Yet another keyboard layout: eNNe

[Sc0tTy]

Updated:
By now Ian and me have made a lot of changes to the layout and word lists which resulted in big advances
I will be keeping this opening post unchanged but the name BoeKoe is no more, its now eNNe.

Enjoy the discussion

Original post:
Hi there,

So yes, this is yet another keyboard layout... And with all layouts, this one fills some specific needs (and comes great responsibility).
In particular this layout is to be used on an ErgoDox keyboard.
The layout I will be introducing shortly is developed for typing English and Dutch (about 50/50 share) and programming in C#, JavaScript, LESS and HTML.

Before I start I would like to clarify some things.

Introduction
I've been typing Colemak for over 5 years and recently one of my TEK's started failing.
So I had to look for a new keyboard and a long story short I purchased a ErgoDox EZ.
One of the downsides of this keyboard is that it has fewer keys than the TEK and moved some other keys around, this forced me to rethink my layout.
Firstly what I did is modify Colemak so that it wouldn't be a problem but I also really felt the need to tweak the layout because some fingers were being overused (mainly my right index finger).
This brought me back to the same journey which resulted in me choosing in Colemak.

Firstly I started analyzing my own WhatPulse data, it contains 2.1 million keypresses.
I also know that key count is not the only thing that makes a layout good (or bad) and after doing a lot of research I decided to base my layout to Ian's X6.4H layout, I preferred it over the BEAKL layout. You can check out Ian's and Den's long conversation here: http://shenafu.com/smf/index.php?topic=89, its a very interesting read.

The X6.4H and the BEAKL layout both have one great omission: the arrow cluster.
According to my data the arrow cluster has a combined usage of over 191.000 key presses.
Which is about 8% of the total amount and therefor it must be easily accessible (specially with copy/paste and such).

I didn't like a some other things about X6.4H and so for the past few weeks I've been tweaking and tweaking and I'm finally comfortable to publish my own layout to the world to get some feedback.
For this first version I only wanted to move around the keys and not assign any extra layer to them, that will be my second step. That step will including assigning F-Keys, moving numbers and punctuations to another layer and assigning keyboard shortcuts to be used while developing and such stuff.

Another step I want to take is to move the modifier keys to the home row with the EZ's "Spacecadet" functionality So please don't worry about Shift/Ctrl/AltGr placement.

BoeKoe
Introducing the BoeKoe layout:

For you non-Dutchies, we have the saying "Boe zegt de Koe" which is a kind of children's saying and translates to "A cow says moo". I found it fitting with the Boe and Koe trigrams being painful to type on this layout. A cow also usually has only two colors and this layout tries to cater to both Dutch and English.

These are its design guidelines:

• Comfortable for both English, Dutch and programming
• Arrow cluster must be on the right hand primary layer. They will also be included in load count for each finger
• Outer index columns must be as light as possible
• Right index finder load must be reduced significantly, because of significant usage with mouse/tablet and arrow cluster
• Right thumb load must not be high, because of significant usage with mouse/tablet
• Middle finger is touch and can handle load

This is the Flipped version which I will be using for testing on the Keyboard Layout Analyzer forks.

Flipping the layout shouldn't impact the layout performance in the KLA tests but it does significantly anyway.
The flipped layout also makes it a little easier to switch from Colemak, especially as I switched the T and S keys already.

I also used the following weight, which I derived from info here: https://geekhack.org/index.php?topic=82183.0


According to my WhatPulse data this results in the following load finger load:


And following column load:


This is the load comparison compared with a standard Colemak layout on the ErgoDox, with the UP arrow key placed where the comma is.


For extra information, this would be the column load using Ian's X6.4H layout according to my WhatPulse data.
With the UP arrow key placed where the comma is AND the H key placed the same as the BoeKoe layout.

As you can see this has a significantly bigger load on the index fingers.
The original X6.4H layout has the H key placed where with the Spacebar, which imho is a really bad choise as the Spacebar is by far the most used key (not counting the E key).

Wordlist
Attached you will find the word list I used for generating this layout.
It includes Ian's following word lists: Most used words exploded, Bigram and Trigram.
And includes a top 50000 Dutch words exploded: but only any word that has a count higher than 1000 you can find it here: https://raw.githubusercontent.com/hermitdave/FrequencyWords/master/content/2016/nl/nl_50k.txt

Scores
These are my scores on the three Keyboard Analyzer forks with the mixed word set.
Den's: http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer%202.html


Ian's: http://kla.keyboard-design.com/

Wierdly I cannot get above the X6 NonGr layout. I beat it in every sub test, but my layout only loses slightly in the "Same hand and thumb use" test.
I'm unsure why that has such a high penalty though.

Patrick's: http://patorjk.com/keyboard-layout-analyzer


BoeKoe performs pretty good on full English key sets too, Alice for example, especially considering the other constraints I put on the layout:



As a disclaimer: I have not typed a single letter on this layout, I will do so starting tomorrow
I wanted to get feedback before I was a week in.

Attached you will also find the two BoeKoe layouts and the two X6.4H layouts.

I look forward to any positive or negative feedback!

 X6 ErgoDox NonGr.txt (13.25 kB - downloaded 209 times.)
 X6 ErgoDox.txt (13.25 kB - downloaded 190 times.)
 BoeKoe F.txt (13.36 kB - downloaded 199 times.)
 BoeKoe.txt (13.35 kB - downloaded 214 times.)
 Wordlist en Woordenlijst.txt (281.15 kB - downloaded 555 times.)

[Sc0tTy]

Updated the post a couple of times with more information etc., should now be complete

[Sc0tTy]

Did some more tweaking, got it below 114 at 113.99



Switching the R and N key will reduce it to .85 but I don't think it's actually better because the ring finger would be moving around more.

Ian told me that having the S on the index isn't the best idea. I tried moving it around but can't seem to find a spot and keep/reduce the same score.

I'm still not happy with the LSM column, but it seems to be the least-worst for now.

 BoeKoe 133.99.txt (13.36 kB - downloaded 191 times.)

[iandoug]

I still get vastly different results from Den original scoring vs current scoring.

Even when testing on plain Dutch (ugly duckling borrowed from http://www.gutenberg.org/cache/epub/25580/pg25580.txt) with angle quotes replaced with normal doublequotes.

Must have a look at Den's current scoring and figure out what is producing such radical differences.

[Sc0tTy]

Just so you know typing the letters é or è is very simple on a dutch layout: hit the back-tick or apostrophe and then hit the key.
Also using those lettters like for instance meer and méér is 99% the same word but just prolonging or shortening the pronunciation.
Therefor replacing the é with e is not a problem, though the ' and `placement should have that usage in mind.
And to be honest, I hardly ever use it. I'd say maybe once a week for één but thats it.

I ran that text on v1 and v2 but see no mayor difference for BoeKoe, so i'm not sure what you are talking about.
v1:


v2:


Or did you mean on X6 NonGr?

[iandoug]

Can you please advise URLs where you ran the tests?

Thanks.

(BTW I speak Afrikaans so am kinda familiar with Dutch. But our keyboards don't have those tricks for diacritics (that I know of... but am on Linux not Windows.)
KLA has no provision of handling anything that's not specified on normal, shift, AltGr or shift-AltGr, so it just complains that the character is not found.

[Sc0tTy]

v1: http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer.html
v2: http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer%202.html

Yeah, I know about the character not found but I ignore those for diacritics.

[iandoug]

Okay.
The version of Seelpy 1.5 on v1 of the Den's analyzer is a bit off. I first fixed it.
Screenshots from Den1, Den2 and my local copy of Den1 attached. Used local copy because Den's Ergolinear/Matrix keymaps are a bit different to my versions.
Test input was your Dutch/English word lists. I'm not sure that it's such a good input test, but in the interests of apples vs apples I used it.

Notice the vastly different ranking for Seelpy 1.5 between Den1 and Den2.
Also note that X4 layout is on plain ANSI, not Ergo anything. Probably a bit awkward to type in real life, it belongs on an ergo layout.

Den1:


Den2:


Den1 local:


So I need to see how Den's scoring changed....

[Sc0tTy]

Hmm very strange indeed.

Does your KLA mirror differ from Den 'v1 local' ? Or is it the same as his 'v1 online'

Yeah the wordlist is not a 100%, it doesn't match my typing entirely. The 'W' and 'Y' for instance are way higher in that wordlist than my actual usage.
But using a book was worse as those use a lot of old or unused words.

Just noticed you got 113.96 instead of 113.99 on Den's v2, thats odd isn't it?


Yesterday I would have sworn I got a 113.81 but due to a browser crash I lost it and wasn't able to reproduce it.
Might he be actively editing it ?
I know he has an account on GH, not sure if he's seen this topic thought.
Edit: nvm, found it and I'm now down to .85

[iandoug]

Does your KLA mirror differ from Den 'v1 local' ? Or is it the same as his 'v1 online'

It should be the same (my local == my live KLA mirror ==moreOrLess== Den1 live)
I say more-or-less because I fiddled a bit with his original version ... think he presented the scores as a fraction of 1, and I scaled it up so that it would have more digits before the decimal, and some after.
Looks like he has copied that approach, there may be slight differences in how we did the rounding.

Yeah the wordlist is not a 100%, it doesn't match my typing entirely. The 'W' and 'Y' for instance are way higher in that wordlist than my actual usage.
But using a book was worse as those use a lot of old or unused words.

I had to do a Dutch setwork in High school (Karakter, it finally came to me now, could not remember this afternoon) and was trying to find it on Gutenberg without knowing the name or author... so just picked Hans Christen Andersen's Ugly Duckling as a stand-in for Alice in Wonderland... :-)

Just noticed you got 113.96 instead of 113.99 on Den's v2, thats odd isn't it?

Yes... I might have slightly different spaces or carriage returns in my input compared to what you have? Which may cause the fractional difference.

Yesterday I would have sworn I got a 113.81 but due to a browser crash I lost it and wasn't able to reproduce it.
Might he be actively editing it ?
I know he has an account on GH, not sure if he's seen this topic thought.[/s]
Edit: nvm, found it and I'm now down to .85

Cool... I'll wait for you to post next version  :-)

Cheers, Ian
BTW GH seems to have a lot less activity than before? Odd. Maybe US summer holidays.

[Sc0tTy]

Meh.. Seems 113.85 will be as low as it can go.





Switching L and D will make it go down to .82 but in my actual usage the D is used more then the L so that would be worse.

 113.85.txt (13.36 kB - downloaded 187 times.)

Tried putting O and L on the right index and moving the rest to the left hand but I couldn't get it under 115 then.

[iandoug]

Here's something for you to aim at then :-)

Your Dutch/Eng input, Den2.
My -+T+- HT02a layout is on my kla.keyboard-design.com site.

