared it would feel bad, but it turns out fine! This is on a standard stagger board, meaning that the (right hand) top row sits slightly to the left. On a matrix keyboard this would be harder.  The pinky slide on the home row turns out to be easy as well.


Quote
"Do you have an ADNW variation that would work on a grid layout keyboard that does not have those extra pinky keys? For example on a Kinesis, Ergodox or Truly Ergonomic there aren't 1-unit keys in the ' and [ position of a QWERTY staggered keyboard. I would be interested to see what you would come up with if constrained to the same 30 positions I used."
- yes there is. The Ergodox version is on patorjk, other versions I can look up, or generate. See if I find time for that tomorrow :-)


Quote
"I'm not sure I understand what the shift-X stats mean. Could you describe them? I think I understand shift-same finger being (relative QWERTY) holding left shift to shift a right hand key then needing to press A since left shift and A are both pinky keys. If shift-adjacent is based on adjacency rules like shift-->S being uncomfortable then I would argue against that stat as moving the pinky one row down to press the shift key has a considerable impact on which rolls are comfortable. For example, though I consider most rolls on the middle row beginning with A to be uncomfortable, I would consider all rolls on the middle row starting with left shift to be comfortable. Shift-alternation I don't have a clue about. :)"
- I have explained that in the beginning of this megapost :-)  but I am doubting somewhat, so I'll have to look that up as well !

Now I need some coffee  :p :eek:
« Last Edit: Thu, 15 January 2015, 07:10:53 by PieterGen »

Offline juyanith

  • Posts: 21
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #18 on: Fri, 16 January 2015, 10:16:44 »
@oneproduct One of the things that I really like about Colemak is the placement of the common control keys (z, x, c, v). It's probably just from too many years of using them in the current (qwerty) location, but I like the fact that they are all on the left hand and bottom row. I use sequences of ctrl-c, mouse click, ctrl-v all the time. (I should probably also mention that I have ctrl on a thumb key so these are very easy one-handed key presses.) How would adding this constraint affect your layout? The simplest option would be to swap 'f' and 'x' but I don't know how that will change the overall results. Given how much work goes into creating these layouts I'm pretty sure that it would not be that easy however.

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #19 on: Fri, 16 January 2015, 14:08:17 »
It would certainly reduce the statistics slightly, but probably not all that negatively. I'm not 100% sold on my layout yet though and am still considering making some changes. :)

I'm also considering switching to a layout that uses E on a thumb key for kinesis/ergodox style keyboards so my attention is somewhat there at the moment.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline juyanith

  • Posts: 21
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #20 on: Fri, 16 January 2015, 19:58:29 »
I understand completely.  :))

I tried playing around with your layout by adding the constraints I mentioned and came up with the following:

p l d w g j f o y q
n r s t m u a e i h
z x c v b , . / ; k


It does score slightly worse, but perhaps not too bad. I think your layout gains a lot of alternation by having all the vowels on one side and then doesn't pay for it with heavy single finger bigrams because the punctuation is below them.

Offline berserkfan

  • Posts: 2136
  • Location: Not CONUS Not CONUS Not CONUS Not CONUS
  • changing diapers is more fun than model f assembly
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #21 on: Tue, 20 January 2015, 09:48:34 »
I'm awed by the quality of discussion going on, and tried out the MTGAP layout. Maybe I need much more getting used to, but I did not find it comfortable typing in English as much as my own modified Colemak. I think one factor is that I am more accustomed to using my index and middle fingers to move around, while the ring finger moves only up and down without moving sideways and the little finger is used to horizontal outward or downward movements. The MTGAP just doesn't seem to work well for the movements that I am used to making or maybe I just haven't tried too hard.

But I am ever reminded that these things are so subjective.
Most of the modding can be done on your own once you break through the psychological barriers.

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #22 on: Tue, 20 January 2015, 10:25:33 »
...tried out the MTGAP layout. Maybe I need much more getting used to, but I did not find it comfortable typing in English as much as my own modified Colemak....the MTGAP just doesn't seem to work well for the movements that I am used to making or maybe I just haven't tried too hard.   

Yes, on the one hand it's a matter of 'getting used to things'. On the other hand we all have our preferences - and that's a great thing or we would all be typing on the same keyboards! Also, I find it depends on your physical keyboard. Rolls can be comfortable on one keyboard but akward on another.

Offline batfink

  • Posts: 57
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #23 on: Wed, 21 January 2015, 11:09:00 »
...but I did not find it comfortable typing in English as much as my own modified Colemak.

I'm interested to know in what way you modified Colemak?

Offline berserkfan

  • Posts: 2136
  • Location: Not CONUS Not CONUS Not CONUS Not CONUS
  • changing diapers is more fun than model f assembly
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #24 on: Fri, 23 January 2015, 20:36:51 »
...tried out the MTGAP layout. Maybe I need much more getting used to, but I did not find it comfortable typing in English as much as my own modified Colemak....the MTGAP just doesn't seem to work well for the movements that I am used to making or maybe I just haven't tried too hard.   

Yes, on the one hand it's a matter of 'getting used to things'. On the other hand we all have our preferences - and that's a great thing or we would all be typing on the same keyboards! Also, I find it depends on your physical keyboard. Rolls can be comfortable on one keyboard but akward on another.

It suddenly occurred to me that because I am using a linear regular matrix keyboard (Tipro 128 keys), rather than your regular 104 key board, rolls are VERY different.
Most of the modding can be done on your own once you break through the psychological barriers.

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #25 on: Sun, 25 January 2015, 16:38:15 »


Made some small updates to the layout and started learning it. Old one on the left and new one on the right. At around 38 WPM at the moment and starting to get a good feel for it. I think it has good potential. Many of the common words that the typing tests start me off with are very comfortable to type.

