proximity prediction

[jab123]

Hi
I love the keyboard! It would be much improved if the text prediction could predict for more words according to the proximity of keys. Proximity prediction of text would basically be a combination of the possible surrounding letters that could make up a word. for example, if i typed "prpplr" the suggestion would be people. another would be, "grlli" as "hello", etc. Would it be possible to come up with an algorithm that could predict based on the proximity of keys hit? the stock android keyboard does this fairly well any way to implement this?

Thanks

[cyril]

Hi
that's quite worrying... proximity correction works since the beginning (at least since there is a dictionary) with the soft keyboard.
On my phone (and the emulator) "prpplr" suggests "people" (though grlli doesn't work for some reason, I will have a look)

Can others confirm ?

[jetrost]

For me, "grlli" brings up "hello" as a suggestion, but doesn't automatically highlight it. "Prpplr" did suggest (and highlight) "people" though.

[cyril]

Ok so it's the same for me... and by the way "grlli" doesn't behave... better with Better Keyboard

[jab123]

sorry i meant that it would help if the proximity applied also to the bottom keys. for example if the needed letter is t, it would also register for the correction for g and h. i know it does this for some but i found most do not allow. thanks

[OKK77]

Proximity prediction can be improved certainly. Android keyboard and HTC Touch Input work very well (with Touch having a slightly better accuracy).

[cyril]

Proximity prediction can be improved certainly. Android keyboard and HTC Touch Input work very well (with Touch having a slightly better accuracy).

Can you give precise examples ?
Smart Keyboard uses the same code as the stock keyboard for proximity prediction, so there should not be any difference! Actually the only difference is that the layout is not the same, so the distance between the keys may be bigger in some cases, and this have an influence on the algorithm (as it detects which keys are close to the pixel on which you pressed the screen)

[OKK77]

Proximity prediction can be improved certainly. Android keyboard and HTC Touch Input work very well (with Touch having a slightly better accuracy).

Can you give precise examples ?
Smart Keyboard uses the same code as the stock keyboard for proximity prediction, so there should not be any difference! Actually the only difference is that the layout is not the same, so the distance between the keys may be bigger in some cases, and this have an influence on the algorithm (as it detects which keys are close to the pixel on which you pressed the screen)

I will take note of the words next time and report here. Anyway, I'm loving this keyboard.

[OKK77]

One strange thing I observed right now:

I was typing the word "bus" and hit "bua". It was not able to predict it at first. So, I deleted the "a" and typed "s". The next "bus" I typed, it predicted fine. "bua" and "bus" were shown and "bus" is highlighted.

The same phenomenon was observed when I typed "wanting" as well.

[jab123]

One strange thing I observed right now:

I was typing the word "bus" and hit "bua". It was not able to predict it at first. So, I deleted the "a" and typed "s". The next "bus" I typed, it predicted fine. "bua" and "bus" were shown and "bus" is highlighted.

The same phenomenon was observed when I typed "wanting" as well.

i found out that the proximity prediction is related to how close to the correct letter than you're pressing. the algorithm tries to predict the likelihood of the key that you pressed is the one that you intended to press. so if you type "bua" and the "a" has been stuck more towards the left, the algorithm will suggest "bua" because it thinks that the "a" is the one you intended. but if you try it again and press "a" slightly towards the "s" then the likelihood of a mistype is represented higher so it suggests "bus".

so when i first suggested a proximity prediction problem i did not know this. i've been making a concerted effort to type more "cleaner". a possible solution would be to implement a larger radius for the algorithm to suggest keystrokes. but this obviously may increase the chance that suggestions would be wrong.

or perhaps a spell checker would fix this problem.

we'll see what goodness cyril has in store for us.