[iandoug]

re -+T+- HT02a, yes I know S is on the index finger.... layout is quite old and before I learned many things..

Think it still manages to work because 2 of the other letters are C and F and the others are lower-used.
S/C and S/F not such common combos as we think.

[Sc0tTy]

Wow re -+T+- HT02a looks cool, didn't know that one. Even after switching T and H it gets a higher score.
I'll have to look in to that one! Though that I and J on the left hand is very painful in Dutch.

In Dutch 'sch' is quite common (during the WW2 it was actually used to identify Germans), so you don't want SC on the same finger too.

Quick note, after moving the Enter to the right thumb on BoeKoe, like +T+, I actually got a score of 110.62
So that's where the big win seems to be.
Like my index, my thumb is overloaded as well and that's why I don't have a big % key on the right thumb on my layout.
I was thinking of putting the Enter key on the inner ErgoDox columns above the backspace, like I have on my TEK. It would mean using 4 physical keys for two actual keys and use both index and middle for both keys on both sides. It would also free a key up on my left Thumb.

I might just switch D and L, its only a .24% increase and it would make typing some bigrams easier, including 'else'. Or maybe that +T+ will inspire me

But for now bed time

[Sc0tTy]

Another update

Switched the O with A and L with D.

One thing I noticed is that the word list contains way to few enters, my real world data has more than 80k presses.
So what I did is to add two more Enters for each Enter and this is my result


And then tried to put the Enter on a couple of places.


108.60 is with Enter on Right Thumb next to the H, but it would make the load on the Right Thumb to high for me.


109.84 is with Enter on the Left Index where the '.' is, it would make the load on the Index just -0.14 then what it originally was on Colemak.

BoeKoe no more ?

110.51 is my new option. The Enter would be on the Left AND Right middle column with the Backspace. Sadly putting the Enter on the left adds 1% to the score. And KLA doesn't understand using both equally. I will also use the Enter and Backspace with both my Index and Middle fingers. So the load would be 1/4th.

Here are the same layouts on the old list.

As you can see the +T+ wins to my 110.51 but the Thumb load in that set is way to low.

 Wordlist en Woordenlijst Triple Enter.txt (298.21 kB - downloaded 3449 times.)

I will have to think of a new name now I guess...

[iandoug]

Can you please post the json for the one you want me to test with?
I have already added about 5 of your layouts to the test panel... :-)

Thanks.

Yes I shall have to ponder this situation... :-)

[iandoug]

BTW you appear to be using both thumbs on each thumb cluster.

[Sc0tTy]

Oh shoot, here you go:
 110.51.txt (13.44 kB - downloaded 194 times.)
 109.84.txt (13.36 kB - downloaded 201 times.)
 108.60.txt (13.44 kB - downloaded 188 times.)

Yeah those outer thumb-cluster were switched on the original ErgoDox KLA layout, never got around to switch them

Cool! Can't wait for your next test panel update

I suggest you use the new 'Tripple Enter' word list btw.

I also reran Alice and Top English Exploded, been a while.
Alice

+T+ wins by a big margin but imho Alice isn't a good real world word list

English Exploded

All BoeKoe layouts win on here to my great surprise

[iandoug]

Also the first and third layouts are illegal for KLA because you need to press two keys at the same time with your thumb to get the shifted characters on the thumb clusters.

[Sc0tTy]

Ah the +- and `~ ic. I'll have to rerun them tomorrow, didn't think of that. I just put them there to put them somewhere. Usage shouldn't be that high. You can put them on a random spot off the cluster, for your test.

[iandoug]

Den 2 scoring, your triple spaced input.

[Sc0tTy]

What did you change?

[iandoug]

Okay played a bit more, enough for now. Your turn to improve. :-)

Scoring on Den2, first with your triple-spaced test, and then with Ugly Duckling in Dutch. (British quotes).

Don't take this personally but as suggestions for improvement: I'm not convinced that your input text is a good test... eg

1. Too much repetition. Nobody is going to type "jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie"  in normal usage.

2. I came to realise that bigrams are actually trigrams, because there is a space between. So the idea of testing "two letters frequently typed together" becomes "two letters frequently typed together, preceded and followed by a space", which is not how words work. I've actually added no-space versions of bi/tri/quad/etc-grams to my tests, and results are different. The position of the space otherwise affects the purpose of these tests too much.

3. As previously mentioned, trying to optimise for more than one language at the same time is very difficult. Even UK/USA English is different ("ou" vs "o", and " vs ' etc). Or English and "programming", before we even get into which programming language.

4. Your input text does not have a wide range of punctuation. Not even a hyphen, question mark, exclamation or doublequote. So there's lots of things you are not evaluating.

5. That's why I tried some plain Dutch text.... :-)

Screenshots, layout and duckling attached.

Dutch is sufficiently different to Afrikaans for me to not "know" what are going to be bad combinations, so I have to rely on the tests.

[Sc0tTy]

I'm always up for improvement! I'm actually a software developer of Continuous Improvement Software : www.cierpa.com

For your Mod Ian you moved the dash, dot, back-tick and equals keys, and switched A and O back, is that correct or did I miss anything?

Okay played a bit more, enough for now. Your turn to improve. :-)

Scoring on Den2, first with your triple-spaced test, and then with Ugly Duckling in Dutch. (British quotes).

Don't take this personally but as suggestions for improvement: I'm not convinced that your input text is a good test... eg

1. Too much repetition. Nobody is going to type "jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie jullie"  in normal usage.

Well, its to increase the usage. Are you saying that randomizing the order of the words would change the score?
If so than I must definitely do that.

2. I came to realise that bigrams are actually trigrams, because there is a space between. So the idea of testing "two letters frequently typed together" becomes "two letters frequently typed together, preceded and followed by a space", which is not how words work. I've actually added no-space versions of bi/tri/quad/etc-grams to my tests, and results are different. The position of the space otherwise affects the purpose of these tests too much.

Hmm yeah, I'd say 90% of bigrams are part of a word and so the test must represent it. Didn't think it would change the score (except for high space usage).
I'll look into that

3. As previously mentioned, trying to optimise for more than one language at the same time is very difficult. Even UK/USA English is different ("ou" vs "o", and " vs ' etc). Or English and "programming", before we even get into which programming language.

4. Your input text does not have a wide range of punctuation. Not even a hyphen, question mark, exclamation or doublequote. So there's lots of things you are not evaluating.

5. That's why I tried some plain Dutch text.... :-)

Screenshots, layout and duckling attached.

Dutch is sufficiently different to Afrikaans for me to not "know" what are going to be bad combinations, so I have to rely on the tests.

Yeah, I have my own stats data for that, which makes some choices easier and some harder.
Here you can see the biggest use punctuations wish the lowest alphabetical keys

Other punctuation usage is even lower so I didn't even bother.
I will also be moving them to a secondary layer

[iandoug]

For your Mod Ian you moved the dash, dot, back-tick and equals keys, and switched A and O back, is that correct or did I miss anything?

I moved the things off the thumb clusters so that KLA would stop complaining about not being able to type some things.

On left hand side, A/O and put period under E. The A/O did not make a major difference I think.

On right hand side, rearranged home row and then some letters on bottom row. That's where the big win was.
But don't know if it will be comfortable to type, either in Dutch or English.

[Sc0tTy]

Okay, will have a look tonight.

[iandoug]

Today's round.... :-)

[tp4tissue]

what's this testing website called again?

[iandoug]

what's this testing website called again?

Patrick's original KLA: http://patorjk.com/keyboard-layout-analyzer/#/main
My fork of Den's fork, with first revised scoring: http://kla.keyboard-design.com/#/main
Den's fork with second revised scoring (where most of above scoring comes from): http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer%202.html#/main

My last comparative round-up of layouts (using Den's first revised scoring):http://www.keyboard-design.com/best-keyboard-layouts.html

Hope that helps :-)

[Sc0tTy]

Wow that 107.15 is insane, can't get under it!
Though I will probably not use it as that load on the Thumb is way to high for me.

What I don't understand is that reduce the pinky load by switching pinky and ring finger colums the score actually stays the same.
Shouldn't that be better as the pinky is not as strong as the index, plus upper pinky key is very hard to reach.


Did you see my questions about the word list earlier? I'm going to try to improve it as some letter usage is way to high or low.

Will try again tomorrow.

[iandoug]

What I don't understand is that reduce the pinky load by switching pinky and ring finger colums the score actually stays the same.
Shouldn't that be better as the pinky is not as strong as the index, plus upper pinky key is very hard to reach.

That sort of question you should ask Den... but it's not just which finger you use but how it relates to what came before and after.  I've seen English layouts with for example the "n" on the pinky... and "n" is 4th most common letter. I also came up with a layout where it was better to put "a" on middle finger and "e" on index, instead of the expected opposite.

Did you see my questions about the word list earlier? I'm going to try to improve it as some letter usage is way to high or low.

Yes. I have come to realize that word lists are not so good for testing keyboards layouts. The ones Patrick uses have two issues:
1. common words... yes they are common, but "and" and "the" are much more common than the bottom end of the list, but each word is only there once. Hence my "exploded common words" as an attempt to remedy this. Even still, it's only the same 200 words over and over again, with a bit of punctuation and assorted capitals.

2. word lists tend to be lower case only, and no punctuation.

3. the SAT list contains a lot of obscure words, as well as a lot of uses of certain letter combinations more frequently than normal usage.

4. Your list (eg that jullie repeated) is actually checking how the keyboard handles "e j". But is that common in Dutch? It's not common in English.

So after running lots of tests (http://www.keyboard-design.com/keyboard-tests.html) I've decided to re-split them into two categories:

1. "real world" tests, which will be (English) prose, and programs. Stuff you might actually type in real life.
2. "optimising" tests, which would have the bi/tri/quad/etc-grams (with and without spaces), number tests, word lists, etc.
3. I also have a bunch of non-English / "English-ish" tests which can be typed on standard American ANSI layout, not quite sure what to do with them. They provide interesting cross-reference as best English layouts also do well on them.

The optimising tests can be used for fine-tuning punctuation, numerals, etc. But as a test of the keyboard as a whole, not so good.

[iandoug]

Wow that 107.15 is insane, can't get under it!

:-)

[Sc0tTy]

Wow that 107.15 is insane, can't get under it!

:-)

Huh... I tried switching the G and W yesterday but didn't lower the score. Nice work though!

[iandoug]

Huh... I tried switching the G and W yesterday but didn't lower the score. Nice work though!

I played some more last night.

BTW these layouts are kinda tailored for your input text, and may not do so well on other inputs.

[Sc0tTy]

I've started working on an updated word list.

