• You are not logged in.

colemak or qgwmlb ?

  • Started by sasqautch112
  • 28 Replies:
  • Reputation: 0
  • Registered: 21-Oct-2012
  • Posts: 1

I went on this website http://mkweb.bcgsc.ca/carpalx/?full_optimization

to figure out what keyboard I want to get, and this site I think makes it pretty clear that colemak is better than dvorak , frustrating because i just switched from qwerty to dvorak. But before I make this last switch to another keyboard I want to make sure I chose the absolutely best keyboard for speed. I want to be a 150 wpm guy. (I understand that there are many arguments that changing keyboard layouts doesnt really increase speed, but i think it can).

so after looking on the site I saw that the best keyboard is apparently qgwmlb, now before I start learning some sort of alien keyboard I am wondering if learning this type of keyboard will keep me from being able to be a stenographer. I'm sure you can be a colemak stenographer, because colemak is so popular, but qgwmlb? uh, oh, I'm not too sure about that one.

Another question is whether or not the sites info is actually accurate, I mean, I don't even know what those numbers are, but it sounds good.




good thing I just bought a blank das keyboard...

Last edited by sasqautch112 (21-Oct-2012 22:03:54)
Offline
  • 0
  • Reputation: 1
  • From: Sofia, Bulgaria
  • Registered: 05-Mar-2011
  • Posts: 387

Firstly – yes, that site is well respected and provides good analysis on different layouts.

Now everybody has different preferences, that's why the "best" layout for different people is different. So you want to type fast, very fast. From my experience on high speeds – above 100wpm – the thing that stops you the MOST is pressing consecutive keys with the same finger. For me personally that's one of the biggest factors that limits your speed.

As you can see from Carpalx' website, the Colemak's same finger ratio is quite low. There are, of course, some situations when you'll have to press two keys with the same finger – that's inevitable – but in a lot of the situations you can "get out" of the same-finger trap by alternating your fingers (pressing one of the keys with a finger that's usually not responsible for that key). Getting to high speeds requires these techniques with any layout, but you probably know that, since I assume your Qwerty speed was very high, if you're aiming for 150 wpm.

And since your Qwerty speed is high, you'll inevitably benefit from switching to Colemak, since it is actually made so the switch from Qwerty is easy (much easier than Dvorak for ex.). Now you might be thinking you don't care about how similar it is to Qwerty, but reaching high speed is a long process, and depending on your learning curve, this "easy switch" thing can be a difference of years compared to other layouts. So you could potentially reach your goal much, much sooner. I'm not just guessing when I say years. A couple of years ago I switched to Dvorak from Qwerty myself and used it for a few months. I noted my progress. Then I found out about Colemak and switched again. Even after typing on Dvorak with almost my Qwerty speed, my progress with Colemak was TWICE as fast as that with Dvorak. And I'm currently typing with a good 40 wpm above my initial Qwerty speed.

About the QGWMLB layout – I have not thoroughly checked its "stats" but I know one thing – if you choose Colemak, you've got this whole forum with Colemakers to turn to for a question, advice, or just a chat. I somewhat doubt that QGWMLB can offer the same :). Also if you compare QGWMLB with Colemak, think about what I said for the difficulty of the switch.

I hope that helps.

PS: Something I forgot to mention: Some of the the information when comparing layouts on Carpalx' website – like "effort" is based on formulas that give different priorities to different stats. They may be different from yours. For example if I was looking for a layout to break speed records, I would consider most of these things, but with different priority than if I was looking for a comfortable layout for long hours of work. In the first case, I'd give most priority to lowest same-finger ratio, whereas in the second I woudn't mind that much typing consecutive keys with the same finger in comparison to something else – like maybe load distribution between fingers.

That's just my personal opinion. Yours and others' might differ. And btw the officially fastest Colemak typist currently is Ryan Heise with 126 wpm for 5 min on hi-games.net

Last edited by pafkata90 (21-Oct-2012 22:55:59)
Offline
  • 0
  • Reputation: 1
  • From: Tampa, FL, USA
  • Registered: 24-Aug-2012
  • Posts: 24

It's really impossible to tell because there just isn't enough real-world data to predict if one or the other would be faster for you.  They are both highly optimized.

I can tell you two things for sure though...choosing an alternative layout is an intensely personal decision.  Even if you go with Colemak, which is inarguably the more popular of the two, you will never encounter a keyboard "in the wild" which is already set up with your preferred layout.  Count on typing QWERTY whenever you're not on one of your computer.  So in some sense it doesn't make a difference as long as you find a layout that works for you.  It's nice to have a community around a layout, but in the end does that matter very much?  Not to your typing.

Second, having learned the Carpalx fully optimized layout, I can tell you it feels great.  I got up to my standard 80 wpm with it, and it was nice.  I forgot how to type QWERTY though, and ended up leaving carpalx behind because it was too hard to maintain typing speed with both, and both matter.  Colemak stands a much better chance of maintaining both because it changes fewer characters.  Where it really shines is low same-finger repetition, which means it's probably within an epsilon of carpalx when it comes to typing speed.

