Gesture engine

[cinnabar777]

@LeanBitLab

Now that we have a working kotlin DTW engine I can integrate the stuff I've been working on in the Java version that was ripped out to make a simple dtw. So much of it should go pretty quick 🤞🏻

The AI have fallen down on integrating settings, so if you could add something into this initial code to the settings, maybe an enable/disable toggle specifically for it, that would properly direct the AI back to the right files, that might get them over this hump. The other thing would be a key code and toolbar button for manual "deep search". That way I nor the AI have to go mucking with anything outside of this main file and settings.

The plan:

1. Get the feature rich database added back
   1. It is already built in Java
   2. Tri-gram+ database
   3. Much of the features won't work without ties into the keyboard, and I'd prefer to leave that final step to you or copilot since it has access to all the code.
2. Add back in gesture engine features
3. Add settings
   1. Test test test
4. Add engine features not yet started
   1. Test test test
5. Optimize as much as possible
6. Initial word training
   1. Maybe you could find out what's happening with all the data HeliBoard is collecting?
7. Fully merge the database into the keyboard
   1. This is a big upgrade to the keyboard
      1. You or copilot on this part 😁
   2. I want to keep this as non-tied in as possible till its fully working and I'll put it in a repo for others to use

Feedback?

[LeanBitLab]

@LeanBitLab

Now that we have a working kotlin DTW engine I can integrate the stuff I've been working on in the Java version that was ripped out to make a simple dtw. So much of it should go pretty quick 🤞🏻

The AI have fallen down on integrating settings, so if you could add something into this initial code to the settings, maybe an enable/disable toggle specifically for it, that would properly direct the AI back to the right files, that might get them over this hump. The other thing would be a key code and toolbar button for manual "deep search". That way I nor the AI have to go mucking with anything outside of this main file and settings.

The plan:

1. Get the feature rich database added back
   1. It is already built in Java
   2. Tri-gram+ database
   3. Much of the features won't work without ties into the keyboard, and I'd prefer to leave that final step to you or copilot since it has access to all the code.
2. Add back in gesture engine features
3. Add settings
   1. Test test test
4. Add engine features not yet started
   1. Test test test
5. Optimize as much as possible
6. Initial word training
   1. Maybe you could find out what's happening with all the data HeliBoard is collecting?
7. Fully merge the database into the keyboard
   1. This is a big upgrade to the keyboard
      1. You or copilot on this part 😁
   2. I want to keep this as non-tied in as possible till its fully working and I'll put it in a repo for others to use

Feedback?

8d49c98

[LeanBitLab]

the kotlin engine seems faster, added some more tweaks like Locale & Script Agnostic, Early Score Pruning, Zero JVM Allocations

[cinnabar777]

I didn't see this. Yes, even in sandbox mode the Java wasn't that fast though i think it was still looking at the whole dictionary when it shouldn't have been, but the kotlin is faster than the java when i used it with a 30,000 word dictionary. This latest kotlin version is noticeably fast, at least i think it is. Things get blurry when you test too many builds.

You might want to hold off on doing to much to the engine itself right now. Use your time for the bugs cause there's a good amount to come and the AIs like to muck things from time to time 🤷🏻‍♂️ it is what it is. But I've got a better pattern of using one to chu check or correct the other 😂

[LeanBitLab]

I didn't see this. Yes, even in sandbox mode the Java wasn't that fast though i think it was still looking at the whole dictionary when it shouldn't have been, but the kotlin is faster than the java when i used it with a 30,000 word dictionary. This latest kotlin version is noticeably fast, at least i think it is. Things get blurry when you test too many builds.

You might want to hold off on doing to much to the engine itself right now. Use your time for the bugs cause there's a good amount to come and the AIs like to muck things from time to time 🤷🏻‍♂️ it is what it is. But I've got a better pattern of using one to chu check or correct the other 😂

gonna fix one at a time now on and release as beta build, as it supposed to be

[LeanBitLab]

The AI have fallen down on integrating settings, so if you could add something into this initial code to the settings, maybe an enable/disable toggle specifically for it, that would properly direct the AI back to the right files, that might get them over this hump. The other thing would be a key code and toolbar button for manual "deep search". That way I nor the AI have to go mucking with anything outside of this main file and settings.

didn't get you fully about this part, ill add the toolbar key, currently added the kotlin engine as third engine option (experimental)

[cinnabar777]

For some reason when they have to add settings into the kotlin file everything breaks. Gemini couldn't figure out where you moved things 😂

So I figured if there was already a tie in then they might get it 🤷🏻‍♂. Hey I'm not a software engineer. That's the AI's job.