Here are the results after removing the bigram and trigram sections


Here are the results after pretying the dutch word list.

I randomized the words and added coma's, periods and capitals.

I'm now going to look where the gaps are in letter count .

[tp4tissue]

Guys, how does the same hand ranking work..



Does it currently mesh with the following conditions  (my opinion)



::: same hand , different fingers   is    good,   

because one handed is easy to time, and have very low probability of mismatch..

This also accounts for rolling letters like  p o i n  on qwerty for point. or  i o n for any TION words.



:::same hand, same fingers,  same row is good.

:::same hand, same finger,   different row is bad. (but not by much)

[iandoug]

Guys, how does the same hand ranking work..

Good question, I've been wondering the same thing. Den added it in V2 of his scoring, and I mostly work with V1 so have not tried to figure it out.
FWIW I normally look at the other metrics (which are not in those summary charts) like hand balance, consecutive finger use, etc. But it's too many screengrabs to post in these discussions.

Does it currently mesh with the following conditions  (my opinion)
::: same hand , different fingers   is    good,   
because one handed is easy to time, and have very low probability of mismatch..
This also accounts for rolling letters like  p o i n  on qwerty for point. or  i o n for any TION words.

I suppose it depends on your design philosophy. KLA does not worry about rolls (inward or outward) or row-jumps etc like some of the other analyzers.
Some people think both hands should do an equal amount of work. Which leads to vowels and consonants getting split, one hand getting the vowels. That makes long strings with one hand unusual.
See (much reduced) screen grab from "hand balance" metrics for a bunch of layouts attached.

:::same hand, same fingers,  same row is good.
:::same hand, same finger,   different row is bad. (but not by much)

I think KLA prefers you to NOT use the same finger twice in a row on different keys. It's a little awkward.
Screenshot of same-finger presses from same test (Alice, using Den2 scoring) attached.

Others with more experience than me may have different views.. :-)

[Sc0tTy]

So I have created a new word list.
I removed the original English word set as it was to small and used the same source that I used for my Dutch set.
Here is my source:  Top 50000.xlsx (2904.3 kB - downloaded 34872 times.)
Basically its any word used over a 1000 times in the original set and then spread according to usage.

The word list is random (seperate for dutch and english) and contains capitals, dots and comas.
I wasn't a 100% happy yet, so I did some manual changes in the .txt for KLA.
You can grab that here:  Wordlist en Woordenlijst 2.txt (309.6 kB - downloaded 4573 times.)

I suggest you use that one now.

Here is the new scoring based on that set:

[Sc0tTy]

Ayt 105.00:


 105.00.txt (13.44 kB - downloaded 190 times.)

[iandoug]

Well done :-)

Actually did consider putting H on S myself but thought it would be a bad idea because I thought (incorrectly it seems) that sh was a very common bigram in English.

[Sc0tTy]

Well done :-)

Actually did consider putting H on S myself but thought it would be a bad idea because I thought (incorrectly it seems) that sh was a very common bigram in English.

Yeah don't like it there either, gonna try to move it.

[Sc0tTy]

I will be changing the name of the layout back to its original: eNNe.
BoeKoe doesn't make any sence now and this is a layout for English and Nederlands.
So if you are making a new list, you can use that name

Because you made such high gains, I feel "Mod Ian" can be dropped because you had a lot of input so it will be the "original".
When its finalized I will make a "low thumb usage" mod for my personal use.

I've also updated the opening post.

Now for another round

[Sc0tTy]

104.59



 104.59.txt (13.45 kB - downloaded 181 times.)

[Sc0tTy]

104.32, I really like the load spread now!



Out of inspiration for the night...

[davkol]

Some people think both hands should do an equal amount of work. Which leads to vowels and consonants getting split, one hand getting the vowels. That makes long strings with one hand unusual.

For clarification, hand balance and hand-alternation frequency are two different things. The latter is much more important for typing rhythm.

Clustering vowels on one half of the keyboard is typical for maximizing hand alternation. There isn't consensus, if it's desirable—because of insufficient data at this point.

I think KLA prefers you to NOT use the same finger twice in a row on different keys. It's a little awkward.

Basically any practically used ranking discourages same-finger ratio and row jumps (separately though).

[Sc0tTy]

Putting them on the other side of the keyboard also makes sure that consecutive finger load is going to be as low as it can get.
While I don't have any data hand alteration in my mind makes the most sense because this way you reduce the awkward positions your hand needs to make because it has time to move to the next letter as the other hand is typing. Thought obviously not completely removes it.

And having the vowels on one side is just the natural way you'll get the most hand alternation, though the U is a weird one as its used way less..

[Snarfangel]

Here is the layout I've been using on my Kinesis for the past year or so:


I like having the letter e under my thumb. The keystrokes actually flow together pretty well, though it takes time to get used to it. I used MTGap 2.0 software and the patorjk site to come up with the basic outline and tweak it. For the data, I used Google Corpus stats from Peter Norvig http://norvig.com/mayzner.html and a gigabyte-plus corpus I made from Project Gutenberg books, years of emails and other writing, and tweet and other corpuses [corpii?].

If you'd like to try the layout in patorjk or similar online program, here it is:

[Sc0tTy]

Here is the layout I've been using on my Kinesis for the past year or so:
(Attachment Link)

I like having the letter e under my thumb. The keystrokes actually flow together pretty well, though it takes time to get used to it. I used MTGap 2.0 software and the patorjk site to come up with the basic outline and tweak it. For the data, I used Google Corpus stats from Peter Norvig http://norvig.com/mayzner.html and a gigabyte-plus corpus I made from Project Gutenberg books, years of emails and other writing, and tweet and other corpuses [corpii?].

If you'd like to try the layout in patorjk or similar online program, here it is:

Some nice ideas there! Like how you put the number layout on 5 keys, might use that one for my layout in step 2.
Yeah I've seen a ton of layouts put the E on the thumb but i really don't like it, my thumb would't be able to handle such a big load.
You have to remember that your four fingers are making a natural movement but your thumb is not, its moving sideways
Putting O and A on the same finger is also a big load.

For what language(s) is this layout? Seems bad for English or Dutch:


Not sure why but a 100% flip results in a lower score :/

Haven't found a difference in the layout

Here's the json:
 eNNe 104.32 F 103.28.txt (13.96 kB - downloaded 199 times.)

[iandoug]

For what language(s) is this layout? Seems bad for English or Dutch:

Depends how you measure :-)

Attached screenshot comparing it to other famous layouts on ErgoDox, using Den1 scoring, input text is Alice.

[iandoug]

104.32, I really like the load spread now!
Out of inspiration for the night...

Was quite a battle. Probably if you mirror it the score will improve again.

[davkol]

Putting them on the other side of the keyboard also makes sure that consecutive finger load is going to be as low as it can get.

It leads to perfectly maximized hand alternation and thus zero same-finger ratio, when typing in a language/orthography, where each and every syllable is formed from exactly one consonant and one vowel (in the same order). Which is never the case.

In practice, for example, Maltron THOR (with maximized vowel _separation_) has a lower same-finger ratio than Dvorak Simplified Keyboard. (calculated on English novels from Project Gutenberg, that ship with carpalx)

While I don't have any data hand alteration in my mind makes the most sense because this way you reduce the awkward positions your hand needs to make because it has time to move to the next letter as the other hand is typing. Thought obviously not completely removes it.

In theory, yes. However, plenty of awkward postures are eliminated by a reasonably natural physical layout (thus Maltron), and it's been suggested that there's an inherent neural delay, when alternating hands.

BTW there are other than biomechanical arguments to be made: for example, error rate. Again, Malt suggested that separation of vowels greatly reduced vowel substitution, an otherwise very common type of error.

[Sc0tTy]

104.32, I really like the load spread now!
Out of inspiration for the night...

Was quite a battle. Probably if you mirror it the score will improve again.

Nice work! Funny, I thought of moving the H there but thought it wouldn't help and so didn't try it. Guess I was wrong
Got a couple of minutes for another round

[iandoug]

I've come to the conclusion that H is a vowel. And so belongs with the vowels. At least as far as keyboard layouts go.

Though the frequency of H in your input text is less than in English, so I guess it's not so common in Dutch as in English. Probably the whole "th" thing not so prevalent.

[Sc0tTy]

Meh, no cigar, seems this might be the end of the road for A-Z optimization.

Yeah, I really like the H there. Typing 'HE' is a nice roll
H is quite common but as common as in English indeed

Wierdly the flip is now a higher score :/




Json:
 eNNe 102.72.txt (13.45 kB - downloaded 167 times.)
 eNNe 102.72 Semi Flip.txt (13.46 kB - downloaded 172 times.)
 eNNe 102.72 Full Flip.txt (13.46 kB - downloaded 175 times.)

[Sc0tTy]

Gonna give it a final go tonight to otherwise I'm going to load it onto my EZ and start learning the layout.

Edit: eNNe 102.46



 eNNe 102.46.txt (13.45 kB - downloaded 155 times.)

Tried moving the Y because typing 'you' wouldn't be nice but was unable to.

Ian, when will you be doing a new layout top list ?

[iandoug]

Gonna give it a final go tonight to otherwise I'm going to load it onto my EZ and start learning the layout.

Okay....

Results attached. Not wild about how much the right index has to fly around though.

Ian, when will you be doing a new layout top list ?

Soon... :-)
Been busy with personal and work stuff. It takes about 3 days to run all the tests... just on Den1 scoring.
I want to modify my fork of Patrick's original to handle the ErgoLinear/matrix layouts and run the same tests on that.

As for Den2 scoring, I still have some issues with it. Have started talking to him about it.
Ideally would also like to run the tests with that scoring, once I agree with it all :-)

Have started with a PHP version of the analyzer code, basically trying to do it without copying/converting the Javascript, which is a lot of work. But it does run a lot faster than the JS.
Will only be able to use it once it produces results 100% in line with existing Patrick/Den1/Den2.

Anyway, here's the results of tonight's drag and drops...

[Sc0tTy]

Hehe I would help you out but my PHP is very rusty and my math is really bad

I see you moved quite a lot on the right side, not sure if the typing experience is actually better.
Can you upload the json?
Without checking it looks like the load on the middel finger has been decreased and moved to the index.

[iandoug]

I see you moved quite a lot on the right side, not sure if the typing experience is actually better.
Can you upload the json?
Without checking it looks like the load on the middel finger has been decreased and moved to the index.

Yup. Right index flying around too much, despite lower score.

Layouts attached.

[iandoug]

BTW, those last layouts were based off my previous one, not your most recent.

Also, I used the list below, so layout is tailored for your input text (and Den2 scoring), scores are going to be different using different inputs and Den1 scoring. Or Patrick's scoring for that matter.
If I use your input text in the next batch of tests I'll first have to replace all the diacritic letters with the plain versions.