And if you need to keep ZXCV locked you don't lose much:
« Last Edit: Sun, 25 January 2015, 16:47:43 by oneproduct »
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #26 on: Mon, 26 January 2015, 04:58:42 »
Could you tell a bit about the algorithm you are using?

It looks like a sane layout ! Hand balance (L/R) looks great too. Another important metric I learned from Carpalx: how the distribution over strings is. MTGAP states that words like (on qwerty) askl or port are the best to type, because they are 2 left, 2 right. Too much alternation (lapsos is slower, too little (sweaterdress) is slower too.

The general left/right statistics don't show this, because behind a nice 50%/50% load, there may be very different distributions.

Look at some imaginairy words. If have put L for all letters that are type with the left hand, and R for all letters that are typed with the right hand. Here are the words, 19 words in total: LLR LLLR LLR LLRRRRR LLR LR LLLLR LLRRRRRRR LLR LLLR LLLLR  RRRRRRL RRRRRLL LR LLR LLLR LR LRRR RLL

So, 40 times L and 40 times R. Meaning, a nice 50% - 50% balance. Total sind es 19 Wörter. The left hand types (cumulatively):
t1 - 22%
t2 - 72% (+50%)
t3- 89% (+17%)
t4- 100% (+11%)

So, from all strings on the left hand, 22% are just 1 letter. 72% of the left hand strings are only 1 or 2 letters. 89% of all left hand strings are 1,2 or 3 letters. All strings are 4 letters or less. There are no 5-letter strings on the left hand. 

On the right side things look differently.
t1 - 74%
t2 -74%(+0%)
t3- 89% (+5%)
t4- 89% (+11%)
t5-  95% (+0%)
t6 - 100% (+5%)

Most strings here are only 1 letter. But - 11% of the strings are 4 letters and 5% of the strings are 6 letters on a row, typed with the right hand.  In this case, the layout seems balanced at first look (a nice 50%/50% balance) but those 50% on the right hand consist of few, but very long strings. This feels like this:
lots of letters on the rigt hand - some left- some right - some left - lots of letters on the right hand.

The best thing is to try a layout out and see how it feels  :)  Next best is to make some metric, like above. One might even use this in the optimization itself, for instance a criteria might be "t3Left > 90%" and "t4Right >90%". Meaning you reject layouts that have too long one hand strings. Or - you give a penality for strings over 4 letters.

Offline vivalarevolución

  • Posts: 2148
  • Location: Naptown, Indiana, USA
  • Keep it real b/c any other way is too stressful
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #27 on: Mon, 26 January 2015, 07:43:43 »
Any chance that the Maltron layout could be added to this analysis? 

http://www.maltron.com/keyboard-info/the-maltron-letter-layout-advantage
Wish I had some gif or quote for this space, but I got nothing

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #28 on: Mon, 26 January 2015, 14:32:43 »
Hi prdlm2009 - Is having the E on a thumb key a good thing? Maltron's choice sounds logical, but the evidence is mixed.

I did found some stuff on the adnw.de website and the adnw google group.  I'll translate some findings:

There have been calculations on " improving Maltron". Maltron is ANISF DTHOR, and at least for the German language, better layouts are conceivable. All while leaving the E at the thumb key.

At the same time, it is clear that the E thumb key brings some weaknesses. 
- Alternation is rather low (only 54% of all digrams are alternating in the stock Maltron layout. This is comparable to Qwerty, but much lower than Dvorak and ADNW, that both are around 70%). The " improved Maltrons" are better, at 56-60%, depending on the language. But still lower).
- it scores worse at hand balance, adjacent keys (Dvorak-style layouts find adjacent keys bad, but Colemak sees it as good) and same finger rate
 
If you' re interested, I can dig deeper in it and find some stats, I don't have time for that now. The message is, anyway, that the E Thumb key has it's advantages is also weaknesses. No such thing as a free lunch, it seems!  :)
 


Offline juyanith

  • Posts: 21
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #29 on: Mon, 26 January 2015, 20:20:38 »
I don't know how much you will find this helpful but I thought I'd post it anyway. I've been using the MTGAP software with modified position values to see what comes up. The main changes I made were in modifying position costs so that finger movements to the bottom row are slightly easier than moving up to the top row. This is how it feels to me when using a column-staggered layout on my ergodox, especially for the ring and pinky fingers. Here are the relevant weights:

Code: [Select]
static int64_t costsCopy[KSIZE_MAX] = {
70,  50,  30,  40,  90,  90,  40,  30,  50,  70,
10,   5,   0,   0,  40,  40,   0,   0,   5,  10,
50,  40,  30,  30,  70,  70,  30,  30,  40,  50,
};

I also set the code up to allow the ZXCV keys to be anywhere on the bottom left row, as I really like to use them with my left hand.  The layout I end up with is:

Hands: 47% 52%
Fingers: 8.0% 11% 14% 14% 0.00% 0.00% 18% 16% 10% 8.0%

 W  G  D  M  J   ?  <  Y  B  :
 R  S  T  H  F   >  A  E  N  I
 X  C  V  L  Z   Q  O  U  P  K

 w  g  d  m  j   /  ,  y  b  ;
 r  s  t  h  f   .  a  e  n  i
 x  c  v  l  z   q  o  u  p  k

Fitness:       15296475
Distance:      12334240
Finger work:   0
Inward rolls:  9.03%
Outward rolls: 3.40%
Same hand:     36.04%
Same finger:   1.49%
Row change:    14.68%
Home jump:     1.68%
Ring jump:     2.29%
To center:     1.39%


For comparison, the last layouts presented by oneproduct are:

Hands: 52% 47%
Fingers: 10% 10% 13% 19% 0.00% 0.00% 14% 19% 9.0% 6.0%

 B  L  D  W  G   J  <  O  Y  Q
 N  R  S  T  M   U  A  E  I  H
 P  Z  C  F  V   X  >  :  ?  K

 b  l  d  w  g   j  ,  o  y  q
 n  r  s  t  m   u  a  e  i  h
 p  z  c  f  v   x  .  ;  /  k