If I had to call it between the two, I'd give Colemak the nod due to support from a number of OS's for easily switching to it, as well as the closer layout to QWERTY.

The stats on the site are good...there's just room for disagreement on which are important. :)

Minimak - Better typing without losing QWERTY
http://www.minimak.org/

Offline
  • 0
  • Reputation: 0
  • From: California Coastal Living!
  • Registered: 24-Aug-2009
  • Posts: 65

Interesting thread.

I'm a happy Colemak user, but would consider looking at the various optimized layout "family" as listed above. I have always held back a bit when reviewing the stats, there is so much info to digest you need to be in a good position to digest it for it to be meaningful. And, if you're not motivated to change, then you're not in a good position for it to be meaningful :)

An Evil Screaming Flying Door Monkey From Hell typing with Colemak saved my life!

Offline
  • 0
  • Reputation: 3
  • Registered: 08-Dec-2010
  • Posts: 656

There are about 30 factors to decide whether the keyboard layout is good enough. The most important factor is  finger travel, but maybe Shai and the Carpalx researcher use different groups of factors, with different percentage of significance.

So depends on different factors we can produce thousand of different keyboard layouts, with efficiency +/- 0.5%. For us it is quite the same in term of efficiency.

QGWMLB and QWFPGJ (Colemak) are just two of those thousands of keyboard layouts that fits the bill. We can choose either of them.

Last edited by Tony_VN (22-Oct-2012 06:02:10)
Offline
  • 0
  • Reputation: 0
  • From: California Coastal Living!
  • Registered: 24-Aug-2009
  • Posts: 65

30 factors! Wow!

I can't believe there is any merit in changing or over-evaluation more than the several layouts that have already shown to be 'efficient' and observing the metrics to see what the focus of the layout was (i.e.: finger combos verses pinky workload, verses... etc). As well as giving it some time to consider peoples evaluation of these layouts. A 1% or less difference in efficiency, not to mention the subjective nature of the evaluations, can't be a difference worth considering. At some point the differences become negligible. I don't mean to say work on improved layouts should stop.

I'm thankful for the Colemak layout, and the website that offers the information needed to make these types of decisions. I also am thankful for the CarpalX site with all the 'mechanical details' though it is a much more in-depth and detailed site, it can also be a lot more than someone can digest.

An Evil Screaming Flying Door Monkey From Hell typing with Colemak saved my life!

Offline
  • 0
  • Reputation: 0
  • From: California Coastal Living!
  • Registered: 24-Aug-2009
  • Posts: 65

1) If you were to look at the metrics only
2) Using with a Kinesis Advantage

The one unique factor would be the ability to use a letter key with a thumb, like the "E" with the Malt layout. There may be a modified Colmak layout for this type of keyboard.

-Colemak
-QGWMLB
-Malt
-Arensito
-Workman

Not a very good post, but I wanted to make a note, since it's seems that's about one of the only things that would possible substantially change anything, and it's relegated to a specific type of keyboard.

An Evil Screaming Flying Door Monkey From Hell typing with Colemak saved my life!

Offline
  • 0
  • Reputation: 7
  • Registered: 21-Apr-2010
  • Posts: 807

> Colemak, which is inarguably the more popular of the two

Oh please, quit the hyperbole!

Is finger travel the most important factor?   Perhaps I should throw away my keyboard and use my blackberry...

--
Physicians deafen our ears with the Honorificabilitudinitatibus of their heavenly Panacaea, their sovereign Guiacum.

Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30

I'm currently deciding between these two layouts. While I do like that Colemak is more popular (which results in a bigger community/support and wider availability), I feel that when you're going for keyboard efficiency, changing keyboard layouts to an alternative will always be a pain regardless of how popular the alternative is.

In other words, if you're going this far down the niche for keyboard optimization, you may as well go all the way down. But then again, a few years from now and research could indicate an entirely new "best" keyboard layout, which would render the use of a completely obscure layout like QGWMLB arguably a waste of time. (But then again, QGWMLB would still be better* than Colemak/Dvorak/..!)

And I do like that Colemak is more QWERTY-compatible, since I aim to maintain both QWERTY (for world-compatibility's sake) and said optimized keyboard layout.

EDIT: On some more reconsideration, I think the best compromise is to know the universal layout (QWERTY), the most efficient* popular alternative (Colemak), and the most efficient* layout (QGWMLB). That way, you have QWERTY in the odd occasion that you need to use it, the highly efficient* and somewhat popular Colemak as your main, and the absolutely most efficient* QGWMLB for experimental reasons/fun. In terms of usage, it would be something like 70-80 WPM on QWERTY, 100+ WPM on Colemak, and 40-50 WPM on QGWMLB.