~/1web/keyboard-site/tools $ cat analysis.txt
 : 50125
e: 36645
a: 21031
t: 20916
n: 20083
i: 16420
r: 15738
o: 15259
s: 14306
d: 10487
l: 8756
g: 7667
u: 6725
c: 6671
m: 6655
h: 6024
p: 4991
k: 4968
w: 4907
b: 4405
v: 4258
f: 3659
j: 2190
.: 2020
z: 1221
y: 1190
,: 536
x: 382
T: 222
q: 185
[rest cut]

[Sc0tTy]

Yeah replacing them is fine Can you upload it when you've done it?

Hmm guess I'll have a look to merge both layouts tonight, though I probably wont keep the N on the index and move the S back

[iandoug]

Yeah replacing them is fine Can you upload it when you've done it?

Attached. Are you sure you want to test your layout with text like
"You we big caesa there son rangers is patient can else gaming legac protest with of even obtained vulcan he tat *****." ?

[iandoug]

Mmm .. I see this forum software does not like the word for "female dog" and replaced it with ***** ....

[Sc0tTy]

Hahah, not sure what you mean. Do you mean if I'm sure I don't want to test my layout with é á à etc?
If so, then yes, as that doesn't really happen in the real world

[iandoug]

Hahah, not sure what you mean. Do you mean if I'm sure I don't want to test my layout with é á à etc?
If so, then yes, as that doesn't really happen in the real world

No, your sample text "You we big caesa there son rangers is patient can else gaming legac protest with of even obtained vulcan he tat"
contains at least three words that I'm not even sure are English. I just pulled that random sentence from the end of your input because I saw it, but I saw other non-English words in other sentences.

Here's another:
"We cauttous she the in acquitted kenn are can slams was wiing contract tamm know calm complicattons badl resting christ feature."

etc.

[Sc0tTy]

Ah yeah, I replaced or removed some letters to get the word list up to my actual input stats.



 eNNe 102.02.txt (13.53 kB - downloaded 168 times.)

I don't understand why having such a big load in the index finger is good for the score :/
The load on the pinky is a lot higher then the middle finger now but switching it seems raises the score

[Sc0tTy]

Was gonna start a new round but v2 shows lower scores and errors in console while loading. I see that Den added v3 of the KLA.

Scores are quite different in v3 too:

[iandoug]

Was gonna start a new round but v2 shows lower scores and errors in console while loading. I see that Den added v3 of the KLA.

Yeah he backported some changes to v2,  maybe had a typo. Haven't checked yet.
V3 has some more serious changes, I think they may change some more soon.
He's punishing the pinkies more now. Also putting a heavy load on index fingers may make more sense than making the ring fingers work more. Also depends a lot on the sequence of letters, hand alternation, etc.
That score you got is pretty low :-)

[Sc0tTy]

Been trying a bit but not sure whats wrong.





The same finger usage for my WIP is 7.12 but when I look it up in the table its lower than the .02 layout, am I missing something?

IMHO the usage of middle finger home and upper row should have zero penalty except for same finger usage, that finger is strong. Index is indeed stronger then Ring, but having a high same finger ratio on the Index isn't that good.

[iandoug]

The same finger usage for my WIP is 7.12 but when I look it up in the table its lower than the .02 layout, am I missing something?

Maybe :-)
Is this Den2 or Den3?
For same-finger usage, are you checking Finger Penalty or Key Presses (selected from the dropdown UNDER the table on Miscellaneous tab).
Den has seriously punished the pinkies now (in v3, more so than in v2), in keeping with his layout design philosophy.

[Sc0tTy]

v3 and that's finger penalty. I would assume that's the one. Finger usage is also lower than .02

[iandoug]

v3 and that's finger penalty. I would assume that's the one. Finger usage is also lower than .02

FWIW I'm still using Den1 scoring until I fully understand and agree with the revised scoring models.

I will be doing some tests over the weekend.

[Sc0tTy]

v3 and that's finger penalty. I would assume that's the one. Finger usage is also lower than .02

FWIW I'm still using Den1 scoring until I fully understand and agree with the revised scoring models.

I will be doing some tests over the weekend.

Okay, haven't looked at that for a while. Are you using your mirror or Den's (there are some minor differences right)?

Not sure why Den changed stuff for his v2 but don't know if we should be using that one anymore...

[iandoug]

Okay, haven't looked at that for a while. Are you using your mirror or Den's (there are some minor differences right)?

Not sure why Den changed stuff for his v2 but don't know if we should be using that one anymore...

I use a local copy which should be the same as my mirror (which SHOULD be essentially the same as Den1... I didn't change any scoring algorithms, just the way the results are presented.)

Results for matching keyboards (eg MTGap, Colemak, etc) seem the same for a quick test with Alice:
http://kla.keyboard-design.com
http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer.html#/main

I have different graphics and less layouts so should run slightly quicker.

If I understood his readme correctly, he made v2 "more correct".

[Sc0tTy]

If I understood his readme correctly, he made v2 "more correct".

Only thing I read on his forum post is this: removed redundant penalty for shifts/modifiers (also removed from v2)
Is that the readme you are referring to?

[iandoug]

If I understood his readme correctly, he made v2 "more correct".

Only thing I read on his forum post is this: removed redundant penalty for shifts/modifiers (also removed from v2)
Is that the readme you are referring to?

Yes. He may have made other changes and forgot to mention them :-)

[Sc0tTy]

Yes. He may have made other changes and forgot to mention them :-)

Ah okay. I guess I'll do a rerun on all 3 versions. I don't pinky usage on the home row.

[Sc0tTy]

Reran on all versions:

v1


v2


v3


I guess Den made more changes to v3, its different to the same run as yesterday and now.
And 102.46 has a better score than 102.02, which imho is how is should.
Not sure what changed though as I didn't see a post.

[Sc0tTy]

Did a rerun of a bunch of layouts on the v3 and I found that 104.35 was the best.

Then started working on a new layout and got this one:


 eNNe 97.72.txt (13.44 kB - downloaded 182 times.)

After putting it in my Excel I must say i quite like the new v3, the result is very close to what I feel is right for my usage

[Sc0tTy]

There is still a bug in the same finger usage calculation though:






Actual values are the same but summary is not...

[iandoug]

There is still a bug in the same finger usage calculation though:
Actual values are the same but summary is not...

Mmm... I also noticed some anomalies with same-finger summaries tonight on Den3.
I have pointed him to this thread, not sure if he is following, but you should maybe point this out to him...

Your scores are getting low...

Cheers, Ian

[Sc0tTy]

Yeah thanks

Tried to lower it but no cigar. Tried putting DNSR on the right home row, but couldn't get the finger usage down.

[Sc0tTy]

Guess 97.72 might be it, couldn't lower it again.

97.48


Moved the H to the index and now both "IK" (most common word in dutch) and "HE" are good rolls
Haven't been able to make the right hand side any better though.

[iandoug]

Moved the H to the index and now both "IK" (most common word in dutch) and "HE" are good rolls
Haven't been able to make the right hand side any better though.

Few points improvement.

[Sc0tTy]

Got it quit a bit lower



 eNNe 96.88.txt (13.53 kB - downloaded 175 times.)

PS you should use the "inline full image option"

[iandoug]

Got it quit a bit lower
PS you should use the "inline full image option"

Congrats. That X/Z was bothering me but didn't think of swapping sides. I think with different input text, it would be better to swap Z and -, certainly in English (and programming) dash/hyphen gets used more than Z. Same with '" and Q.

I don't use inline images because I thought I first had to upload them somewhere.  But seems not. Will try next time.
BTW I also put C at bottom in attempt to be a little ctrl-z/x/c/v friendly. You now have both X and C that can move down. V will be a problem.. :-). Maybe put right Crtl on thumb next to Y, then Ctrl V/C/X are doable on one hand. (and similarly on left, then ctrl A and Z are easy.

[Sc0tTy]

Ahh was wondering why you swapped the B and C. Yeah I'm probably going to bind copy/paste keys so don't really need that I think.

Yeah dash and quote get used more then Z and Q. Maybe I need to update the word list so that we can actually put those keys on better spots while not losing score.

On a side-note.
Do you know of any articles/research into putting the home pinky keys on the bottom row (where the Z and / are on qwerty)?
The pinkys are a lot shorter and on columnar layouts, like the TEK and ErgoDox, that might be a much better place.
When I put my pinky's there it feels a lot more natural.

[iandoug]

Do you know of any articles/research into putting the home pinky keys on the bottom row (where the Z and / are on qwerty)?
The pinkys are a lot shorter and on columnar layouts, like the TEK and ErgoDox, that might be a much better place.
When I put my pinky's there it feels a lot more natural.

Haven't seen anything like that, but your point is valid (especially in my case, with my short pinkies (pics somewhere on Den's site).
It is similar to your idea of putting what are traditional "pinky" keys on ring and middle in eNNe... in truth I actually use my ring for those outside keys (on MS Ergo Natural keyboard), because of short pinkies.

Downside is that for pinky then to move to top two rows is increased distance, which will affect score.

So what PROBABLY should happen is that the form factor should take it into account, and lower the entire pinky columns by one unit or so.
Think I must see if I can build that into the ErgoLinear layout. The lack of stagger in ErgoLinear bothers me, but when I tried it long ago on early Programmer's Keyboard layouts, it just looked "messy", and I ended up with a regular grid.

I was actually browsing online for books on "keyboard design" and only found one short book from around 1993, no longer in print.

There's probably more "useful" info to be had by doing searches on the Patent websites... there's a LOT of patents on keyboards, most of which it seems never made it to production. So just wasted money by inventor, and block against people trying to improve. Also US patent system seems oblivious to prior art and grants patents on 'obvious' things. Which is why I decided against trying to patent either ErgoLinear or Seelpy (which is actually a "fresh" idea in keyboard design).

eg I took this one
https://www.google.es/patents/US3929216
and improved it (at least as far as KLA is concerned).
Here's scoring on Alice (Den 1), against top layouts in same form factor:



 But the text provides useful info, and there's a long list of other patents at the bottom which may provide further ideas....

[iandoug]

On a side-note.
Do you know of any articles/research into putting the home pinky keys on the bottom row (where the Z and / are on qwerty)?

Have you see MK-Type V2.Staggerfix layout?

http://kla.keyboard-design.com/#/config

Under ANSI section.

[Sc0tTy]

Ah yeah, but I guess thats not really research or an article:)
Was hoping to find some pro's/con's and such.

Is does remind me that I can change the default home key for the pink's though

[iandoug]

Ah yeah, but I guess thats not really research or an article:)
Was hoping to find some pro's/con's and such.