Fitness:       20020690
Distance:      13510390
Finger work:   354280
Inward rolls:  4.31%
Outward rolls: 0.80%
Same hand:     29.27%
Same finger:   1.48%
Row change:    13.48%
Home jump:     1.02%
Ring jump:     6.12%
To center:     4.90%


Hands: 50% 49%
Fingers: 9.0% 10% 13% 19% 0.00% 0.00% 14% 19% 9.0% 7.0%

 P  L  D  W  G   J  <  O  Y  K
 N  R  S  T  M   U  A  E  I  H
 Z  X  C  V  F   Q  >  :  ?  B

 p  l  d  w  g   j  ,  o  y  k
 n  r  s  t  m   u  a  e  i  h
 z  x  c  v  f   q  .  ;  /  b

Fitness:       19312485
Distance:      13951180
Finger work:   0
Inward rolls:  4.32%
Outward rolls: 0.82%
Same hand:     30.51%
Same finger:   1.57%
Row change:    14.60%
Home jump:     1.18%
Ring jump:     6.93%
To center:     5.09%


MTGAP seems to really favor inward rolls and for reasons I don't quite understand it loves to arrange A, E, O, U in a grid rather than have them all on the home row. Perhaps it has something to do with the row change or home/ring jump penalties. Oh, and ignore the "Finger work:   0" stat. For some reason this seems to happen sometimes but I don't really know why.

I'm not sure what I'm going to do with this (if anything) but maybe it is useful as a comparison. I rather like the layout oneproduct has presented but I'm just not in a position to really try out either one at the moment.  :eek:

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #30 on: Tue, 27 January 2015, 01:12:44 »
Could you tell a bit about the algorithm you are using?

It looks like a sane layout ! Hand balance (L/R) looks great too. Another important metric I learned from Carpalx: how the distribution over strings is. MTGAP states that words like (on qwerty) askl or port are the best to type, because they are 2 left, 2 right. Too much alternation (lapsos is slower, too little (sweaterdress) is slower too.

The best thing is to try a layout out and see how it feels  :)  Next best is to make some metric, like above. One might even use this in the optimization itself, for instance a criteria might be "t3Left > 90%" and "t4Right >90%". Meaning you reject layouts that have too long one hand strings. Or - you give a penality for strings over 4 letters.

I did notice this string length concept while trying out my layout. I am beginning to think that there might be too much alternation in mine. I'll try to add something that measures this.

My algorithm is actually pretty simple and doesn't measure too many more advanced things like this. I've mostly just been measuring the things you see listed in the screenshot and assigning different weights to each one. However I feel that even a limited set of measurements like this tends to do the large part of defining what a good layout looks like. After that I hand tweak it based on perceived "comfort" and experiences using it to type common words.

I'm also having some doubts about how I feel about having WE IO on the top row be considered similarly to ASDF JKL;
Although they are indeed very comfortable to reach too, I find that moving between them and the homerow too often is somewhat disruptive and confusing. I think I may go back to putting all the major keys on the homerow (Colemak-like style) and then putting the next most frequent ones up there. Notably this means moving O down to the homerow and swapping M with L or D.

I'm very interested in this string length notion though. I'll start adding some measurements for that tomorrow if I can.
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #31 on: Tue, 27 January 2015, 05:15:40 »
MTGAP seems to really favor inward rolls
- you can set the penalties for in and out rolls at run time. Something like inRoll -20 outRoll 30  and so on.  AFAIK you cannot set penalties for use of adjacent fingers, so you'll see a lot of Colemak style rolls. Which you may or may not like (I don't, I prefer Ring-Index over Ring-Middle. Dr. Dvorak (and his brain grand childs, the ADNW layout and other Dvorak style layouts) thought the same, but it is a personal thing, and also dependent on the keyboard. On flat chiclet keyboards, Colemak rolls are fine). 
 
Quote
for reasons I don't quite understand it loves to arrange A, E, O, U in a grid rather than have them all on the home row. Perhaps it has something to do with the row change or home/ring jump penalties.
  I think it is those penalties indeed

Quote
and ignore the "Finger work:   0" stat. For some reason this seems to happen sometimes but I don't really know why.
Happened to me too when I optimized with MTGAP last summer. Michael, the dev of MTGAP thinks it is a bug somewhere in the code.

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #32 on: Tue, 27 January 2015, 05:20:52 »
My algorithm is actually pretty simple and doesn't measure too many more advanced things like this. I've mostly just been measuring the things you see listed in the screenshot and assigning different weights to each one. However I feel that even a limited set of measurements like this tends to do the large part of defining what a good layout looks like. After that I hand tweak it based on perceived "comfort" and experiences using it to type common words.

I understand you make the layouts by hand, and then score them using some formula with the criteria you mentioned? Scoring like that is fine. The problem with hand making layouts is however that everything influences everything, a computer can come up with layouts that you could never think of. You layouts seem good though  :thumb:

Quote
...I feel that even a limited set of measurements like this tends to do the large part of defining what a good layout looks like. After that I hand tweak it based on perceived "comfort" and experiences using it to type common words.
That is OK. All those numbers are just 'constructs', in the end it is about you typing on a keyboard and how that feels to you. 

Offline oneproduct

  • Thread Starter
  • Posts: 857
  • Location: Montreal, Canada
  • @Ubisoft
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #33 on: Tue, 27 January 2015, 07:56:50 »
I do actually use simulated annealing to generate the layouts algorithmically, but it's just based on the statistics you see. I do that in a different program that I made before making this UI program. I just do hand tweaks in the UI later, and since the algorithm used to generate them automatically only uses the stats that are visible to me in the hand tweaking UI, I can see all the things that I am affecting. :)

According to my algorithms, my hand tweaks actually make things "worse" but that's because the algorithm doesn't mention certain "comfort" related things (like which fingers take part in a roll for example).