Should QGWMLB ever be dethroned, you can switch to that layout as your experimental one. And should it become more and more universally accepted as the best, then you can slowly replace Colemak with it. Of course, this is even assuming you can retain the muscle memory of three separate layouts, but I'm sure that after a lot of time, this could happen. I'm going to go ahead and start learning Colemak, and once I become quite fluent, I'll probably continue following this plan and experiment with QGWMLB.

*: in carpalx standards.

Last edited by nil (04-Nov-2012 21:42:57)

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 0
  • From: California Coastal Living!
  • Registered: 24-Aug-2009
  • Posts: 65

That's really way too complex buddy.

I believe that the ability to use an alternate layout will become increasingly easier as time goes forward...so the need to 'know' qwerty will diminish.  As much as it would be nice to 'just start from scratch' the reality is that it's not that easy because of the original qwerty foothold. At some point we just need to decide on a layout because there's effectively no difference/gain. It seems that Colemak is an excellent starting point... and although there are some small improvements over Colemak, but due to the remnants of qwerty, perhaps it's just best to go to Colemak and call it good. Or for the slightly more ambitious to adopt QGWMLB (or theoretical equivalent), but the problem is that there is no good argument for QGWMLB over MANY other layout incarnations...so there could be an endless 'this is better!' discussions and no accepted standard. Thus our dilemma.

So, I figure at this point I've got 30 years to live, and I've made the big move to Colemak, and until there's an equally big move forward in layouts, input device, or how an alternate layout can 'follow' you and be implemented at your location, etc....I'll save myself the major inconvenience of making another layout change for minimal gains. I'll keep the alternate layout ideas an an interesting point of ergo discussion :)

An Evil Screaming Flying Door Monkey From Hell typing with Colemak saved my life!

Offline
  • 0
  • Reputation: 1
  • From: Sofia, Bulgaria
  • Registered: 05-Mar-2011
  • Posts: 387

That's all correct unless switching for you is actually not "major inconvenience". Most people describe it as such, and it could be if you have to type a lot on daily basis – like at work, but if you approach it as fun experiment and challenge for your free time it could actually be enjoyable and fun experience.

I've switched twice my English layout (both times cold turkey) and designed my own native, which took a lot of testing with different layouts, but I haven't experienced the "torture" many describe. If I ever decide to try a new layout, I probably won't go cold turkey cause I really doubt the improvement over Colemak will be big enough to be worth it. But I am still open to try out something new as a brain challenge (like CHORDMAK maybe)

I'm currently using Colemak for English and my custom layout for Bulgarian and I can still almost touch type Qwerty with only a couple of looks down at the keyboard with decent speed. And I haven't spent any extra effort in maintaining these layouts or had any problems with confusion.

Last edited by pafkata90 (04-Nov-2012 22:43:19)
Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30
input nirvana said:

I believe that the ability to use an alternate layout will become increasingly easier as time goes forward...so the need to 'know' qwerty will diminish.  As much as it would be nice to 'just start from scratch' the reality is that it's not that easy because of the original qwerty foothold. At some point we just need to decide on a layout because there's effectively no difference/gain. It seems that Colemak is an excellent starting point... and although there are some small improvements over Colemak, but due to the remnants of qwerty, perhaps it's just best to go to Colemak and call it good.

I guess that's where we have our differences. I'm not sure I can compromise with a "good enough" solution as opposed to a "best" solution.

input nirvana said:

Or for the slightly more ambitious to adopt QGWMLB (or theoretical equivalent), but the problem is that there is no good argument for QGWMLB over MANY other layout incarnations...so there could be an endless 'this is better!' discussions and no accepted standard. Thus our dilemma.

But there is such a heuristic, and that is carpalx's typing effort model and its parameters. I haven't see any typing model as comprehensive as carpalx's nor have I seen a strong critique against it. The most popular I've heard is that it doesn't take into account penalties for same-finger typing (e.g. typing "rtfgvb" in QWERTY), but it does, and says it right on the homepage (search "reduction in same-finger typing"). The only legitimate concern carpalx's model I've found is truly determining the priority of the parameters and penalties, which is a problem in all models.

input nirvana said:

So, I figure at this point I've got 30 years to live, and I've made the big move to Colemak, and until there's an equally big move forward in layouts, input device, or how an alternate layout can 'follow' you and be implemented at your location, etc....I'll save myself the major inconvenience of making another layout change for minimal gains. I'll keep the alternate layout ideas an an interesting point of ergo discussion :)

I agree. After the vast improvement from QWERTY to any of the alternatives, any other change has a rather marginal difference. From here, it's not necessarily about efficiency a la time investment, but about theoretical efficiency and having fun with arguably more efficient layouts.