I trust you read both my posts above.

Anyway after many failures, we have a new working lightbulb.

[Sc0tTy]

Whups, you trusted wrong missed that one

Haven't seen anything like that, but your point is valid (especially in my case, with my short pinkies (pics somewhere on Den's site).
It is similar to your idea of putting what are traditional "pinky" keys on ring and middle in eNNe... in truth I actually use my ring for those outside keys (on MS Ergo Natural keyboard), because of short pinkies.

Downside is that for pinky then to move to top two rows is increased distance, which will affect score.

So what PROBABLY should happen is that the form factor should take it into account, and lower the entire pinky columns by one unit or so.
Think I must see if I can build that into the ErgoLinear layout. The lack of stagger in ErgoLinear bothers me, but when I tried it long ago on early Programmer's Keyboard layouts, it just looked "messy", and I ended up with a regular grid.

Yeah thats exactly what I was thinking!
The "normal top row" would be a bit harder to reach indeed but I'd just put something with low usage there.
I'd have to move PgUp/Dn on the ErgoDox though. On my TEK I'd probably just switch the middle and lower row for the pinky.

Same goes for all the keys on the right, ctrl etc.

I use my middle and index fingers for the TAB and key above as you can see in the layout.

I was actually browsing online for books on "keyboard design" and only found one short book from around 1993, no longer in print.

There's probably more "useful" info to be had by doing searches on the Patent websites... there's a LOT of patents on keyboards, most of which it seems never made it to production. So just wasted money by inventor, and block against people trying to improve. Also US patent system seems oblivious to prior art and grants patents on 'obvious' things. Which is why I decided against trying to patent either ErgoLinear or Seelpy (which is actually a "fresh" idea in keyboard design).

eg I took this one
https://www.google.es/patents/US3929216
and improved it (at least as far as KLA is concerned).
Here's scoring on Alice (Den 1), against top layouts in same form factor:

(Attachment Link)

But the text provides useful info, and there's a long list of other patents at the bottom which may provide further ideas....

I'll have a look tonight

Anyway after many failures, we have a new working lightbulb.

(Attachment Link)
(Attachment Link)

Nice
That Y is on a bad position I think, the most used word with that letter is probably 'YOU' which would be bad like that

[Sc0tTy]

Updated the finger usage on your 96.40 to match the actual finger and then made a pinky mod.





Edit: One of thing I already noticed is that my thumb now naturally lands on the inner thumb keys. Before it was a tiny stretch or I'd only tap it partially.

[iandoug]

Updated the finger usage on your 96.40 to match the actual finger and then made a pinky mod.

Bit puzzled as to why your scores and my scores are slightly different ... I noticed this morning when I was playing with your 96.88, I got 96.89 (or vice versa) and now you're getting 96.42 where I had 96.40.
Did you change your input file perhaps?
OS and browser? I'm using Firefox on Linux.

[Sc0tTy]

Updated the finger usage on your 96.40 to match the actual finger and then made a pinky mod.

Bit puzzled as to why your scores and my scores are slightly different ... I noticed this morning when I was playing with your 96.88, I got 96.89 (or vice versa) and now you're getting 96.42 where I had 96.40.
Did you change your input file perhaps?
OS and browser? I'm using Firefox on Linux.

Nope I actually gos 96.39 when I ran your JSON, I changed some key-finger settings though and that's why its .42.

Oddly when I switch F/Z my score gets worse but when I switch B/C for example there is no change ;/

[Sc0tTy]

Have been trying to optimize the layout but couldn't get it lower, but I'm very happy with it as is.
So added all of the miscellaneous keys like pgup/pgdn to the P layout.


 eNNe 96.40PM.txt (13.6 kB - downloaded 199 times.)

Guess I'll start moving the modifier and misc keys around now.

Think this might be is unless you have another light bulb moment

[iandoug]

Do you know of any articles/research into putting the home pinky keys on the bottom row (where the Z and / are on qwerty)?

Looking at staggering the ErgoLinear layout.

Question is, how much is enough/correct?

Looking at ErgoDox, it appears middle finger row is shifted around 12% relative to index and ring.

Do you find that enough/too much/too little? (Hello Goldilocks)

And your pinky going down one whole row... how is that?

Thanks, Ian

[Sc0tTy]

Sorry for the very late response but my laptop died and didn't have an opportunity at work to check GH... Finally got a new laptop today

Anyway...

The finger placement feels "right" but I haven't typed much on the keyboard since then so I cannot give you any more feedback on it, yet.
My main concern now, is reaching some other keys might have a bigger penalty, I'll get back to you on that

[Sc0tTy]

So did all my mods on the eNNe layout:
- P: Pinky mod
- M: Added Moddifiers
- A: Arrowcluster now on the spot I want it
- T: Thumbcluster Flip, where it will actually be in real life
- S: Shifts to Pinkies
- F: Flipped


And I have a 113.33 layout that I'm quite happy with (still need to flip it):

Any thoughts ?

Won't keep that Enter there, will probably move it to the big inner column or something.

[iandoug]

Any thoughts ?

Sorry, don't think I got an email for your last post.

Was reading your original post again, the reason there are no arrow keys on the X6.4 layout is because they are not necessary for KLA, and are left as an exercise for the reader... :-)

Thanks for your mod to X6.4, will add it to my collection to evaluate.

I should maybe take another look at non-AltGr layouts... we've been so focused on them because usually they out-perform everything else.

Cheers, Ian

[ADNW]

Hi, I somehow lost my login stuff, so made a new account. I have been around here much longer 

Anyway, patorjk has a nice analyzer, but you know that the scoring of layouts depends on:
- the layout
- the analyzer 

Different analyzers will give different results, depending on what that analyser thinks is important. I tried a few:
- the MTGAP analyzer
- the Carpalx analyzer
- The ADNW analyzer
- the Patorjk (web based) analyzer.

I don't agree with Carpalx assumptions, so high ranking layouts in Carpalx I don't find very good in real life. I myself like the ADNW analyzer best. Can also be tweaked to your liking, that is by setting penalities for all sorts of things.

BTW, your layout has resemblance to the AdNW layout. Can't seem to post images, look at adnw.de

[iandoug]

Hi, I somehow lost my login stuff, so made a new account. I have been around here much longer 
I don't agree with Carpalx assumptions, so high ranking layouts in Carpalx I don't find very good in real life. I myself like the ADNW analyzer best. Can also be tweaked to your liking, that is by setting penalities for all sorts of things.

What was your previous handle?

FWIW, Den (shenafu.com) has made some changes to Patrick's code: (two versions at least, with different emphasis).
first version includes adding vertical distance as well as horizontal:
http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer.html#/main

Current version has much stronger focus on avoiding pinky use, and avoiding horizontal index movement.
http://shenafu.com/code/keyboard/klatest/#/main

So different layouts do better at the different versions. Version 1 scores are largely in line with Patrick's original, except for the AltGr style layouts (Arensito, AdNW, etc).

From what I've read of CarpalX his methodology is largely in line with Patrick's although I don't think Patrick/KLA worries about rolls or row-jumps.

[Sc0tTy]

Wow.. Haven't checked back in a while. I see my scoring is totally off now with all the new rules in KLA.


This is my current layout:

http://configure.ergodox-ez.com/keyboard_layouts/kljbmn/edit

Have the eNNe home-row pegged down now (no speedtest yet) and I must say I really like the pinky mod!
Been working on positioning the keyboard shortcut layer and stuff.
The Q has been moved to another layer, thinking of doing the same for the Z and Y.
The punctuation's all have their own layer but they might have to move when i do the Z and Y move.
I have shortcuts for Ctrl +Z, X, C, V , Y on the base layer which is great. Saves a lot of pinky strain and bad positioning of fingers.

Was gonna decide what keys I would move next for typing but now with new scoring I think I need to redo some tests. Also I already know I need 4 layer toggles and that won't work how eNNe is now.
I'm also still not sure if I want to use the thumb keys for layer switching, I might move them all to the pinky.

[Sc0tTy]

First evening of hunting for a score, coming close X7.1 layout:


Shift keys are now under Pinky home key and X, Z, Y, Q have been moved to the "home cluster" on the Alt-Layer:


@Ian: Whats the main Idea behind the "move the pinky home key to the middle finger upper row" concept ?
I really like the "home cluster" idea, but what is the research that this is faster/better?
Or is the only upside to lower usage of the pinky? Because now the middle finger gets a bigger load.

I'm trying to put all modifiers and layer-switch keys on the pinky, as its already used to it and its better at it then the thumb (down movement vs sideways movement).
Also with the pinky home letters gone it should be able to take that load no problem

[iandoug]

First evening of hunting for a score, coming close X7.1 layout:

@Ian: Whats the main Idea behind the "move the pinky home key to the middle finger upper row" concept ?
I really like the "home cluster" idea, but what is the research that this is faster/better?
Or is the only upside to lower usage of the pinky? Because now the middle finger gets a bigger load.

1. It's illegal to beat X7.1H ... :-) ... but you're getting very close so I'll have to see where I can improve. But see note below.

2. You lost me here : "move the pinky home key to the middle finger upper row" .... where did you see that? First I've heard of the idea. So I thought maybe there was an error in one of my layouts, but 7.1 looks okay on Klatest.

3. The 'home cluster' aka home block is Den's idea... he is very focused on reducing pinky effort. I think because his model is based on 'weighted effort' where using the pinky gets punished because it is weak. Which means Dvorak and Colemak get smacked because they have common vowels on pinkies.

Been reading some research papers, saw one that says index is actually stronger and more flexible than middle, so in theory it should have lowest effort score, so that it can do more work. Which makes sense because most layouts give the index two columns to worry about. However Klatest also punishes horizontal motion, so inner index column is also punished. All scoring algos that I know of assume middle finger is strongest.

All of which makes finding a winning layout a bit of a challenge :-). Hence my "note" comment above: X7.1H is designed to win on Klatest, but it does not do so well on Den1 scoring, and presumably also not on Patrick's original scoring. Best layout on Den1 is still X6.3H (for English and probably Code).

I've just run a fresh batch of tests using over 300 layouts, and Den1 scoring. Will upload results in next few days, need to fix the database browser program to handle some things differently.
Will then repeat the process using Klatest scoring, and Patrick's original.

I don't think I'll be able to add your latest eNNe versions to the tests... KLA does not handle layers apart from Shift, AltGr, and Shift-AltGr. Which may mean that your scores above are not accurate... did you check the Javascript console output for error messages?

Must figure out how to subscribe to these threads...

Cheers, Ian

[Sc0tTy]

1. It's illegal to beat X7.1H ... :-) ... but you're getting very close so I'll have to see where I can improve. But see note below.

