Author Topic: Balanced Keyboard Layout  (Read 54267 times)

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Re: Balanced Keyboard Layout
« Reply #1200 on: December 31, 2017, 11:36:54 PM »
You really put in a lot effort for this.

You're okay with Den3 at the current state?

I accept your reasoning behind it (ie punishing pinky usage). I may not agree with the exact weightings but have not looked at it hard enough to debate it.

Think it will be more useful to first run all the layouts through all the tests and see how they stack up... if for example, QWERTY suddenly comes out ahead of Dvorak, then we will need to talk... :-)

First need to get it running locally and then see where I need to put in extra "write to console" stuff if necessary, to get the data and error conditions I need. Should only be in one or two places I think.

I think I will change the weightings in the "average" calculations, at the moment each score is added once, except for Patrick English, and Den 1 English, which get added twice each, in an attempt to balance the numerous word-based scores.

Think I will change that to Patrick English x 2, Den 1 English x 3, and Den 3 English x 1, because Den 3 already includes some word-based metrics in its score.

My gut feel is still that Den 1 is "most accurate", in being largely agnostic about any particular design philosophy (where Den 3 takes the deliberate position against pinkies. There may be some people (eg Colemak with both A and O on pinkies) who would challenge this.)
But my gut has been known to be wrong :-)

Cheers, Ian
« Last Edit: January 02, 2018, 03:52:31 AM by iandoug »

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Same finger bigrams
« Reply #1201 on: January 02, 2018, 02:28:52 PM »
Hi

Added another metric of my own invention, which possibly has some similarities to methodologies used by the scientists in one of those studies comparing QWERTY with Dvorak etc.

It's based on the probability of one given letter following another ( http://www.prooffreader.com/2014/09/how-often-does-given-letter-follow.html ) (took me ages to manually type all those probabilities).

Anyway, we see which letters are on a given finger, then look up the probability in both directions for all letters on same finger (ie so we add "er" and "re", for example), and add them all up, and rank the results.

I ignored Capital letters, and also when both letters were on same key (as in CPL layouts), since the idea is to see how much a given finger has to fly around.

No distances were harmed in this production. That's under ToDo, with Row Jumps, Rolls, etc... when I figure out some way to score it.

Anyway the top end looks like this:


1   seelpy-1-1.en.ergolinear   15.48
1   seelpy-1.en.ergolinear   15.48
3   seelpy-1-4.en.ergolinear   15.61
4   beakrak-1-2.en.ergodox   16.06
5   hieaqmtsrn.en.ansi   16.39
6   proxkb.en.ergodox   17.19
6   proxkb-thumbshift.en.ergodox   17.19
8   maltron-us-90-mod-Andreas.en.ergodox   17.51
9   hieamtsrn.en.ansi   17.61
10   dvormax.en.ansi   17.77
11   opuiany.en.ergodox   19.14
12   balance-twelve.en.ansi   20.60
13   seelpy-1-8.en.ergolinear   21.03
14   beakl-clp-0.en.matrix   21.27
15   widely-alternating-A.en.ansi   21.33
16   beakl-modified.en.ergodox   21.88
17   mtgap-mod-joey2216.en.ansi   21.97
18   x1-ou.en.ergolinear   22.71
18   m2-tweak.en.ansi   22.71
20   maltron-us.en.ergolinear   22.86
20   maltron-us-90.en.ergodox   22.86


and the bottom end like this:


354   a-joy.en.ansi   238.65
354   a-joy-rehomed.en.ansi   238.65
356   qwerty-mod-noah.en.ansi   239.73
357   blick.en.ansi   241.01
358   neo2.en.ansi   242.76
359   ina-dv.en.ansi   247.47
360   ward.en.ansi   247.70
361   hoke.en.ansi   255.81
362   reverse-qwerty.en.ansi   258.62
363   rhythmic.en.ansi   269.31
364   tyler.en.ansi   270.50
365   qwerty-q-layout.en.ansi   272.72
366   cyberswarm.en.ansi   274.09
367   dvorak-onehand-right.en.ansi   275.74
368   wong-poiea.en.ansi   293.10
369   trotman.en.ansi   295.47
370   culemak-mod-Ian-hp.en.ansi   299.42
371   as-in-red-hot.en.ansi   308.38
372   mcgunnigle-peoples.en.ansi   343.95
373   tnwmlc.en.ansi   366.12
374   dvorak-onehand-left.en.ansi   384.13


For a change the bottom end is not QWERTY and Alphabetics-dominant.