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 0
  • From: California Coastal Living!
  • Registered: 24-Aug-2009
  • Posts: 65
nil said:
input nirvana said:

I believe that the ability to use an alternate layout will become increasingly easier as time goes forward...so the need to 'know' qwerty will diminish.  As much as it would be nice to 'just start from scratch' the reality is that it's not that easy because of the original qwerty foothold. At some point we just need to decide on a layout because there's effectively no difference/gain. It seems that Colemak is an excellent starting point... and although there are some small improvements over Colemak, but due to the remnants of qwerty, perhaps it's just best to go to Colemak and call it good.

I guess that's where we have our differences. I'm not sure I can compromise with a "good enough" solution as opposed to a "best" solution.

input nirvana said:

Or for the slightly more ambitious to adopt QGWMLB (or theoretical equivalent), but the problem is that there is no good argument for QGWMLB over MANY other layout incarnations...so there could be an endless 'this is better!' discussions and no accepted standard. Thus our dilemma.

But there is such a heuristic, and that is carpalx's typing effort model and its parameters. I haven't see any typing model as comprehensive as carpalx's nor have I seen a strong critique against it. The most popular I've heard is that it doesn't take into account penalties for same-finger typing (e.g. typing "rtfgvb" in QWERTY), but it does, and says it right on the homepage (search "reduction in same-finger typing"). The only legitimate concern carpalx's model I've found is truly determining the priority of the parameters and penalties, which is a problem in all models.

input nirvana said:

So, I figure at this point I've got 30 years to live, and I've made the big move to Colemak, and until there's an equally big move forward in layouts, input device, or how an alternate layout can 'follow' you and be implemented at your location, etc....I'll save myself the major inconvenience of making another layout change for minimal gains. I'll keep the alternate layout ideas an an interesting point of ergo discussion :)

I agree. After the vast improvement from QWERTY to any of the alternatives, any other change has a rather marginal difference. From here, it's not necessarily about efficiency a la time investment, but about theoretical efficiency and having fun with arguably more efficient layouts.

I think it was Tony_VN had pointed out that there are many incarnations along the lines of QGWMLB (but not shown on the CarpalX website) that have effectively very similar matrices, typing distance, digraphs/trigraphs, etc. and perhaps many more to come....the issue being here is that there really can not be one 'best' solution... there are several/many 'best' solutions depending on your scale of judgement and methodology. This is by nature imprecise. That being said, there is not necessarily a 'wrong' solution either.

IE: Qwertiy is what it is. Then came Colemak. Colemak is much better in many ways (and manages to retain some legacy qwerty items). Great. But then came QGWMLB. Excellent, because QGWMLB is a tiny little better than Colemak in some ways (but does not have any legacy qwerty items). No problem, let's all switch to QGWMLB and we're finished! Here's the problem: If we know QGWMLB is the best and the final...let's do it. But....the issue is that there are at this point several if not many possible layouts that are equivalent to QGWMLB and unless someone can come up with a somewhat reasonable end, then it becomes like chasing the tail. There is no clear layout to switch to because the various differences all appeal to a slightly different matrix to address corresponding different issues. All this for an alleged fraction of a percent that may in all reality not be a factor at the end of a year. Do we want to create the best layout? Yes. I feel there needs to be something more definitive to justify such a minor switch.

DISCLAIMER: My problem is that I use a Kinesis and will always use a Kinesis or Maltron or my own creation that will also use thumb keys. So, you can see my dilemma. My layouts, buy using thumb keys, are far more 'efficient' when you put an "E" under a thumb (english language) and therefore I'm biased with my layout goal. QGWMLB is not good enough! Hahaha I've noted MALT and ARENSITO takes this into account just like WORKMAN takes a 'matrix' style keyboard into account. So it's not just about the layout...the keyboard is a factor as well.

Last edited by Input Nirvana (05-Nov-2012 19:43:37)

An Evil Screaming Flying Door Monkey From Hell typing with Colemak saved my life!

Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30

I understand your assessment and respect your right to an opinion. However, I don't believe in using just a "good enough" solution and even if you constrain yourself with more parameters (such as ZXCV fixed or more "QWERTY-friendly"), you can always find even alternatives to Colemak which (marginally) improve it in every imaginable way.

input nirvana said:

But....the issue is that there are at this point several if not many possible layouts that are equivalent to QGWMLB and unless someone can come up with a somewhat reasonable end, then it becomes like chasing the tail. There is no clear layout to switch to because the various differences all appeal to a slightly different matrix to address corresponding different issues.

But I don't think there are "several if not many possible layouts that are equivalent to QGWMLB." If we're adhering solely to Carpalx's parameters and model priorities, then all such heuristics (e.g. simulated annealing) converge to the same solution, which is QGWMLB. Should we find a better adjustment for the model's priorities, then sure, we could potentially get an entirely new layout. But we'll still be quite confident that QGWMLB is better than previously iterated/compared models, e.g., Colemak.

input nirvana said:

All this for an alleged fraction of a percent that may in all reality not be a factor at the end of a year. Do we want to create the best layout? Yes. I feel there needs to be something more definitive to justify such a minor switch.