The task I have right now is to add more kinds of statistics to measure for both algorithmic and hand tweaking purposes. :)
« Last Edit: Tue, 27 January 2015, 08:05:00 by oneproduct »
Layout: Colemak
Fastest typing speed: 131 WPM on typeracer, 136 WPM on 10fastfingers.
Daily driver: Filco Tenkeyless MX Brown with ergonomically weighted, lubed springs.
Ergo keyboards: Truly Ergonomic, Kinesis Advantage, Ergodox

Offline juyanith

  • Posts: 21
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #34 on: Tue, 27 January 2015, 08:12:00 »
@PieterGen Thanks for the info. I had already assumed that there was a bug in the code, but I don't really have a good way to try and debug it. I did consider changing the inRoll and outRoll parameters but I'm still on the fence about it. I think I rather like the inward rolls so I left things as they were designed in the MTGAP code. As an interesting side note I was watching my son practice his piano lessons the other day and noticed that there were a lot of outward rolls in what he was playing. Could it be that preference for (or against) rolls depends a lot on how well you have trained your fingers to use them? (I personally have no experience with the playing the piano so I can't speak to how relevant this may be to typing.)

Offline jdcarpe

  • * Curator
  • Posts: 8856
  • Location: Odessa, TX
  • Live long, and prosper.
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #35 on: Tue, 27 January 2015, 08:13:07 »
Anyone want to try this with Vibex's JD40 layout? :)

J W U P V   B Y M D K
A I S T L   R N E O
Z X C G [Enter] F H Q
KMAC :: LZ-GH :: WASD CODE :: WASD v2 :: GH60 :: Alps64 :: JD45 :: IBM Model M :: IBM 4704 "Pingmaster"

http://jd40.info :: http://jd45.info


in memoriam

"When I was a kid, I used to take things apart and never put them back together."

Offline juyanith

  • Posts: 21
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #36 on: Tue, 27 January 2015, 14:26:44 »
@jdcarpe I don't know if this will help much but here is the comparison using MTGAP (with my changes). I had to add keys for the blanks and used comma (,) for [enter] as it has a similar usage rate according to the data table in MTGAP. I don't think the comparison is really fair because of those differences and the layer usage though. I actually was playing around with my own code to try to compare keystrokes when using layers but it's not really ready for prime time, so to speak.

Hands: 48% 51%
Fingers: 8.0% 9.0% 12% 18% 0.00% 0.00% 19% 19% 11% 2.0%

 J  W  U  P  V   B  Y  M  D  K
 A  I  S  T  L   R  N  E  O  >
 Z  X  C  G  ;   ,  F  H  Q  ?

 j  w  u  p  v   b  y  m  d  k
 a  i  s  t  l   r  n  e  o  .
 z  x  c  g  ;   ,  f  h  q  /

Fitness:       29717909
Distance:      13861985
Finger work:   0
Inward rolls:  4.27%
Outward rolls: 1.95%
Same hand:     50.90%
Same finger:   8.36%
Row change:    25.26%
Home jump:     1.11%
Ring jump:     3.02%
To center:     13.80%

Offline jacobolus

  • Posts: 3634
  • Location: San Francisco, CA
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #37 on: Tue, 27 January 2015, 16:58:35 »
There have been calculations on " improving Maltron".
Is there more detail about what these “calculations” are based on? For example, when judging other keyboard layouts are they including spacebar presses when calculating “alternation”? If hand alternation including the thumb is going to be considered something worth measuring, then each layout should probably be analyzed at least twice, once assuming the use of the right thumb for spacebar, once assuming the use of the left thumb. Otherwise you can’t possibly get a fair comparison to the Malt layout which uses one thumb for E.

Quote
it is clear that the E thumb key brings some weaknesses. [...] - it scores worse at hand balance, adjacent keys (Dvorak-style layouts find adjacent keys bad, but Colemak sees it as good) and same finger rate [...] The message is, anyway, that the E Thumb key has it's advantages is also weaknesses. No such thing as a free lunch, it seems
This seems like arguments about the Maltron layout overall, not specifically about having E on the left thumb.

Offline Input Nirvana

  • Master of the Calculated Risk
  • Posts: 2313
  • Location: Somewhere in the San Francisco Bay area/Best Coast
  • If I tell ya, I'll hafta kill ya
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #38 on: Tue, 27 January 2015, 23:11:47 »
oneproduct: Really nice work!

And thanks to you other guys for really kicking in some great information, support and otherwise. Some heady stuff in this thread for layout geeks like me (and you). Even though I'm a die hard Kinesis Advantage user, this work still pertains to any physical configuartion.

Slightly off topic-
There was a thread, either on GH or maybe Colmak, about the Maltron layout efficiency (also including the "E" in the thumb cluster and the dramatic effects of that), If that sounds familiar to anyone please ping me.
Kinesis Advantage cut into 2 halves | RollerMouse Free 2 | Apple Magic Trackpad | Alphagrip | Colemak | on Mac+Hackintosh
Evil Screaming Flying Door Monkeys From Hell                     Proudly GeekWhacking since 2009
Staying in touch via Tapatalk VIP                                        Thanks much, Smallfry  
I AM THE REAPER . . . BECAUSE I KILL IT
~retired from forum activities 2015~

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #39 on: Wed, 28 January 2015, 05:55:13 »
@jacobolus
Quote
Quote from: PieterGen on 26-01-2015, 21:32:43
There have been calculations on " improving Maltron".
Is there more detail about what these “calculations” are based on? For example, when judging other keyboard layouts are they including spacebar presses when calculating “alternation”? If hand alternation including the thumb is going to be considered something worth measuring, then each layout should probably be analyzed at least twice, once assuming the use of the right thumb for spacebar, once assuming the use of the left thumb. Otherwise you can’t possibly get a fair comparison to the Malt layout which uses one thumb for E.