The Famous Brands stack up like this: (all standard ANSI variants)


Colemak 46.68
Klausler 60.06
Dangvu 64.53
MTGap 70.13
Workman 81.62
Capewell 94.94
QGMLWY 109.93
Dvorak 129.02
Norman 162.00
QWERTY 181.65


Gory details here: http://www.keyboard-design.com/same-finger-bigrams.html

The details of how the score was arrived at are on each layout's page.

The graph showing the spread over all the layouts is actually the smoothest progression of the lot. Which kinda gives me some sort of confidence in the metric... :-) (FWIW)

Cheers, Ian
« Last Edit: January 02, 2018, 02:40:00 PM by iandoug »

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Re: Same finger bigrams
« Reply #1202 on: January 03, 2018, 01:22:28 AM »
Added another metric of my own invention, which possibly has some similarities to methodologies used by the scientists in one of those studies comparing QWERTY with Dvorak etc.

The Famous Brands stack up like this: (all standard ANSI variants)


Colemak 46.68
Klausler 60.06
Dangvu 64.53
MTGap 70.13
Workman 81.62
Capewell 94.94
QGMLWY 109.93
Dvorak 129.02
Norman 162.00
QWERTY 181.65


After a restless night of tossing and turning, I finally gave in to the voice at the back of my head and changed the calculation a bit.

Consider qu and et. The chances of u following q, and t following e, are:

qu: 98.7
et: 2.66

So if a layout had qu on same finger, I would have added 98.7 to the score, and only 2.66 if et were on same finger... but clearly having et on same finger is much larger problem than qu, since et are two most common letters.

So now I multiply the above numbers by the frequency of the first letter in English, so we have

qu: 98.7 * 0.1 == 9.87
et: 2.66 * 12.1 == 32.186

which is a much better reflection of reality.

So the Famous Brands now stack up as follows:


Colemak 82.48
MTGap 93.05
Capewell 128.75
Klausler 181.35
Dvorak 196.04
Dangvu 204.9
Workman 212.57
QGMLWY 350.73
Norman 503.58
QWERTY 525.73


The relative percentages are going to have a much wider range, since the scores now range from 8.01 to 1,757.11.
« Last Edit: January 03, 2018, 12:03:17 PM by iandoug »

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Corpus update
« Reply #1203 on: January 03, 2018, 12:09:57 PM »
Hi

FWIW an update to the corpus used in tests.

At the moment, for comparison purposes, I am only using the files in EnglishTexts and Programming.

The WordLists, MixedLanguages, and DigitsEtc are useful for fine-tuning particular aspects of a layout, or just cross-comparisons to things non-English or rare English. But don't think it's fair to compare layouts using them as inputs.... my focus is on English and Programming. I believe if you're going to enter hundreds of digits, you should use a numpad....

These versions SHOULD be free of non-ANSI/ASCII characters, ie only stuff found on normal ANSI 104 keyboard. Think one of programming files still has two(?) arrows (or something) hiding somewhere.

The home-alone.txt is the list of words from the various home-key, home-block, easy-block and one-handed tests... the sample long words from the various layouts, deduped. (home == home key/block, alone == one-handed). Yes I know my sense of humour is weird.

Cheers, Ian
« Last Edit: January 03, 2018, 12:20:39 PM by iandoug »

Den

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 1089
  • Selina is my Superstar
    • View Profile
    • Amuseum
Re: Balanced Keyboard Layout
« Reply #1204 on: January 03, 2018, 11:40:14 PM »
Numpad revisited. Prioritize 1, 0 at the strongest fingers.