As I said before, I agree completely. After the switch from QWERTY, any other layout switch really only makes a marginal improvement. But say you're still at QWERTY and deciding a new layout. Do you go for Colemak just because it's "good enough" and somewhat popular, or do you go for a model that even if dethroned in the future, will still be better than Colemak? Also, as a remark, many experienced typists who know more than one layout simply enjoy switching layouts and experimenting, not because of some false hope that there's an efficiency gain from the poor time investment, but simply because of the fun involved with trying new models.

Last edited by nil (05-Nov-2012 08:32:55)

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 124
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,842

I think you need to realize that CarpalX ≠ typing! If you choose to completely trust its modeling (and out-of-the-box to boot!) then feel free – and it is an enchantingly simple way of choosing. Other modeling efforts of varying quality give different results of course.

The seeding (i.e., the choice of weights and parameters) and interpretation of the models will always be heavily influenced by the researcher's preferences... which again are influenced by layout and typing history of that individual, along with individual needs and peculiarities (such as being a programmer from Russia with weak pinkies or whatever).

To take myself as an example, I don't feel that the upper-row positions for the middle and ring fingers are worse than their home row counterparts (then again, I have long fingers). I hate same-finger digraphs with a vengeance. I don't mind inward rolls and I don't necessarily feel that hand alternation makes me euphoric (then again, I've played the piano). My pinkies are strong but don't like to stretch outwards. I abhor wrist twists, hence my ceaseless advocation of the Angle (and Wide) ergonomic mods; these again change the effort somewhat for some keys (left-hand bottom row much easier on the wrist, the Colemak B position a lot easier, the Colemak G position a little harder but acceptably so). I type a lot of English but also some Norwegian (which has some fun digraphs) and code. Your mileage can and will vary!

I think your confidence may be a little on the high side. Reading CarpalX's own pages on the modeling of typing effort I at least perceive more humility. ;)

Last edited by DreymaR (05-Nov-2012 10:34:27)

*** Learn Colemak in 2–5 steps with Tarmak! ***
*** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30
DreymaR said:

I think you need to realize that CarpalX ≠ typing! If you choose to completely trust its modeling then feel free – and it is an enchantingly simple way of choosing. Other modeling efforts of varying quality give different results of course.

And the seeding (i.e., the choice of weights and parameters) and interpretation of the models will always be heavily influenced by the researcher's preferences... which again are influenced by layout and typing history of that individual, along with individual needs and peculiarities (such as being a programmer from Russia with weak pinkies or whatever).

I think your confidence may be a little on the high side. ;)

It's less a trust/confidence and more an actual agreement with his priorities and parameters. I've read through all the documentation and I'm in agreement with almost all the work.

It's also a nice assurance that QGMLWB is "completely open source," meaning that I can see how it's developed from bottom-up. Look at other alternative keyboard layouts (e.g. Colemak), and I really am forced to instill a heavy amount of trust into the researcher's parameters, and whether or not they suit my own preferences. After all, doing statistical comparisons can only mean so much, when the foundations of the development may be completely different and instantly tell you what the said typing model is all about. Taking Colemak as an example again, I don't agree with the design philosophy to keep it "QWERTY-friendly" (fix 'ZXCV' and preserve as many others as possible), nor do I agree with the philosophy to weigh in aesthetics into the layout. I feel these add unnecessary constraints, and so I would be more in tune with QGMLWB.


EDIT: Also, I should add that I believe in an efficient placement of the entire keyboard, which includes how one arranges modifiers and other commands. For example, I really like how close Colemak puts Backspace; I always felt a shame that such a commonly used key is so far away. In my current setup, I'm using QGMLWB with Backspace mapped to Tab, Tab mapped to Esc's location, Delete mapped to Backspace's location, and overloading both CTRL and Escape into Capslock. (I'm a Vim guy.) I'm also experimenting with other modifiers (e.g. Hyper, Meta) and other features (sticky shifts).

Last edited by nil (05-Nov-2012 11:48:31)

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 0
  • From: California Coastal Living!
  • Registered: 24-Aug-2009
  • Posts: 65

Just to clarify...I'm not specifically arguing any of your points, they are all valid in principle, if not practical applications...I agree except for the one tiny item that I personally wouldn't switch unless the next layout is a clear, definite and relatively final layout decision. I recognize QGMLWB 'may' be that, as there are others...but they have not shown this to be the case, definitively. And, any change from any of the QGMLWB/equal layouts would be pointless other than an exercise in study.

If I was starting from scratch, I may probably choose to try a QGMLWB or equivalent (still assuming there has been a solid end game shown with the various metrics)....because it seems to make sense. But, I started qwerty.....then Colemak...so there is track record that needs to be converted. Today, if there were a multiple sourced, tried and true, highly optimized layout that could easily be the 'end game'...I would (and still may) switch from the 'standard' Colemak to a partially or fully optimized Colemak or QGMLWB/equal. BUT....at this point I don't believe that. I'm not right or wrong...it's just my current assessment. Keep in mind, I have another goal with my Kinesis/thumb keys layout....so I'm going to be somewhat biased until I get to the point where the KINESIS/MALTRON keyboard physical layout is full optimized.

