Engram Scoring Model has no meaningful (efficiency) range

[eugenesvk]

While reading the "Engram Scoring Model scores (x100) for layouts" I've noticed that there is almost no difference between various layouts. For example, on the "Google bigrams" corpus, QWERTY is 62.19 vs Engram's 62.48, so just a 0.5% (!) improvement, nothing that would entice anyone to go through the pain of switching to a non-standard layout.

The KLA scores not only have some range (70.13 vs. 53.06 for Engram vs QWERTY, so a 32% win), but are also somewhat tied to real-word efficiency, e.g., the first factor distance your fingers moved is easy to understand (though it's just a proxy for the real effort, so any more direct measure of the effort would be better)

Is there a way to present this data in a more meaningful way that would approximate real-word efficiency improvements in using different layouts to make the comparison easier?

[binarybottle]

@eugenesvk -- Thank you for your comment!

When scoring systems result in similar scores for dissimilar inputs, this could either mean that the scoring method is insensitive to differences in the inputs, or that the score values represent nonlinear differences. A colorful example of this is distance measures in the CIE 1931 color space (http://www.colorbasics.com/CIESystem/). Percent differences are relative comparisons -- in this case, the absolute values aren't important. In the Jupyter notebook, you can see how the Engram scoring model incorporates interpretable, physical measures, and you can just use these if you'd like, but I feel strongly that this type of optimization problem demands accounting for a variety of relevant factors simultaneously. That is why I created the scoring model.

I welcome suggestions for presenting results in a more helpful way that approximate real-word efficiency improvements!

[eugenesvk]

in this case, the absolute values aren't important

But then how would you decide if your layout is meaningfully better to justify the switching costs?

demands accounting for a variety of relevant factors simultaneously

That's obviously true, we do need a multi-factor model, that's why I can't really use the individual metrics directly

I welcome suggestions for presenting results in a more helpful way that approximate real-word efficiency improvements!

My first naive thought is that you'd need to normalize each factor on, e.g., a 0-100 scale, where <10 is awful and >90 is excellent, then there is this obvious challenge of which range of the underlying "physical" metrics (lateral distance, same finger utilization, roll-ins etc.) maps to this scale. Maybe for some metrics you could get down to some fundamental value like improving speed or reducing harm, e.g., in the following hypothetical example of perfect knowledge, for a daily use of 8 hours of heavy typing:

• a lateral distance of 10 meter a day would lead to a RSI within 1 year
• 5 meters a day would give you 5 years before you get a RSI
• 1 meter would not lead to any
• 0.5 meters would not lead to any even if you type 24 hours a day :), so that's the "100" ceiling for this metric, above which any optimization is practically meaningless Though mostly you'd not have this level of insight into the metrics tying them to real world values of speed/harm, but then that's where you'd need an keyboard design expert such as yourself :) to weigh in, that's would still be better than no equalization of parameters' value ranges

[binarybottle]

But then how would you decide if your layout is meaningfully better to justify the switching costs?

Regarding absolute values, there are different ways to focus on seemingly small differences. One way is to threshold to a meaningful range (like changing the dynamic range of intensity values in an image), and another is to change to a nonlinear scale (think gamma correction of an image).

To determine if any measure is meaningful, we first need to establish what is the goal. For some, it would be increased speed or accuracy. In my case, I'm more interested in improved comfort and safety. To take this work to the next stage, I would need to have a sample of people adopt the Engram layout and either record subjective questionnaire responses, objective physiological (e.g., electrodermal activity) data, or monitor longitudinal data of injuries or recoveries. And to make meaningful comparisons with other layouts, this would mean putting them on an even footing, which would probably entail a random control trial of young people randomly assigned a layout, or layouts, to learn. Needless to say, I'm not in a position to conduct such a trial, and it would take considerable work to gain access to an injured population willing to take part in such a study.

[Tristen-Sinanju]

as we know, it's entirely subjective, but i find it meaningful to point out that i switched to this layout because i experience a lot of pain while typing. doesn't particularly matter what kind or style of keyboard i use. (i will be consulting with a doctor soon) ... but for the common English language, engram has helped a lot. i can even sustain 60plus on good days and burst into the hundreds. it's a little weird with "ye olde english"... but seeing the corpus the layout was tested against, that makes sense, and is completely understandable. (it's a mostly dead way of writing, after all)

the lowered lateral motion has been great. i can plant my wrists and don't have to rotate them in any weird directions for word combinations.... overall, it's been great for my particular case.

i can't even remember any of the previous layouts i learned, if i'm honest

[binarybottle]

@Tristen-Sinanju -- Thank you for telling me this. First, I'm terribly sorry to hear about your pain (have you tried voice dictation?). It means a lot to me too hear that the work I've put into the Engram project has been helpful for you. I really would like to find ways to measure the difference this layout makes for people. Until you wrote about your own condition, I hadn't even considered including subjective self reports (testimonials) in the site, but an account of your experience may help others until we can furnish objective evidence substantiating these experiences.

[Tristen-Sinanju]

Hopefully, we're just looking at a case of carpal tunnel, but i'm pretty sure there are other underlying conditions i hope to identify soon. life is what life is; i don't feel too much one way or another... but if testimonials are helpful, i've been using engram about 9 months as my primary layout.

overall the things that have been most beneficial about this layout have been the preference towards towards a healthy amount of alternation and comfortable rolls. the position, and use of the bottom two corners as preferential areas isn't something i've seen a lot in other layouts, and i often feel like those are wasted keys in other layouts (where you could put a perfectly good letter, but they put something like 'z' 'q' or punctuation instead).

I, personally, can't reach the upper outer corners without an alt-finger or a wrist rotation and stretch, but it's easy enough to alt-finger with a ring finger, or reassign on a programmable keyboard (my thumb 'b' and 'q', for instance)... and the corners are easier to reach on a standard stagger... so no problem. things like 'science' and 'ice' and redirects like that are still difficult for me, but i assume that's because of the current degradations to my coordination

i really dislike the "jo" bigram, and any "jo-" trigrams that are left hand only ... but their frequency is too low to be a sustainable complaint. (to that end, i'm considering a j-to-thumb-q, q-to-v, v-to-j rotation, but i haven't tried it yet... there's no way i can put q onto left hand since its most frequent bigram is 'qu', hah)

the main driver was a preference against center column and lateral movement. in no particular order... dvorak's 'i' 'D' 'L', for instance were problematic to me. 'i d" was in the center column, and 'L' is in an upper corner (which we know i can't really reach) ... or for mtgap. 'y' 'w' (too frequent of letters to not be able to reach them, and engram moved them in) and punctuation on the bottom index is way better served with a letter. workman ... it had all the vowels on the wrong side and i couldn't cope XD. (it wasn't particularly good anyway... nice rolls, but the rate of sfb's was uncomfortable)

all of the layouts i moved to and moved from had a preference against center column usage, increasingly, till i landed on engram for the foreseeable future.

[binarybottle]

Thank you, @Tristen-Sinanju. All the best in reducing your pain.