Code: [Select]
   +=*
   8236
 -.9014
  ,:57/
Support me on Patreon

I saw. I conquered. I came.

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Re: Same finger bigrams
« Reply #1205 on: January 04, 2018, 02:57:19 PM »

Colemak 82.48
MTGap 93.05
Capewell 128.75
Klausler 181.35
Dvorak 196.04
Dangvu 204.9
Workman 212.57
QGMLWY 350.73
Norman 503.58
QWERTY 525.73


Voice in my head was still complaining... something about finger usage.... so I did a quick experiment, and multiplied each letter pair on same finger by a weighting per finger (simplified model: thumb = 1, pinky = 5).

Overall resulting order was sorta the same, but some differences, here's the revised list:


Colemak 210.78
MTGap 264.3
Capewell 377.79
Klausler 440.98
Dangvu 482.96
Dvorak 523.69
Workman 567.26
QGMLWY 802.54
QWERTY 1426.34
Norman 1446.09


Which is not bad, and except for QGMLWY, compares order-wise with the same-finger list (here in reverse order) on https://normanlayout.info/compare.html


    Norman: 4.7%
    QWERTY: 4.6%
    Minimak: 3.1%
    Asset: 2.6%
    Workman: 2.4%
    Dvorak: 2.2%
    Klausler: 1.6%
    Capewell: 1.4%
    Arensito: 1.3%
    Colemak: 1.3%
    QGMLWY: 1.3%


Norman is known to have bad same-finger metric.
I assume QGMLWY is out of order for me because of the finger weighting.
It's score was calculated thus: (first digit is KLA finger number)

1 => 'qdz': 0.42
2 => 'gsx': 20.0084
3 => 'mtc': 94.0185
4 => 'lwnrvj': 111.011
5 => ' ': 0
6 => '': 0
7 => 'yfiakp': 447.6328
8 => 'ue': 31.4118
9 => 'bo': 98.0376
10 => 'h': 0
11 => '': 0

Cheers, Ian
« Last Edit: January 04, 2018, 03:01:57 PM by iandoug »

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
GhettoDox
« Reply #1206 on: January 05, 2018, 04:12:26 AM »

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Keyboard for mouths
« Reply #1207 on: January 13, 2018, 07:35:41 AM »
Stumbled across this:

http://www.musanim.com/mam/qhkcstfpn.html

Tried playing around with it but requirement of having vowels and liquids on home row works at odds with putting common letters on home row. T, N and D can easily be put in better (but not optimal) spots.
Also have H and E on same finger is never going to work.

Came to the conclusion that it may be comfortable for the mind, but we type with fingers not brainwaves, so comfortable for fingers is more important.
« Last Edit: January 13, 2018, 09:41:30 AM by iandoug »

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
not good
« Reply #1208 on: January 16, 2018, 07:26:15 AM »
smashed my left wrist. back on ansi keyboard and mouse on right hand and mostly one handed typing for now. At least it has mech switches (gateron red I think)

Den

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 1089
  • Selina is my Superstar
    • View Profile
    • Amuseum
Re: Balanced Keyboard Layout
« Reply #1209 on: Yesterday at 12:57:24 PM »
Sorry to hear that.

Are you now gonna devise the perfect one hand layout? For that, I suggest sticky modifiers--so you don't have to hold down shifts, which would demand dexterity.

iandoug

  • Hero Member
  • *****
  • Posts: 752
    • View Profile
    • Keyboard Design
Re: Balanced Keyboard Layout
« Reply #1210 on: Yesterday at 01:44:39 PM »
Sorry to hear that.

Are you now gonna devise the perfect one hand layout? For that, I suggest sticky modifiers--so you don't have to hold down shifts, which would demand dexterity.

I did wonder if the universe was pointing me in that direction.

More worryingly so because a few days ago I had been thinking about one-handed layouts for some reason ... but more along the lines of "I'm glad I don't need to work like that" .... and now here I am doing that...

I'll see if the universe throws any bright ideas my way over the next few weeks....

Cheers, Ian