ildarPlease please fix!
I've added docstrings for all non-trivial functions in the gesture system. Now I'm working on assessing its accuracy by simulating noisy gestures.
The current system uses a corner matching algorithm to classify gestures. The Keyboard object starts by loading a dictionary and feeds it to the gesture detector object which then generates paths of corners for every word. To do so it first creates a sequence of points by getting the location of the center of each key needed to type the word, then it keeps those points that are at an angle > 170° from each others.
Afterward, when the user traces a gesture, the points the gesture goes through are passed to the gesture detector through keyboard events. Once the gesture is over, only the corners (angles > 170°) are kept. Then the dictionary if filtered to keep only the words who start by the key that was pressed at the start of the user gesture. A cumulative euclidean distance is then calculated between the corners of those words and the corners of the traced gesture.
We keep the n-lowest distances for suggestions, with the lowest being the typed word.
ildaryet 1 question: I see that CI is changed (M$ launched its own?). But where are CI artifacts? Namely built APKs to install and test?
ildarah! found em': https://github.com/AnySoftKeyboard/AnySoftKeyboard/runs/336914112
ildarinsane I mean