Here is the page. Please use Google Translate if you can't read German. By the way, my own language is Dutch. And Dutch-German is like French-Spanish. That is: related, but also very different  :(   

I stil haven't found time to look into the code and see how it is exactly calculated. jacobolus' remarks are justified.

Quote
Quote
it is clear that the E thumb key brings some weaknesses. [...] - it scores worse at hand balance, adjacent keys (Dvorak-style layouts find adjacent keys bad, but Colemak sees it as good) and same finger rate [...] The message is, anyway, that the E Thumb key has it's advantages is also weaknesses. No such thing as a free lunch, it seems
This seems like arguments about the Maltron layout overall, not specifically about having E on the left thumb.
  True

@Input Nirvana
Quote
There was a thread, either on GH or maybe Colmak, about the Maltron layout efficiency (also including the "E" in the thumb cluster and the dramatic effects of that)
Thanks, hope you'll find it !

Offline jacobolus

  • Posts: 3634
  • Location: San Francisco, CA
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #40 on: Wed, 28 January 2015, 15:14:07 »
Here is the page.
Yeah, so as they point out on that page, it’s hard to compare to other layouts because their analysis usually excludes spacebar (which is IMO a bad idea, but whatever). So it looks like they have one analysis excluding thumb keys entirely, one analysis excluding spacebar, and one analysis with both E and space.

They also mention that among the original goals of the Malt layout is to reduce error rate by separating commonly mixed up letters (like vowels) onto different fingers. Apparently in some study, 40% of errors on QWERTY (presumably on a standard keyboard) were substitution errors where the wrong key was pressed, and 20% were omitted letters; on the Malt layout, only 30% and 10% of errors were substitutions or omissions, respectively. Overall error rate was much lower, and most errors were transpositions. See http://web.archive.org/web/20071009094803/http://www.ergo-comp.com/articles/theeffect.html
(Actually, it looks like it was a combination of multiple studies possibly with different methodologies, and the paper is written by Malt, so I’m not convinced it’s a super reliable source. But maybe a reasonable starting point.)

I wonder if there have been any good statistics gathered about errors in ADNW, Colemak, Arensito, Capewell, etc. layouts.

It would be great to have some software set up to measure precise timing of various keypresses, so that we could actually see some evidence about which key combinations are slow or error prone on various layouts, and quantify what kind of time is lost to errors or slowdowns on particularly challenging sequences, as well as quantifying which sequences were fastest and most fluid. The fundamental flaw of all the various layout design heuristics I’ve seen is that there’s little concrete evidence from real typing to back them up.
« Last Edit: Wed, 28 January 2015, 15:25:13 by jacobolus »

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #41 on: Thu, 29 January 2015, 08:45:56 »
@jacobolus - I don't find the Malt study super reliable either and I don't see the problem of "commonly mixed letters". Of course, as with other aspects, data of real life typing is what is needed.

Problems I see with the Malt concept of "less errors by separating commonly mixed up letters":

1. The concept of "less errors by ..."  in itself. It is counter intuitive. So my "Null hypothesis"  would be that such effect does not exist :-)   But prove me wrong ! 

2. How do you reserach this? Even if I write with pen and paper, I make spelling mistakes, including mixing up letters. These mistakes would have to be subtracted, since they are not keybaord dependent. Also, there is the language-effect. In other languages I make more mistakes then in my mother language (Dutch). For instance, in English I tend to type keybaord You see I have the A and O mixed up. On a qwerty board the A and O are on different hands, and yet I mix them up anyway

And then there is also this weird effect that the Qwerty layout has influenced words and baby names ..... self fulfilling prophecy keyboard, anyone?

Dvorak-style layouts like Dvorak, Neo, Klausel, ADNW, Balance12, Carpalx have all vowels on one hand. I never felt I mix them up more then on a qwerty layout. I would be interested in data, to see if typers on layouts that have the vowels on both left and right hand make less vowel-mistakes. These layouts are among others Qwerty, Colemak, Arensito, Capewell, Norman, Workman.

3. layouts are compromises. This is yet another optimization factor, which will hurt other factors such as alternation, in/outward rolls, homerow jumps, same finger use, balance over fingers/ hands, etc. You see this already in the Malt layout, which has a much lower alternation then Dvorak. And those are only averages. Behind that is the distribution over 1,2,3,4,5,6 etc or longer strings on one hand. If vowels and consonants are on one hand, there is a much higher risk of typing long words or even several words with that same hand, which is tiring and unpleasant.

As carpalx states: QWERTY forces the typist to use the same hand repeatedly, which limits the amount of rest and increases effort. On a qwerty layout, 5% of all left hand runs are 6 letters or more ! Words like "dearest".  Those sort of words you will encounter much more on a layout that mixes vowels & consonants.

I'm not saying that mixing vowels & consonants is bad in itself. I do say that such a choice - like other choices - will come at a price.

The idea to put a much used letter on a thumb key seems sane though. Mtgap did some full keyboard optimization. You should do that for a board with thumb keys.

The logical order would then be:
1. build an ergonomic board (I call this the hard layout, so where the physical keys are)
2. make a soft layout (that is: what letters do you put on those keys). Ideally you optimize the whole board, including modifiers, number keys etc. Do so with as little as possible restrictions. So allow the letters to land on the thumb keys. 

The limit to this freedom is you memory and your sense of beauty. A layout like this can be nice and optimal, but is hard to remember and ugly:
x 3 6 5 q / " 9 2 8 0
u l c o ; v m d p ) j
a r s e , f h t n i -
( ' w . = k y g b _
4 1             z 7


So all in all, more research is needed !!  Actually, anyone looking for a PhD subject - this is a great one, right?
« Last Edit: Thu, 29 January 2015, 08:48:50 by PieterGen »

Offline jacobolus

  • Posts: 3634
  • Location: San Francisco, CA
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #42 on: Thu, 29 January 2015, 16:28:47 »
If vowels and consonants are on one hand, there is a much higher risk of typing long words or even several words with that same hand, which is tiring and unpleasant.