Guess, it's a good thing I don't like laws

2. You lost me here : "move the pinky home key to the middle finger upper row" .... where did you see that? First I've heard of the idea. So I thought maybe there was an error in one of my layouts, but 7.1 looks okay on Klatest.

3. The 'home cluster' aka home block is Den's idea... he is very focused on reducing pinky effort. I think because his model is based on 'weighted effort' where using the pinky gets punished because it is weak. Which means Dvorak and Colemak get smacked because they have common vowels on pinkies.

Well when I started doing eNNe I wanted to do home-block but would get a lower score so I like that its not a big of an issue now. But what I mean is that now your middle finger will do a lot more work (distance) with two of the main letters. Wheres when its behind the pinky you spread the load a bit. Just wondering if there was more info on it

Been reading some research papers, saw one that says index is actually stronger and more flexible than middle, so in theory it should have lowest effort score, so that it can do more work. Which makes sense because most layouts give the index two columns to worry about. However Klatest also punishes horizontal motion, so inner index column is also punished. All scoring algos that I know of assume middle finger is strongest.

Guess my thoughts agree with that research then
With the pinky mod I'm using the upper row is actually a bit harder for the index and ring finger to reach, the bottom row is easyer now. Without the pinky mod the upper row is easier, just so you know.

I've just run a fresh batch of tests using over 300 layouts, and Den1 scoring. Will upload results in next few days, need to fix the database browser program to handle some things differently.
Will then repeat the process using Klatest scoring, and Patrick's original.

I don't think I'll be able to add your latest eNNe versions to the tests... KLA does not handle layers apart from Shift, AltGr, and Shift-AltGr. Which may mean that your scores above are not accurate... did you check the Javascript console output for error messages?

No need to do it, this eNNe is still WIP

Must figure out how to subscribe to these threads...

Cheers, Ian

Have you tried the "Notify" button at the bottom? You should be able to get email notifications.

[iandoug]

Have you tried the "Notify" button at the bottom? You should be able to get email notifications.

From what I can see (both here and in my Notification settings) I'm supposed to get notifications for this topic, but don't seem to... maybe the backend does not like my postgrey filter on my mail server.

[iandoug]

Have you tried the "Notify" button at the bottom? You should be able to get email notifications.

From what I can see (both here and in my Notification settings) I'm supposed to get notifications for this topic, but don't seem to... maybe the backend does not like my postgrey filter on my mail server.

Notification finally arrived. Guess the greylisting slowed it down.

[Sc0tTy]

Ah okay great


Getting closer


For some wierd reason the right pinky has a to high count, not sure where that's coming from.
And even though Enter is way higher then AltGr, if I switch them my score drops and I have no idea why.

Switching spacebar with AltGr also lowers the score even though it (i know from experience) would be better to have the layer toggle key on the other hand.

[iandoug]

For some wierd reason the right pinky has a to high count, not sure where that's coming from.
And even though Enter is way higher then AltGr, if I switch them my score drops and I have no idea why.

Switching spacebar with AltGr also lowers the score even though it (i know from experience) would be better to have the layer toggle key on the other hand.

Um, think your left pinky has higher usage than right pinky.

Interesting idea putting the letters on altgr/shift-altgr like that... not something I have thought of before ...

[iandoug]

You could try this for Dutch:

[Sc0tTy]

Nope only makes it worse.

This is what I mean with the pinky:



Count says 11200 but only has the J key there now and a way lower keycount

[iandoug]

I moved the right pinky home onto the j key. I know this breaks your setup, but it seemed better than moving the j down to where home is, and leaving a large gap to the top keys.

[Sc0tTy]

Well I want the shift modifier under the pinky home as this key is used a lot more then the J key.

But that doen't answer my question, where is that high pinky count coming from?
Is finger keycount displaying the keycount with a penalty ?
Don't know what the pinky penalty is but 11120/2240=5

[iandoug]

Well I want the shift modifier under the pinky home as this key is used a lot more then the J key.

But that doen't answer my question, where is that high pinky count coming from?
Is finger keycount displaying the keycount with a penalty ?
Don't know what the pinky penalty is but 11120/2240=5

:-) That's why I put the shift on the right thumb. And j home on the right pinky. That layout got better scores than X7.1H on Ugly Duckling in Dutch. Which is why I suggested it ;-)


I didn't see such high usage on right pinky as what you are seeing ... might be Input Text dependant? Where you have a lot of characters on left hand that need to be uppercased?

[Sc0tTy]

Well I want the shift modifier under the pinky home as this key is used a lot more then the J key.

But that doen't answer my question, where is that high pinky count coming from?
Is finger keycount displaying the keycount with a penalty ?
Don't know what the pinky penalty is but 11120/2240=5

:-) That's why I put the shift on the right thumb. And j home on the right pinky. That layout got better scores than X7.1H on Ugly Duckling in Dutch. Which is why I suggested it ;-)


I didn't see such high usage on right pinky as what you are seeing ... might be Input Text dependant? Where you have a lot of characters on left hand that need to be uppercased?

As you can see in the images, that score is only with the J on the upper pink row, the shift is still on the thumb. Hence my questioning.

[iandoug]

As you can see in the images, that score is only with the J on the upper pink row, the shift is still on the thumb. Hence my questioning.

You need to move pinky home to the j .

[Sc0tTy]

As you can see in the images, that score is only with the J on the upper pink row, the shift is still on the thumb. Hence my questioning.

You need to move pinky home to the j .

So the high keycount is 2210*5 to calculate for the penalty ?
I just want to confirm that

[Sc0tTy]

Well I did it


Big difference compared to original to the "original final eNNe layout"



Getting pretty excited about it too, this is the reduction in pure finger stress according to my typing statistics:

Stats for when the layout is flipped btw and includes arrow cluster (not in KLA stats)
Good to see how putting those QYZX letters on the homeblock helps a lot on the layout

This is on the dutch/english word list I've used previously.

[iandoug]

Good to see how putting those QYZX letters on the homeblock helps a lot on the layout

This is on the dutch/english word list I've used previously.

Played around a bit... attached :-)

[Sc0tTy]

Good to see how putting those QYZX letters on the homeblock helps a lot on the layout

This is on the dutch/english word list I've used previously.

Played around a bit... attached :-)

Not bad

Must say I don't think its a real world improvement. The J is quite common in Dutch so I'd prefer to have it on the primary layer.
J is 0.67% compared to F at 0.63% and quite a few hits will be CTRL+F and CTRL+SHIFT+F which ill be putting on a separate key.

Also the shift next to the most used key is also not that great imho. Its also a modifier and sideways action with thumb is not great imho.
And I use that key to switch to a layout too.

But you have me an idea and shaved another point off


I also switched the P and W which lowers the score with .02 but in my personal stats W is 1.13% vs P at 0.78% usage. So that should be a lot better in real world (though I'm not sure how much of that is me hitting CTRL+W )
EDIT: Guess that got revert somehow lol, see now its back. Anyways score with P/W switch is 108.29

I also switched the H and U because i think it will make it a little easier to type stuff like "choose".

[Sc0tTy]

So after a while of life getting in the way, I started working on my punctuation and numpad layers when I suddenly got another idea.
That change ended up taking off more then a whole point: 107.12. Then I noticed that having the dash instead of the coma in the normal grid was making it artificially lower (also on the previous 108.29 layout) so I fixed that and this is the result:




I moved the X to the 2nd layer, moved Y to the index finger and then moved the J to the other side, should be okay enough to type "jij" and "je" (though it's not my preference). I was now able to move the CTRL above the SHIFT.

I was then checking the key data in KLA with my own WhatPulse data and decided to fix the inconsistency of the T and Z key's. Their frequency was to low, Z is actually used more frequently than F so I updated my data set and added more of both keys.

I then moved the Z to the middle and F to the index finger and got a little lower score



When looking at the heatmap in KLA v2 you'll see that the biggest load is now on the index finger (corresponding with my data its a 0.36% difference)


So obviously I tried switching the T and N but that gave me quite a bigger score, almost 6pt increase.

The only reason I can think of is that having R, T and L keys would be a lot of keys on just one finger (only the same finger is significantly higher).
I have not yet gotten around that issue to see if I can get done but decided to post an update as it has been a while
Turns out the index sees a big encrease in same finger usage. I guess even though the T is used a lot, its not used as much as N with other vowels:


Attached you'll find the wordlist with extra TZ's and both KLA exports.
I'll be using that wordlist in the future but somewhere in the future I want to change it so that it uses actually words based on the stats and the top word list. Because I do feel that removing and adding random characters does skew KLA's results. Although this layout does look very good in my WhatPulse data.

@Ian: I'm now getting a 4% gap on X7.1H Ergolinear and a 17% gap on BEAKL 9 (pretty significant I would say) on my wordlist, I wanted to know what is your main wordlist for working on a layout and if you know what layout Den uses. So that I can test it against those wordlists and see how it does.

On a side note:
I've been using the pinky mod for quite a while now and I must say that it makes the ErgoDox thumb cluster a lot better to use but also makes everything a little easyer on the pinky. I'll write a longer post in a separate thread for that one of these days.

[Snarfangel]

Just a quick note, since it's time for me to head off to sleep. I went back to some of the old layouts I generated with the MTGap software, and found one that seemed to work unusually well with the current round of keyboard layout analyzers -- lower in rolls than the one I was using, but quite a bit better in hand alternation and lower in same finger use for consecutive letters, and significantly better scores. After trying a variety of inputs and tweaking things in the keyboard layout analyzers at Shenafu's site (https://shenafu.com/code/keyboard/klatest/#/main) as well as http://kla.keyboard-design.com/ I came up with this somewhat unusual layout for my Kinesis Advantage2 that I might as well call KFOU DHAI Y (the letters on the left hand):



Comparing to other layouts with the wordlist and layout provided by Sc0tTy, here is the heatmap I get for the collected works of Mark Twain:


And the results of both sets of tests -- Shenafu's ver. 3 on the left, KLA on the right:


And here is the layout you can plop into any of the above keyboard layout analyzers:
 KFOU_DHAI_Y_Kinesis.txt (14.71 kB - downloaded 165 times.)

I have another one with a bunch of foreign characters mapped to AltGR, but I still need to see if I can improve on the high ASCII side. I might post it in the next few days for completeness, though.

[Sc0tTy]

That's an interesting layout!
That location of the right shift modifier is quite interesting, have you checked the scores when putting the left shift modifier in the same place?
Because I do think its on the wrong place currently for a key that's hit over 4%.

What's the reason for the Enter on the right pinky's second layer?
The CTRL is on a bad place though, what finger do you use for it, the middle?
Do you really use the middle and index finger to hit those delete and backspace keys?

I also assume that you have another layer modifier on the other side otherwise it would make hitting some of the keys on the same grid very hard.