[LeanBitLab]

2. I want to keep this as non-tied in as possible till its fully working and I'll put it in a repo for others to use

well the kotlin file is on the app now, lets leave it like there for the testing , i can make it hidden until debug mode on , lets start with something

[cinnabar777]

There are a few things that you ought to do instead of me having the ai direct me to certain files. I'll let you know as best I can with AI help 😂. Currently the database needs to be tied in to the dictionary but I'll give you that with the next code after I get a little more done.

[LeanBitLab]

There are a few things that you ought to do instead of me having the ai direct me to certain files. I'll let you know as best I can with AI help 😂. Currently the database needs to be tied in to the dictionary but I'll give you that with the next code after I get a little more done.

Currently(upcoming release) with debug mode on you can check the gesture engine, the toolbar key currently just added as an placeholder, we can add the function later after we got a clear plan

[cinnabar777]

Made net progress today 😂

Database integrated. There is an included fix from Gemini to tie it in, causing index issues.

Added back some gesture features: wiggle, circle, punctuation gesture for apostrophe, Deepseek found a Gemini bug 🤣 that seems too have improved typing, seems noticeably improve. 🤷🏻‍♂️

Attached are the complete dtw engine file plus a fix, not sure if that's in the SwipeGestureEngineKotlin.kt, pretty sure it is not.

Drop this in, tie in database and dictionary and then the fun stuff.

Gesture_Update.zip

[LeanBitLab]

facing some out of memory crashes right now, same thing happened on the early java, mainly having issues on the non english

[cinnabar777]

Bummer!

Let me know if there is something I need too do.

What setting file exactly should I be adding all the settings to? So I know when the ai directs me to do the wrong thing. Thanks!

[LeanBitLab]

Made net progress today 😂

Database integrated. There is an included fix from Gemini to tie it in, causing index issues.

Added back some gesture features: wiggle, circle, punctuation gesture for apostrophe, Deepseek found a Gemini bug 🤣 that seems too have improved typing, seems noticeably improve. 🤷🏻‍♂️

Attached are the complete dtw engine file plus a fix, not sure if that's in the SwipeGestureEngineKotlin.kt, pretty sure it is not.

Drop this in, tie in database and dictionary and then the fun stuff.

Gesture_Update.zip

Made net progress today 😂

Database integrated. There is an included fix from Gemini to tie it in, causing index issues.

Added back some gesture features: wiggle, circle, punctuation gesture for apostrophe, Deepseek found a Gemini bug 🤣 that seems too have improved typing, seems noticeably improve. 🤷🏻‍♂️

Attached are the complete dtw engine file plus a fix, not sure if that's in the SwipeGestureEngineKotlin.kt, pretty sure it is not.

Drop this in, tie in database and dictionary and then the fun stuff.

Gesture_Update.zip

lets go all in for kotlin engine, we dont need an seperate java engine

[LeanBitLab]

the kotlin engine will be removed from next build, lets add it after testing and if it become stable enough to replace java engine, it will be replaced

[LeanBitLab]

it complicated the welcome wizard and other settings

[cinnabar777]

That is actually better. I can just hack away bit by bit till the parts are there.

[cinnabar777]

Here is a current update, some fixes to the data and much better gesture and suggestions IMO

update.zip

[cinnabar777]

BTW

The big features in the database are the learning and cache that keeps words with their gesture and those suggestions in cache for proof reading until the keyboard closes then adds them to learned words, and of course syncing.

There was some tie in that i did with the Java version when i added dwell time over a letter, that should be in the engine now. If that's obvious to find it would be great to have it hooked up during this next phase. If not no big it was easy for the ai.

Nonetheless i think this last update is a good base. It is reasonably accurate with decent suggestions. Many iterations that i tried either had good accuracy and terrible suggestions or good suggestions and not so good accuracy. And this is with it having no training or learning properly. Ought to be good. 🤞🏻

[cinnabar777]

I just remembered that the database, whatever it is called, has a word limit on it for internally learned words BUT some people may be using their user personal dictionary as a small dictionary thus they could have significantly more words to sync than can be handled. I understand why people do this, i did it on HeliBoard for a while because it greatly helps gesture typing, however, the whole purpose of this project, for me at least, is to eliminate the need to do that.

[LeanBitLab]

I agree completely. A personal learning method is far superior to a universal one. The keyboard should become more personalized the more it is used, while universal methods can provide decent general suggestions, users eventually prefer and rely on the personalized suggestions that reflect their individual habits.

[LeanBitLab]

I agree completely. A personal learning method is far superior to a universal one. The keyboard should become more personalized the more it is used, while universal methods can provide decent general suggestions, users eventually prefer and rely on the personalized suggestions that reflect their individual habits.

not only suggestions and gesture typing i prefer this method for everything