As carpalx states: QWERTY forces the typist to use the same hand repeatedly, which limits the amount of rest and increases effort. On a qwerty layout, 5% of all left hand runs are 6 letters or more ! Words like "dearest".  Those sort of words you will encounter much more on a layout that mixes vowels & consonants.
The word dearest on QWERTY is indeed pretty tough. By standard fingering, it’s 3-3-5-2-3-4-2. I find the fastest way for me to type that is with three phrases – d, e-a-r, e-s-t – each of which uses all separate fingers, so doesn’t require anything to be lifted in between. For me, it’s hard to shorten the time gap between d and e, and especially the gap between e-a-r and e-s-t. If there were a couple of letters typed with the other hand in between that would be a big overall speed boost for me.

Some motions on one hand can be really fast though. For instance, using standard fingering, the sequence p-o-i-n can be typed as a phrase in one motion very fast with no finger reaching in the middle of the motion. So a word like, say, 'pointing' is extremely quick and easy to type, because it can be broken up as basically two phrases p-o-i-n+t, i-n+g with only a short gap in between them. However, because none of the letters in 'pointing' are in the 'home' positions on the QWERTY keyboard, it would get scored extremely harshly by pretty much every layout scoring heuristic I’ve seen.

If someone wanted to do a real well-funded study here, I’d love to have actual measurements of a few dozen experienced typists on different layouts, ideally with precise/accurate (to the 10s of milliseconds, say) timings for every key typed, and ideally even a top-down video of their hand motions, so the fingering for specific sequences could be investigated. It would be great to have some kind of real quantification of how much slowdown is caused by each typing error, which sequences are fast or slow for different typists, which words are fastest or slowest from one layout to another, etc.

All the studies I’ve ever seen used very crude measurements, and all the alternative layouts were developed using a bunch of unproven conjecture about what features of a layout might be good or bad, fast or slow. Some of the built-in assumptions may be outdated, for instance Dvorak was designed based on the physical keyboard of a manual typewriter, for which typing on the bottom row requires moving the whole hand down and back (so Dvorak tries to avoid the bottom row as much as possible). On any computer keyboard, the bottom row is much easier than that to reach, though it’s obviously easier for the index fingers than others; on a completely flat laptop keyboard, or for instance using flat profile keycaps like DSA, the top row is relatively more challenging to reach than it would be on a typewriter, while the bottom row is relatively easier.

Indeed, for a completely flat laptop keyboard, it no longer makes much sense IMO to think of separate 'rows' at all, and we could I believe get better layouts by starting with a 'home' finger position something like (QWERTY) f-e-w-q and k-o-p-[ – though probably not centered precisely on those keys – next figuring out which key could be most easily pressed by each finger, and then finally arranging the letters to be most convenient from that baseline (some of them on the QWERTY number row). With such an arrangement, a whole new set of heuristics would be required, compared to the ones used to calculate all the alternative keyboard layouts I’ve ever seen. Depending on the keyboard firmware or the operating system, this might not work perfectly (we might not have full control to change the position of all the modifiers for example), so a different home position might need to be chosen.

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #43 on: Fri, 30 January 2015, 09:07:52 »
@PieterGen I did consider changing the inRoll and outRoll parameters but I'm still on the fence about it. I think I rather like the inward rolls so I left things as they were designed in the MTGAP code.
  I did play with it, changing InRoll, outRoll and sameHand - everything comes at a price though. More nice rolls = more ugly rolls as well, usually. Like medication: the stronger the effects, the stronger the side effects :)

Could it be that preference for (or against) rolls depends a lot on how well you have trained your fingers to use them?
I think so! Also, how long are you fingers, how is your keyboard etc. To me, outrolls are not so bad. What I do not like are "broken rolls" such as (qwerty) WRE  Even worse are things like WREST

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #44 on: Fri, 30 January 2015, 09:10:41 »
Anyone want to try this with Vibex's JD40 layout? :)

J W U P V   B Y M D K
A I S T L   R N E O
Z X C G [Enter] F H Q

Interesting layout! For fun I shall calculated how it scores (using the ADNW scoring algorithm - no absolute truth!  You could also score it in Patorjk 's website)

Offline vivalarevolución

  • Posts: 2148
  • Location: Naptown, Indiana, USA
  • Keep it real b/c any other way is too stressful
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #45 on: Tue, 03 February 2015, 16:28:33 »
Hi prdlm2009 - Is having the E on a thumb key a good thing? Maltron's choice sounds logical, but the evidence is mixed.

I did found some stuff on the adnw.de website and the adnw google group.  I'll translate some findings:

There have been calculations on " improving Maltron". Maltron is ANISF DTHOR, and at least for the German language, better layouts are conceivable. All while leaving the E at the thumb key.

At the same time, it is clear that the E thumb key brings some weaknesses. 
- Alternation is rather low (only 54% of all digrams are alternating in the stock Maltron layout. This is comparable to Qwerty, but much lower than Dvorak and ADNW, that both are around 70%). The " improved Maltrons" are better, at 56-60%, depending on the language. But still lower).
- it scores worse at hand balance, adjacent keys (Dvorak-style layouts find adjacent keys bad, but Colemak sees it as good) and same finger rate
 
If you' re interested, I can dig deeper in it and find some stats, I don't have time for that now. The message is, anyway, that the E Thumb key has it's advantages is also weaknesses. No such thing as a free lunch, it seems!  :)

Ok thanks.  I just saw your response.  Your analysis so far is more than helpful.  Maybe it is time for me to learn another new layout, or modify my current layout.