I guess you win the most for having the E on the thumb cluster. This layout wouldn't work for me because my right thumb cluster needs a low key count.
Though that shift placement I might use that somehow, I need to update my dataset to represent the actual shift count, its way to low currently and it might effect eNNe.

You might need to check the Enter usage, because it doesn't show on the heatmap and that's very strange. You can see on my eNNe heatmap it should show up on at least my wordlist.

[Snarfangel]

That's an interesting layout!
That location of the right shift modifier is quite interesting, have you checked the scores when putting the left shift modifier in the same place?

Not really, I like to hit shift with one hand and the key I want shifted with the other.

Unless you mean put the left shift on the standard QWERTY location for "B"  ) -- then yes. And everywhere around there. The left thumb is where the left shift likes to be. I even looked into changing <left shift><space> to Enter, freeing up the right thumb for a shift key, but haven't found a way that the analyzers like.

Because I do think its on the wrong place currently for a key that's hit over 4%.

It does look wonky. I think it's mostly the asymmetry. But the stats got better when I moved from the normal shift position. Using my Kinesis Advantage2 layout, I first put it on the right pinky single quote-double quote key, then to the up-arrow under the right index finger, then the down arrow under the right forefinger (mirroring backspace on the left side), and then finally where N is on the standard Kinesis. Here is a picture, which also helps with the next question:

What's the reason for the Enter on the right pinky's second layer?
The CTRL is on a bad place though, what finger do you use for it, the middle?
Do you really use the middle and index finger to hit those delete and backspace keys?

The Enter key is part of the embedded keypad -- you just hit the button on the upper right with the blue letters. I just wanted to account for it, since I do use it when typing in long strings of numbers. I'm going to set up a set of non-standard characters there too -- with the AltGr key -- but the keyboard layout analyzer doesn't handle four characters plus numbers.

The Control keys work pretty well, actually. The are under the pinky, and I just have to curl my hand a bit. It's a little easier for me to hit than the normal shift keys, and quite a bit easier than the top of the thumb cluster, which is where Kinesis puts them.

I also assume that you have another layer modifier on the other side otherwise it would make hitting some of the keys on the same grid very hard.

I guess you win the most for having the E on the thumb cluster. This layout wouldn't work for me because my right thumb cluster needs a low key count.
Though that shift placement I might use that somehow, I need to update my dataset to represent the actual shift count, its way to low currently and it might effect eNNe.

You might need to check the Enter usage, because it doesn't show on the heatmap and that's very strange. You can see on my eNNe heatmap it should show up on at least my wordlist.

I'll check the Enter usage. That does look odd.

[Sc0tTy]

That's an interesting layout!
That location of the right shift modifier is quite interesting, have you checked the scores when putting the left shift modifier in the same place?

Not really, I like to hit shift with one hand and the key I want shifted with the other.

Unless you mean put the left shift on the standard QWERTY location for "B"  ) -- then yes. And everywhere around there. The left thumb is where the left shift likes to be. I even looked into changing <left shift><space> to Enter, freeing up the right thumb for a shift key, but haven't found a way that the analyzers like.

I guess the thumbcluster might be positioned differently and/or your hands are bigger. For me the thumb where your 'e' is placed is the only natural place for the thumb and thats only after doing the pinky mod.

What's the reason for the Enter on the right pinky's second layer?
The CTRL is on a bad place though, what finger do you use for it, the middle?
Do you really use the middle and index finger to hit those delete and backspace keys?

The Enter key is part of the embedded keypad -- you just hit the button on the upper right with the blue letters. I just wanted to account for it, since I do use it when typing in long strings of numbers. I'm going to set up a set of non-standard characters there too -- with the AltGr key -- but the keyboard layout analyzer doesn't handle four characters plus numbers.

Ahh the enter under the pinky makes sense for the numpad layer indeed.

The Control keys work pretty well, actually. The are under the pinky, and I just have to curl my hand a bit. It's a little easier for me to hit than the normal shift keys, and quite a bit easier than the top of the thumb cluster, which is where Kinesis puts them.

Looks like its a little to the right from where the CTRL usually is on a normal keyboard. And I already have a bad pinky from that position, your CTRL would probably be worse for me.

[Snarfangel]

I think the previous heat map was from my "Big bunch o' text" corpus I got by combining all of the inputs on the page (from "English: ANSI characters" to "Tech: Currency Modern..."). Not sure why it didn't show the return key. Here are our two layouts for comparison:


When I re-did Mark Twain's Project Gutenberg corpus with the High ASCII/Foreign character version (had to remove the number pad, which dropped the score a couple of points, even though it's still there), I got the following. The character count on the bottom of each picture is while I had the mouse over the "Enter" key:


No clue why some blank keys have color.

[Sc0tTy]

Yeah eNNe has been optimized for lower pinky usage so it wont get a top score on V2.

I guess there is something iffy going on with Ian's version of KLA. You can find V2 of Den's KLA here: http://shenafu.com/code/keyboard/Keyboard%20Layout%20Analyzer%202.html

I see you also use non-standard alphabet characters in your layout as well.
Dutch has some of those characters but they're not used much so I don't need to take that into account.

AltGr and Enter are not going to end up on those places but KLA gives me a massive penalty if I move them.
It doesn't understand that if I place it on the left and right side it will need to put half on each side

[Sc0tTy]

I updated my wordlist (attached) with the right shift modifier count. And it confirms my choice for putting it directly under the pinky:




What I noticed is that CTRL and SHIFT need to be placed quite close to each other because their combined usage is also quite high.
This is another reason for me not the put the shift under the right thumb hitting CTRL+SHIFT+LEFT (my most used combo, according to data) a pain.
But the main issue is that my right thumb can't handle the load and I also hold my pen there which is my input device. Plus the holding the thumb down like that is not its most natural motion: side-ways vs inwards.

I'm sad KLA can't handle CTRL, UP, DN, LEFT and RIGHT because it would show that eNNe should be even better in real world scenarios. Because now it looks like a lot of precious real estate is unused compared to other layouts.

Looks like I'm pretty much done with the main layer unless me, Ian or anybody else gets another light bulb moment. I'll go back to figuring out where to put the layer modifiers and edit my other layers

I do believe that modifiers should be on both sides of the board for their optimum usage and under fingers that move downwards and not sideways

[Snarfangel]

I noticed that a lot of keyboard layouts influenced by MTGAP software put the E and the O on the same finger, since they are both common letters that rarely occur together (at least semi-rarely). So I thought it might be interesting to put both in the left thumb cluster, and moving my shift to mirror the one on the right side.

It actually seems to have fairly significantly improved the score for the final layout. I still have a ton of tweaking to do, and it might feel weird when I try it out, but it's a lot of fun seeing an unexpected result like that. Of course, if you don't like E under the thumb, you probably don't want E and O under it, but I'm willing to give it a shot.

[vvp]

Isn't there a program which would find the best layout based on the given word list, the physical switch positions, and the weights of the optimized attributed (distance, same finger, same hand, ...)?

It does not look like too hard(*). It would save a tremendous amount of time for people who care about layout optimization ... not me, obviously, since I use only a slightly modified qwerty

(*) Well at least in the case the almost best solution is good enough. Finding global maximum may be more complicated.

[davkol]

There's about a dozen of such programs, but the weighs are entirely subjective.

[Sc0tTy]

I noticed that a lot of keyboard layouts influenced by MTGAP software put the E and the O on the same finger, since they are both common letters that rarely occur together (at least semi-rarely). So I thought it might be interesting to put both in the left thumb cluster, and moving my shift to mirror the one on the right side.

It actually seems to have fairly significantly improved the score for the final layout. I still have a ton of tweaking to do, and it might feel weird when I try it out, but it's a lot of fun seeing an unexpected result like that. Of course, if you don't like E under the thumb, you probably don't want E and O under it, but I'm willing to give it a shot.

Yeah it might sound like a good idea in theory but having the O so far away sounds like a bad idea.

OE combination is pretty common in Dutch actually and is one of the few things I'm not sure about for this layout.

[Sc0tTy]

I'm trying to do some updates to the layout but it seems http://shenafu.com has been down for a while.

Does anybody know what happend to it? Or if there is a mirror of his KLA ?

Found them: http://ieants.cc/code/keyboard/kla3 and http://ieants.cc/code/keyboard/klatest

I'm thinking of adding the auto shifting feature of the EZ to the layout, this should free up a lot of shift usage

[batfink]

If you're looking to test this on a variety of analyzers, there is another fork here, with well documented changes (I think with heavier emphasis on avoiding horizontal hand motions).

Would be interesting to see how the results compare.

If only there were an analyzer that could compare all the different analyzers!

[Snarfangel]

I actually used the keyboard layout analyzer at https://stevep99.github.io/keyboard-layout-analyzer/#/config and came up with the following Kinesis setup (I have an Advantage2):




It has really good stats for a Kinesis layout, and with this analyzer it's significantly better than my current layout. Since it's a couple of years since I learned my current layout, I'm about due for a change.  )

About the only thing I don't like the look of is the left and right shift keys aren't symmetric, but I really wanted backspace under my thumb (I use it a lot).

I'll attach the layout data file so others can import it and play with it if they like. I am probably weird, but I really like seeing the layouts other people come up with, and the design decisions that went into them.

 FinalBestKeyboard07252020.txt (15.68 kB - downloaded 110 times.)

[Sc0tTy]

So.... After discovering the Auto Shift feature of the ErgoDox EZ and testing it out a whole new world opened up to me.
I'm enormously convinced that, except for non-staggered layouts, it's the single best thing that has been invented in Ergo-Keyboardland.

Having XZFYQ on a secondary layer is not very efficient when having Auto Shift enabled so I'm changing the layout and moving them to the primary layer.
I then gained the insight in having most of the punctuations on the primary layer as wel as I hit those more than most of my shortcuts.
I also moved more keys to the left side of the board.
So I've come up with this layout (and it's looking quite final).

Sadly having no shifts in the KLA results in invalid reports but IMHO so I moved it to the thumb cluster for the most optimum result. Imho this still downgrades the result. I've kept shift/ctrl on the original spots in the actual layout thought because they are used a lot in during text editing operations and other tasks.

Most layouts i've seen never focus on those kinds of operations while being a mayor part of the usage: like the arrow key cluster usage.



https://configure.ergodox-ez.com/ergodox-ez/layouts/QaGnB/latest/0



I also updated the word list a bit to better reflect the 'J' usage in dutch and 'jij' trigrams:
Here is the new heatmap based on my latest WhatPulse data (over 5 years of data by now)


Hope to hear your thoughts!

