I trained some models that spit out predictions based on the input of [CMDeviceMotion] . These were done (the data collected) in portrait mode. My idea is to just transform them i.e. data points of the mlmodelc (90º / -90º, another variable, sigh) and make it work for landscape mode. Has anyone done this with good results? Are there any gotchas here?
I reckon I will have to transform the raw data and then generate a new model . . . .