I read through the java file. The learning does not seem to be doing much if anything at all.
After using it a lot here is my take:
- The start and end are locked down very well.
- It is fluid and fast. No problem with long words.
- The suggestions fit the gesture, maybe to well 😂
Now for the not so good:
- it has a preference for possessives: "word's" over "words" every time no mater how many times I correct it thus the learning doesn't seem present. And it is usually picking the lower frequency word over the higher.
- often chooses the lower frequency word: "wheat" over "what". This again doesn't change after selecting the correct word many times.
- gives preference to letters you fly over. Even when it is a lower frequency word and the main points of direction change fit the higher frequency word.
- no user input helps it with double letters.
- the beam path for error seems large especially for long straight fast swipes.
- f=1 words will not gesture no mater how many times you tap them out. You have to add them to the personal dictionary. I'm sure there are other f numbers, didn't check.
- does not recognize apostrophe comma or period in gesture.
- cannot deplete words
- does not apply blacklist
That sounds like it is horrible but it really isn't.
If this is something you feel like playing with for fun a couple things, if possible, that might help dial in things:
User settings for
- radius of error, beam width, for
- start point
- end point
- path, especially long straighter shots
- points where direction changes
- multiplier for word frequency above x
- penalty for low frequency below x
The other stuff like circle for double letter etc would be great, but the settings to see if it can be dialed in could prove fruitful.
The fact that it is so small and is very fluid and does as well as it does is actually impressive. I actually think that a few additions would make it very good though currently I cannot say the learning is doing anything whatsoever, so that alone working properly might be a big boost. 🤷🏻♂️
Edit
There is some learning taking place. It seems you need to do the word sometimes many times then switch keyboards but then there are some that will not alter, and the other big problem is the ranking of words by frequency, to many low frequency words fill all the suggestions while higher frequency words that fit the gesture are not present. But there is some slow learning, crank that up 😁
I read through the java file. The learning does not seem to be doing much if anything at all.
After using it a lot here is my take:
Now for the not so good:
That sounds like it is horrible but it really isn't.
If this is something you feel like playing with for fun a couple things, if possible, that might help dial in things:
User settings for
The other stuff like circle for double letter etc would be great, but the settings to see if it can be dialed in could prove fruitful.
The fact that it is so small and is very fluid and does as well as it does is actually impressive. I actually think that a few additions would make it very good though currently I cannot say the learning is doing anything whatsoever, so that alone working properly might be a big boost. 🤷🏻♂️
Edit
There is some learning taking place. It seems you need to do the word sometimes many times then switch keyboards but then there are some that will not alter, and the other big problem is the ranking of words by frequency, to many low frequency words fill all the suggestions while higher frequency words that fit the gesture are not present. But there is some slow learning, crank that up 😁