Download for layout:  eNNe AS 78.38 (15.97 kB - downloaded 98 times.)
Updated wordlist:  Wordlist en Woordenlijst 2X TZJ.txt (339.29 kB - downloaded 137 times.)

[iandoug]

Re eNNE AS 78.38:

Error with eNNe AS 78.38.json : eNNe AS 78.38, duplicate character on keyboard: ; (59)
Error with eNNe AS 78.38.json : eNNe AS 78.38, duplicate character on keyboard: ! (33)
Error with eNNe AS 78.38.json : eNNe AS 78.38, duplicate character on keyboard: x (120)
Error with eNNe AS 78.38.json : eNNe AS 78.38, characters not on keyboard: X^&
Error with eNNe AS 78.38.json : eNNe AS 78.38, no Enter key assigned

Cheers, Ian

[Sc0tTy]

O indeed! Fixed now



Ian, what's the main Corpus you use for your layouts?
And what are your thoughts on the changes I've made?

[iandoug]

O indeed! Fixed now

Ian, what's the main Corpus you use for your layouts?
And what are your thoughts on the changes I've made?

1. more fixing needed :-)
Error with eNNe AS 83.86.en.json : eNNe AS 83.86, duplicate character on keyboard: ; (59)
Error with eNNe AS 83.86.en.json : eNNe AS 83.86, characters not on keyboard: ^

re the corpus, I did discuss it on Den's site a while back.. . basically it dawned on me that most if not all of the texts we were using were not good... they may be "real world" but their letter frequency does not match English (or Dutch, for that matter). That applies especially to the assorted word lists.

So I went searchabout .. if I found stuff online (news or blogposts) that looked like they might have correct frequency, I cleaned them up and ran them through a checker. Most failed.

Did the same with assorted old books on Gutenberg.

My checker tries to find the texts that most match English letter frequency. I was doing "similarity checks" to the list of characters in descending order, but existing similarity algos didn't work they way I needed them to, so I rolled my own. Ended up with three, slightly different.

Two check all characters, the other just checks the 11 most common letters (as in: your home row and space bar.... or somesuch).

At the moment I only have 8 texts that have the top 10 or 11 characters in the correct order (score 66 or 65 in Top11 column on spreadsheet) . I use this metric because letter distribution in a text is some sort of Zipf distribution curve, and the most used letters are the most important. The FreqMatch column is most important, lower == better. The Similarity column has higher== better.

I uploaded them to Den's site the other day, might as well put it here too. Note that most of them are from a blog by John Ward, it just so happens that he occasionally churns out pieces that have the right frequency distribution. They are probably copyrighted but I guess we can plead some sort of fair use case. Two are old books, and one an op-ed piece from Russia Today. Note I'm not concerned with what they say, just the letter frequency ..... :-)

Texts and analysis attached. The .csv is tab-delimited.

I don't know what to do about "programming" inputs... original idea was a mishmash of languages from RosettaCode. Should probably limit it to "popular" languages, but that's a moving target and the well known lists are heavily criticised. There's still tons of COBOL and Fortran being written/maintained but those programmers don't need to ask Google or StackOverflow for help....

So at the moment I'm just going with these English tests, and sometime KLE home page for "web dev stuff" ... but that overuses certain things like the letter k and doublequotes etc.

As for your layouts, will take a look. At the moment trying to do a round-up of recent new/updated layouts to do another round of testing. So if anyone reading this has a new layout, feel free to submit.
You can check most of what I already have here:
https://www.keyboard-design.com/internet-letter-layout-db.html

I have a bunch of others to add as well (and probably some others from Den)

May  6  2018  YPHINAFU.txt
May 11  2018  beakl9-ansi-shifted-mod.txt
Dec 10  2017 'eNNe KLA3 P 109.54.json.txt'
Jul 19  2018  phynnboi.txt
Feb 11  2019  quesnel.mtgap.beakl.ansi.txt
May  6  2018  snarfangel-UP_OIANY_4.txt
May  6  2018  spindle.txt
Aug 13 07:50  FinalBestKeyboard07252020.json
Aug 13 07:48  KFOU_DHAI_Y_Kinesis.json
Aug 13 07:31  Qwicker-KFLY.kla.json
Aug 13 07:31  Qwicker-Mod-H.kla.json
Aug 13 07:32  Qwickly-Mod-B.kla.json
Aug  9 22:31  astarte.en.ansi.json
Mar  2 09:05  balanced13-iso-shifted-final-mod-ian.json
Mar  2 07:58  balanced13-iso-shifted-final.json
Mar  2 07:57  balanced13-iso-shifted-no-spacefn.json
Mar  4 10:24  balanced13-iso-shifted-v2-mod-ian.json
Mar  4 09:34  balanced13-iso-shifted-v2.json
May 31 11:47  beakl-15.en.matrix.json
Feb 11  2019  beakl15ModPqIntl3.kla.json
May 31 22:28  beakl19.matrix.json
Jun  5 08:05  beaklArr29k1.matrix.json
Mar 11  2019  code.en.ansi.json
Aug 13 10:47 'eNNe AS 83.86.en.json'
Dec 12  2017 'eNNe KLA3 P 108.27.json'
Dec 10  2017 'eNNe KLA3 P 109.54.json'
Dec 10  2017 'eNNe KLA3 P 110.97.json'
Dec  7  2017 'eNNe KLA3 P 111.69.json'
Aug 12 23:21  hycis.en.ansi.json
Jun  9 15:00 'kla BEAKL-19 ergodox.json'
Jun  9 16:31 'kla X1 Atreus-Ergodox 44-keys.json'
Aug 10 23:13  ntsc.en.matrix.json
Aug 10 16:32  ough.en.ergolinear.json
Aug 11 09:44  ough.en.matrix.json
Aug 11 10:07  ougw.en.matrix.json
Dec 29  2017  power.en.ansi.json
Aug 12 21:33  shz.en.ansi.json
Aug 12 10:14  shz.en.matrix.json
Aug 12 13:38  shz.en.matrix.zkq-dblequote.json
Aug 13 08:42  vflm.en.ansi.json
Aug 11 09:43  wiea.en.matrix.json
Dec 30  2017  zx-1.en.ansi.json

Last present : ANSI (30-keys, not full optimization) version of my most recent layout ... scores well.

Cheers, Ian

[Sc0tTy]

Nice write up!
Yeah I looked at your site a couple of weeks back and didn't see much updates then. its a real nice resource.

One of my issues with all corpuses are that they don't represent a real world scenario. That blog writer probably did several edits performing several different text operations.
I use KLA to see if the layout performs okay on the corpusses but I use my type data as the main source of input.

For instance over 11% of all keys hit are: pgup, pgdn, home, end and the four arrow keys. That is quite a lot.

Sadly its not possible to detect the percentage of auto shifted characters, I'd really like to see the reduction size of the shift key and that finger usage.

And in that regard I updated the layout a bit for a better match in my heat map and alsy the weight for the ring finger is better on top than on the bottom. It did reduce the score a little:

[Snarfangel]

I just wanted to mention that "finalbestkeyboard" just means the best one I could come up with (plus, I was running out of names!   ) . I did reprogram my Kinesis and am using it now, though.

[iandoug]

Nice write up!
Yeah I looked at your site a couple of weeks back and didn't see much updates then. its a real nice resource.

Have been occupied analyzing Giza the last few years, but trying to spend some time daily on various sites now.

One of my issues with all corpuses are that they don't represent a real world scenario. That blog writer probably did several edits performing several different text operations.
I use KLA to see if the layout performs okay on the corpusses but I use my type data as the main source of input.

For instance over 11% of all keys hit are: pgup, pgdn, home, end and the four arrow keys. That is quite a lot.

Yeah. I'm actually busy with design for my second physical board, and first is STILL not done (fiddling with QMK at the moment, that's a whole nuther story...)

But I also use the Nav cluster a great deal (not to mention backspace ....) and putting those on a layer won't work for me. So trying to find something that works.

Two years ago (on Den's site) we discussed mini-joysticks or D-pads for the arrow keys or PgUp/PgDn/Home/End, but their build quality is sadly lacking.
Now Den himself is starting to think along those lines ... :-)

The problem I think is you often need to hold shift or ctrl when using the Nav cluster.  I think I should run a keystroke recorder and see what my actual usage is like, because most of those keys are done in "automatic mode" (ie not conscious, the fingers just move....) and we're really aware  of just how much we use the nav cluster.

Maybe tweak KLA to handle "key capture logs", at least on local versions (wouldn't want to do that on a remote site ...)

Cheers, Ian

[Sc0tTy]

Ooo physical keyboard?! A whole new design or based on existing designs. I'm looking at a Dactyl ManuForm 6x6 atm for my next keyboard.

Yeah Ctrl and Shift are used heavily with the Nav-cluster, here you can see my top 4 key combo's according to my WhatPulse data (though key combo tracking seems a bit buggy)
.

Its the main reason why I don't like having those modifiers in the thumb cluster but on the sides. Same for other key combinations. If you need to press 2 modifiers for a shortcut you're doomed on the thumb cluster.

You could start by recording your data with WhatPulse, that's done securely. Sadly it doesn't record the actual letters you've typed so you don't know the bi- and trigrams you're typing. But it does allow you to make heat maps with the type data in Excel.
For instance with the layout above this is the finger usage:

(pinky is a bit skewed due to shift key)

[Snarfangel]

(snipped)
As for your layouts, will take a look. At the moment trying to do a round-up of recent new/updated layouts to do another round of testing. So if anyone reading this has a new layout, feel free to submit.

You shouldn't have asked!

I used the keyboard layout analyzer here: https://ieants.cc/code/keyboard/klatest/?#/config

I like the results better than my previous one -- that one overloaded the pinkies with O and R.

I need to head to work, so I may post more later.



 YOP_UIAN 1.txt (15.67 kB - downloaded 110 times.)

[Snarfangel]

To add a bit to the previous post, when tweaking things I used a large corpus made up of the collected works of Mark Twain (and other works in Project Gutenberg), a bunch of programming snippets from the Rosetta Code repository, a nice collection of my own writing, plus a plethora of other stuff. It didn't quite match the etaoin srhldcum fpgwybvkxjqz that Peter Norvig (replicating Mark Mayzner) found, more like etaoni srhdlucm fwgpvbvkxjzq (i-n switched, etc). Still, it is almost 30 million characters and 516,085 lines of text.

AS a quick-and-dirty test, I replaced the QWERTY keyboard in the dozen default keyboards at https://ieants.cc/code/keyboard/klatest/?#/config and used my text corpus, and got:



Not really fair to non-kinesis / Ergodox keyboards, though I did some comparisons earlier. The layout fire for the KLA is provided in the previous post.

Heatmaps look like:


(I'm using the layout as I type, albeit a bit slowly.)