From reading the post I understand that they're taking down the launcher, or that it otherwise won't work anymore. And I have to ask - why? Why just not leave it there? Is there something in it that requires Internet connection to access their servers? It's a launcher, even predictive smart whatever features should not require Internet access.
The prediction is likely entirely centralized machine learning; similar to how Netflix predictions work. Tell the server <x apps installed, y location, z time of day>, get a response showing <show apps a, b, q>.
This is exactly the kind of data that should not fly over the wire unless it's absolutely necessary, which in this case I believe it isn't.
I can't imagine what kind of machine learning they'd have to be using to make it not work on a phone. It doesn't take much computing power to do a decent predictor. I'm going to assume process laziness here - being used to the idea that if everything is running on your server, you can tweak stuff there and have it immediately working on everyone's (Internet-connected) endpoints. It makes sense for websites, but IMO it's a wrong approach for devices.
Big data learning. They're (probably) inferring how you do stuff based also on how other people do. In fact, there's no way to recommend apps not already installed on your phone without consulting a server.
I think I misunderstood their app's description. I thought it was about recommending things to launch out of the things you have installed.
EDIT: And I'd pay for a launcher that learns from my interactions with it off-line, and recommends me apps based on context such as location, time of day, previously launched apps, etc. Such a thing does not need "big data learning". It's an undergrad-level machine learning exercise.
> It doesn't take much computing power to do a decent predictor
Collaborative filtering, a standard recommendation method, requires a great deal of computing power. Depending on the feature engineering, this could result in "big data" (whatever that means) even considering only one users' activity in isolation.