My personal preference is to stay away from the bottom row and heavy pinky usage as much as possible, so those are my main goals with any layout.  The only fingers that seem to like the bottom row are my index fingers.  The rest put up a hard fight.
Wish I had some gif or quote for this space, but I got nothing

Offline PieterGen

  • Posts: 135
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #46 on: Wed, 04 February 2015, 05:34:59 »
@prdlm2009 - Seeing you are in Indiana, USA, I assume you write mostly in English, which is a big advantage, because many layouts were specifically optimized for English. For instance Colemak. You may have less need for a custom layout. I had to make a custom one, because I type mostly in Dutch, for which "English" layouts are suboptimal.

As you know, on patorjk's keyboard analyser is an easy way to see how well various layouts work for your texts. One may argue with pat's way of calculating things, but it is a nice start and very easy to use. This one is very nice too: a website where you can  try out several layouts in the browser. Recommended ! If the existing layouts don't work well enough for you, you can always modify one or calculate a completely custom one.

BTW, I too hate to use the pinky & ring fingers on the bottom row. On a standard qwerty keyboard I don't use the left pinky on the bottom row. W S and Z are typed with the ring finger.  I hate the shifts on the pinkies as well. Maybe I should try out that spacebar/shift idea again:
- Press space, keep it pressed and type a letter (let's say abc) = ABC
- Press space = nothing happens
- Release space = space

I tried it before, what I liked was being able to shift with thumbs, what I did not like was that space comes slow, it only appears on the release of the space bar, which "feels" slow.
« Last Edit: Wed, 04 February 2015, 05:36:59 by PieterGen »

Offline vivalarevolución

  • Posts: 2148
  • Location: Naptown, Indiana, USA
  • Keep it real b/c any other way is too stressful
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #47 on: Wed, 04 February 2015, 11:37:47 »
@prdlm2009 - Seeing you are in Indiana, USA, I assume you write mostly in English, which is a big advantage, because many layouts were specifically optimized for English. For instance Colemak. You may have less need for a custom layout. I had to make a custom one, because I type mostly in Dutch, for which "English" layouts are suboptimal.

As you know, on patorjk's keyboard analyser is an easy way to see how well various layouts work for your texts. One may argue with pat's way of calculating things, but it is a nice start and very easy to use. This one is very nice too: a website where you can  try out several layouts in the browser. Recommended ! If the existing layouts don't work well enough for you, you can always modify one or calculate a completely custom one.

BTW, I too hate to use the pinky & ring fingers on the bottom row. On a standard qwerty keyboard I don't use the left pinky on the bottom row. W S and Z are typed with the ring finger.  I hate the shifts on the pinkies as well. Maybe I should try out that spacebar/shift idea again:
- Press space, keep it pressed and type a letter (let's say abc) = ABC
- Press space = nothing happens
- Release space = space

I tried it before, what I liked was being able to shift with thumbs, what I did not like was that space comes slow, it only appears on the release of the space bar, which "feels" slow.

Thanks for the reply and links.  I never considered the advantage that I might have for typing in English.  I do like a custom layout that does allow to keep my fingers on the home row for the most common letters, and QWERTY does an awful job of accounting for that preference.  Anything that can be done to reduce hand and finger fatigue is helpful.

The Maltron layout, to my knowledge, was designed to maximize the amount of words that can be typed on the home row, as to minimize vertical finger movement and finger fatigue.  Other considerations, like alternation, same finger usage, and inward/outward rolls did not seem like they were considered when the layout was designed.  Some parts of the Maltron layout, like "L" in the upper right corner, seem like poor choices.

I created a modified Maltron layout last night with E on the home row, among other changes,  so I don't have to completely relearn a new layout in the short-term, and I will see how that works for my needs.  Alternation, same finger, rolls, those things I have not considered for this layout.

The shift on the pinky is an annoyance.  I have experimented with different strategies for the shift, such as moving it up a row, putting it on the bottom row thumb key on a custom keyboard with a more robust bottom row.  But I keep it on the pinky out of habit.  My favorite shift placement?  2 unit vertical key on the Maltron keyboard.  That way, the whole weight of the pinky can be used for shift.
Wish I had some gif or quote for this space, but I got nothing

Offline Steven Tammen

  • Posts: 11
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #48 on: Sat, 25 July 2015, 22:59:49 »
Oneproduct:

Sorry to necro this thread, but I’m very, very interested in hearing how things are going now that you've used your layout for a while (like so interested that I made a GH account specifically so I could join in on this thread). I also have a couple of questions related to your layout, and hopefully you can help me understand.

I’ve been doing tons of research on keyboard layouts since I want to learn touch typing and get away from my QWERTY hunt-and-peck. I’ve looked at and experimented with a whole bunch of layouts (Dvorak, Colemak, MTGAP, the Carpalx ones, Arsentio, Capewell) and yours seems to be the one I’m liking the most because of its combination of super high alternation with good inward rolls. From my experience and research, it seems like travel distance is overemphasized in most analyses (e.g., Carpalx), as is finger balance (to an extent). I think things that need to be optimized for go in the order of:

1. low same finger in the typing of digrams

2. high alternation AND inward rolling (not one or the other but both). I personally like middle-index and pinky-index rolls best, and then ring-middle after that. I’m not as big fan of pinky-ring or pinky-middle, but I think the latter one is less uncomfortable than the former.

3. low finger travel distance

4. low outward rolling, with the exception of index to middle (which I have found to be significantly less bad than other outward rolls).

5. low trigram “cycling” on same hand typing,  e.g., “afs” or “ljk” on QWERTY. This is because such movements necessarily involve opposite directional motion and are easy to mess up.

6. Finger layout preferences (probably the most subjective heuristic). I’m going to go through mine below.

- I like workman-style positions: bottom row for index (without extension), top row for middle/ring. I find this makes more of a difference on pure staggered layouts than matrix layouts (TypeMatrix, Planck) and staggered column layouts (ergodox, Kinesis)—but it is still more favorable even on columnar layouts since it seems to conform more to natural hand positions.
 