:)

An Evil Screaming Flying Door Monkey From Hell typing with Colemak saved my life!

Offline
  • 0
  • Reputation: 124
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,842

About the "aesthetics" issue: It's not about pretty or not so this word's misleading here. It's about the utility and recognizability, especially in a QWERTY-centric world. If it isn't broken (noticeably), don't "fix" it. (Or should that be, fix it - as in, fix that key's position? Hehe.)

Therefore, I believe in keeping the symbol keys in place unless you have special needs (in which case you should strive to use a modular approach - I love those!). I have used Dvorak so I know that messing with those keys is a pain in the butt. Same with the ZXCV "shortcut" keys, where Dvorak had me closing windows instead of pasting into them because it put the W and V next to each other on the right-hand side which means horror stretches if you try to mouse while copy/pasting.

That's the kind of considerations we're really discussing here, not just "ooh, that layout's purdy!". That much said, I do agree with Shai that if you make a butt-ugly layout it's going to be much less popular with the general public. Given how little the difference in effort seems to be between the multitude of "best" and "next-best" layouts while considering the substantial uncertainty in the model results, you might as well introduce some fairly useful extra constraints to get a layout that's just as efficient as far as we can prove but also nicer.

But we'll agree to disagree on that point like we do on the virtue of not moving too many keys, right? ;)

Last edited by DreymaR (05-Nov-2012 16:33:20)

*** Learn Colemak in 2–5 steps with Tarmak! ***
*** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30
DreymaR said:

About the "aesthetics" issue:
...
I have used Dvorak so I know that messing with those keys is a pain in the butt. Same with the ZXCV "shortcut" keys, where Dvorak had me closing windows instead of pasting into them because it put the W and V next to each other on the right-hand side which means horror stretches if you try to mouse while copy/pasting.

You can argue however useful the constraints may be on a practical standpoint, but efficiency-wise, they will always be unneeded constraints, hence slowing down the layout. I'm an optimization guy, so even if it's a loss of .0001%, making it more QWERTY-friendly or tuning in aesthetics will be bad in my perspective.

DreymaR said:

That's the kind of considerations we're really discussing here, not just "ooh, that layout's purdy!". That much said, I do agree with Shai that if you make a butt-ugly layout it's going to be much less popular with the general public. Given how little the difference in effort seems to be between the multitude of "best" and "next-best" layouts while considering the substantial uncertainty in the model results, you might as well introduce some fairly useful extra constraints to get a layout that's just as efficient as far as we can prove but also nicer.

But the layout isn't as efficient "as far as we can prove." Looking back at the previously linked thread, Shai made the recognition that the semicolon was more effective when placed between B and M. But he made the aesthetic-choice (to leave it in the QWERTY position) rather than the efficient-choice (to move it). And however useful this is from a "fix QWERTY as much as possible" perspective, it's very clearly a loss in efficiency. Now, just how much of a loss? Well, that's something we can't quite determine exactly, but it is at the very least a loss, no matter how big or small.

DreymaR said:

But we'll agree to disagree on that point like we do on the virtue of not moving too many keys, right? ;)

I guess so. :]

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 124
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,842

My strong feeling is that the uncertainties in your modeling are far greater than what you realize. So you follow that model to its conclusion but you really don't know whether it is in fact better than the top competition. Your talk of .0001 % forces me to smile wryly. Again, CarpalX is a lot more modest in his claims there. If you think his calculations with their inherent uncertainties constitute proof of his layouts' efficiencies then you're bolder than both me and him.

Last edited by DreymaR (05-Nov-2012 22:55:50)

*** Learn Colemak in 2–5 steps with Tarmak! ***
*** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

Offline
  • 0
  • Reputation: 1
  • From: Tampa, FL, USA
  • Registered: 24-Aug-2012
  • Posts: 24

I have to agree with DreymaR here.  These designs are stochastic in nature...they are not subject to absolutes, only characteristics in response to certain families of inputs such as which language you type in, whether you are writing formally or casually, whether you are programming, etc.  No one design can be totally optimized.  Even carPalx's designs are the results of thousands of simulations where there may be a different "best" design at the end of each run.  He just chooses one that wins more frequently out of those.

Minimak - Better typing without losing QWERTY
http://www.minimak.org/

Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30
lilleyt said:

I have to agree with DreymaR here.  These designs are stochastic in nature...they are not subject to absolutes, only characteristics in response to certain families of inputs such as which language you type in, whether you are writing formally or casually, whether you are programming, etc.  No one design can be totally optimized.  Even carPalx's designs are the results of thousands of simulations where there may be a different "best" design at the end of each run.

You may have misunderstood me. I've never said that there exists one true layout god. When I'm talking about "efficiency" or "effectiveness," I mean what is normally meant by this (i.e. based on overall averages and logical parameters). I do not mean that layout A will always be better than layout B at every single typing task ever known.

lilleyt said:

He just chooses one that wins more frequently out of those.

This is not technically correct. The algorithm runs the program, outputs a layout, and measures its typing effort, reiterating this process as many times as desired; then the layout with the least typing effort is designated as the optimized one. You should not be critiquing the computation itself, as it is logically sound. The problem, acknowledged by Krzywinski himself, is setting the right parameters.

DreymaR said:

.. So you follow that model to its conclusion but you really don't know whether it is in fact better than the top competition. Your talk of .0001 % forces me to smile wryly.

I'm not sure what you're arguing here. There is no doubt that the aesthetic-choice is inherently less effective than the effective-choice. Any other belief would be flat-out denial. Colemak was not inherently designed to be as efficient as possible, but merely an "efficient enough" layout that's also quite QWERTY-compatible. And as stated before, just how less efficient does its inefficient constraints make it? We don't know, but at the very least we do know that they are indeed inefficient constraints.

DreymaR said:

Again, CarpalX is a lot more modest in his claims there. If you think his calculations with their inherent uncertainties constitute proof of his layouts' efficiencies then you're bolder than both me and him.

Do I think QGWMLB is the most efficient layout possible? Certainly not. I never claimed that it is the absolute "bee's knees". There is still a lot to be desired (mostly stemming from individual preferences, such as typing "jn" in QWERTY using the middle then index fingers, or hand rolling vs alternate hands). As such, a lot of our decisions should be based not only on statistics we don't fully understand, but on the core issue at hand: the design philosophy of the layout itself. If a layout is designed to be bad, then you can surely expect it to be worse than a layout designed to be good. Analogously, if Colemak is designed to add inefficient constraints, then it is unlikely for it to be as effective as one designed without those constraints. That's the simple matter here, and you can debate the practicality of these constraints all you want, but the fact of the matter is that they are inefficient constraints that worsen the layout. Whether or not we can truly determine whether Colemak or QGWMLB is more effective, I'd say that the arguments for QGWMLB make it the more likely candidate.

EDIT: Note that this is merely a probabilistic argument. But given how individualistic preferences can change a layout's parameters, a probabilistic argument is likely the best argument we can ever get.

EDIT2: Also, I realize the arguing against Colemak on a Colemak forum would be like arguing against Christianity in a Christian church. Hence, I don't expect to see anyone agreeing with my arguments or perspective, but I'm at least glad I got them across.

Last edited by nil (06-Nov-2012 01:29:58)

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 1
  • From: Tampa, FL, USA
  • Registered: 24-Aug-2012
  • Posts: 24

Sure thing, but I think you might want to go back and reread the tutorial on simulated annealing:

"Usually multiple runs of the simulation are necessary to explore the family of solutions to identify the best one because the phase space of solutions is generally complicated and many local minima exist.

Simulated annealing is one of many types of stochastic optimization algorithms. Because simulated annealing has its roots in physics, the quantity that measures a solution's fitness is frequently refered to as the energy. The algorithm's role is to therefore find the solution for which the energy is minimum. The process of finding this solution begins with starting with some initial guess, which is frequently a random state.

For example, you would start with a keyboard layout on which the keys are randomly organized. Since each potential solution has an associated energy, you compute the energy (E0) and set this value aside. Next, you adjust the solution slightly to obtain a marginally different solution. The size of this adjustment is cruicial. If the step is too small, it make take forever to move to a significantly different state. If the step is too large, you risk jumping about the solution space so quickly that you will not converge to a solution. In the case of the keyboard, a reasonable step from one solution to another is a single or double keys swap. Once the new solution has been derived, you compute its energy (E1).

Now that you have E0 and E1 is where things get interesting. Let dE = E1 - E0 (I am using dE loosely here - it's not a differential quantity but just a difference). You might suppose that if dE < 0 then you should value the new solution over the old one, since its energy is lower. You'd be right - in this case, simulated annealing transitions to the new solution and continues from there. Let's skip over the dE = 0 case for now. What about when dE > 0? In this case, the new solution is less desirable than the old one, since its energy is higher. Do we discard the new solution? Let's consider what would happen if we did.

If all transitions with dE > 0 are rejected, the algorithm will be greedy and descend towards the lowest-energy solution in the vicinity of the current state. If the step from one solution to another is small, then a minimum energy state will be found. However, this may be a local minimum instead of a global one. If all paths from the local minimum to the global minimum involve an increase in energy (imagine the local minimum being a trough), these paths will never be explored. To mitigate this, and this is where the "stochastic" part of simulated annealing comes into play, the algorithm probabilistically accepts transitions associated with dE > 0 according to the following probability



where dE is the energy difference and T is a parameter that is interpreted as a temperature. The basis for this interpretation originates from the original motivation for the form of of the transition probability.