- I also favor bottom row pinky versus top row pinky (for similar reasons as the workman finger preferences—better hand position overall), something I was pleased to find holds in your updated layout (i.e., P, K have higher occurrence frequencies than their respective top counterparts, B, Q).

- In terms of lateral index extension versus top row index extension, I think lateral (homerow) extension is better (i.e., “g” in QWERTY is better than “r”). This holds for W and M on your layout, and perhaps U and , on the other side too (I’ll talk about that again later).

- Finally, I find the bottom row positions for the ring finger to be the second worst space on the board behind top row pinky, which is reflected in your layout too (with Z and ?). The ring finger is the most “dependent” of all the fingers, and getting it to go down smoothly without messing with the middle finger, e.g., is something I struggle with more than the bottom corner with the pinky.

Overall I was super psyched when I saw your layout because it follows my finger preferences to the letter. I think my reasons for having these preferences are more than simply subjective, but each person has their own comfortable positions for their particular hand size and morphology. I was just surprised to find a fully optimized layout that perfectly lined up with mine.

7. Balancing finger load. I agree with you and PieterGen that the middle fingers are “underrated,” and feel that layouts should have pretty even balance between middle and index fingers (lower frequency letters on the index reach keys, e.g., “b” and “y” in QWERTY, with higher frequency letters on “i” and “e” QWERTY keys, should lead to pretty balanced loads). Your updated layout has the index doing more on the left, and then the middle doing more on the right (probably because that’s where E is), but it’s not crazy unbalanced on either side. The pinkies shouldn’t get quite as much as the rest, but they shouldn’t be nearly as babied as they are in the QGMLWB layout from Carpalx, in my opinion, where they get about half as much as most other layouts..

8. Finally, my last heuristic is balanced hand load. In my opinion this is not particularly useful on its own because you can have balanced-ish hand loads with low alternation (a la Colemak), but there’s no need to try to unbalance it for any reason. It should be “good enough” naturally as a by-product of optimized hand alternation.

Feel free to comment on any of the heuristics—good, bad, or otherwise. This is just what I’ve come to from my own experimentation and through a lotta hours of research. I’m now going to move into some of the questions I had regarding your layout. Most of these are from things you mentioned in the thread, but a couple come from me.

1) At the top you say “I also made a few changes where the layout became more similar to some other layouts (for familiarity) where there was a very minimal decrease in heuristic viability.” Could you tell me what these changes were? IMO, it doesn’t seem worth compromising efficiency to make transition easier for some small demographic of people, but I don’t want to misjudge what you’re saying. I hope the changes were really small, otherwise the perfectionist part of me is going to have a fit  ;D

2) Later on you also say “While trying my layout now I just realized that it's fairly uncomfortable to do 2-grams with H(vowel) which are very common. That's likely why MTGAP puts N with the vowels instead of H. I might try to fiddle around with something similar.” Did you ever go anywhere with this? I saw later that you said “but any consonant that you put with the vowels other than H results in around 5% less alternation,” which seems like an acceptable reason to leave it as is. There will always be tradeoffs.

3) Did you ever compare your layout to “Balance 12” by Sasha Viminitz (or the updated version)? You can see the layouts down in the comments on this page here: https://mathematicalmulticore.wordpress.com/the-keyboard-layout-project/. I’ve copied the two I am interested in below. These were the only other layouts I was seriously considering other than yours and MTGAP, but since you’ve already made a good case why yours is better than MTGAP, I would be curious to see how these two stack up. She said she tried to optimize hand alternation and inward rolls at the expense of finger/hand balance, which lines up with my heuristics, as does your design philosophy.

Balance 12:   

\ / # $ % ^ & * _ +
B Y O U @ K D C L P Q { }
H I E A ; M T S R N V
X [ ] : ! W G F J Z

1 2 3 4 5 6 7 8 9 0 – =
b y o u ‘ k d c l p q ” —
h i e a , m t s r n v
x ( ) . ? w g f j z

Updated Balance 12:

\ / # $ % ^ & * _ +
B Y O U @ V C D L P Q { }
H I E A ; F S T R N K
X [ ] : ! W G M J Z

1 2 3 4 5 6 7 8 9 0 – =
b y o u ‘ v c d l p q ” —
h i e a , f s t r n k
x ( ) . ? w g m j z

4) Did you ever research punctuation frequencies? I haven’t been able to find any good information on the internet about them, but it is the only place where you might possibly change something on your layout for the better, depending. There were some possibilities I thought of, so see what you think:

- Case 1: The comma has higher frequency than U. In this case, you could try swapping the U and the comma on your updated layout, which would give you the nice ou digram roll like MTGAP has. I was really curious if you had tried this, because that roll was something that made MTGAP look appealing. It would also put “you” as one nice inward roll, and IIRC you mentioned that you liked having that word easily typable for IM’ing and whatnot.

- Case 2: The comma has a higher frequency than the period. Since we are in agreement about the workman positions, this would seem to suggest that you should swap the comma and the period since the bottom index position is superior to the top index position.

- Every other case: Anything else would really depend on how valuable/harmful switching for the ou digram turns out to be.

I’m sorry this got long, but I’m really interested in this topic, and I don’t have the technical wherewithal to run the optimization algorithms myself (I don’t even know how to pull something from github :-[). I don’t want to dump work on you just for my benefit, but if you are interested, could you answer some of my questions anyway? It’s for the greater good of humanity ;D

Offline Snarfangel

  • Posts: 276
Re: Layout Analysis: Dvorak/Colemak/MTGAP/Carpalx/Workman/my own
« Reply #49 on: Sun, 26 July 2015, 00:29:01 »
I'm interested as well. Since the last update by Oneproduct, I've gotten my hands on a Maltron, and am attempting to convert it to this:

106550-0

I still need to program Soarer's converter, do some minor wiring (if I can puzzle it out), and change the number pad, but I've altered this:

106552-1

Into this:

106554-2
« Last Edit: Sun, 26 July 2015, 00:30:37 by Snarfangel »