Given a candidate transition and dE > 0, the algorithm samples a uniformly distributed random number r in the range [0,1) and if r

The role of the parameter T is very important in the probability equation. This parameter is continuously adjusted during the simulation (typically monotonically decreased). The manner in which this parameter is adjusted is called the cooling schedule, again with the interpration of T as a temperature. Initially, T is large thereby allowing transitions with a large dE to be accepted. As the simulation progresses, T is made smaller, thereby reducing the size of excursion in to higher energy solutions. Towards the end of the simulation, T is very small and transitions with only very slight increases in E are allowed.

In carpalx, T is cooled exponentially as a function of the iteration, i.



The exponential form of T(i) has nothing to do with the exponential form of the transition probability. It's just a cooling schedule that I adapted, admittedly without testing any other form.

The effect of the cooling schedule is to allow the algorithm freedom of movement throughout the solution space at first (large T), then progressively converge to a solution with a minimum (medium T) and converge to a minimum (small T). The optimization simulation is typically run many times, each time with a different starting state.

parameter values
Choosing values for p0, T0, k and N should be done carefully. The possible range of values for E must be anticipated and T will need to be scaled appropriately to produce useful probability values.


Figure 1 | Cooling schedule for the optimization. As the optimization runs, the probability of accepting a less desirable layout exponentially decreases.
For carpalx, the typing effort ranges from 9 (QWERTY, mod_01) to about 4.5-5 (optimized, mod_01). Consequently, I have chosen the parameters as follows

T0 = 10
k = 10
T0 scales the temperature so that at maximum energy, E/T is approximately 1. With a value of k=10, the temperature drops by 1/e for every 1/10th of the simulation. Initially, the temperature is T(i=0) = 10. At i=1000, (1/10th of the simulation), T(1000)=10/e. At i=2000, (2/10th of the simulation), T(2000)=10/e2, and so on. At the end of the simulation, T(end) = 10/e10 = 4.5x10-4. If you want the temperature to drop faster, use a larger value of k. To start with a cooler system, set a lower T0.

For all simulations, I've used

p0 = 1
N = 10000
"

tl;dr: There is a single corpus of input, against which he chooses a random starting configuration for a keyboard.  Each step introduces a random keyswitch which is evaluated for goodness.  If it's better, it's kept.  If not, there is a probability it will be accepted anyway, a probability which decreases as each step is taken.  Finally, the algorithm converges on a minimum that is best for the area within a monotonic slope from that point.

Because the problem space is non-monotonic, the given result for a run cannot be proven to be optimum across the problem space.  Because the problem space is np-hard, it cannot be solved in polynomial time (ie, brute-forced) so your result is only an heuristically chosen optimum and cannot be proven so, only disproven.

Minimak - Better typing without losing QWERTY
http://www.minimak.org/

Offline
  • 0
  • Reputation: 0
  • From: Berkeley
  • Registered: 04-Nov-2012
  • Posts: 30
lilleyt said:

Sure thing, but I think you might want to go back and reread the tutorial on simulated annealing:
...
Because the problem space is non-monotonic, the given result for a run cannot be proven to be optimum across the problem space.  Because the problem space is np-hard, it cannot be solved in polynomial time (ie, brute-forced) so your result is only an heuristically chosen optimum and cannot be proven so, only disproven.

Of course. We're not 100% sure that the layout is even the lowest in typing effort given the current parameters. But given enough simulations, one becomes only more and more confident that it is. After all, it's a heuristic argument, and theoretically proving that it's the absolute lowest, while practically impossible, is not necessary on the practical standpoint. So long as we are 99% sure it has the lowest typing effort (using whatever confidence level you desire, of course) is fine by me.

And besides, the computation is universally accepted, and having taken a grad course on ergodic theory myself, I'm well aware of stochastic processes and the problems of its application. But it isn't really the main problem here; the main problem is dealing with the parameters. Even the smallest adjustment would radically change the converging solution (say you'd prefer .1% more use of the pinky before it should be penalized). And as such, arguments shouldn't be made according to what layout is more effective than another (e.g. Colemak vs Dvorak), but rather what parameters are more preferable/effective than others (i.e. just how much hand rolling do you want versus hand-alternating? how much pinky-usage is enough?).

Colemak (start 11.5.12): ~80 WPM.
QWERTY: ~90 WPM.

Offline
  • 0
  • Reputation: 1
  • From: Tampa, FL, USA
  • Registered: 24-Aug-2012
  • Posts: 24

I think we have different definitions of several of our terms, since to my eye you seem to be contradicting your earlier points.  Clearly you've got a strong grasp of it though, so I'm not worried.  Cheerio.

Last edited by lilleyt (06-Nov-2012 05:06:43)

Minimak - Better typing without losing QWERTY
http://www.minimak.org/

Offline
  • 0