DataTaunew | comments | leaders | submitlogin
Show DataTau: Upcoming Feature Engineering as a Service tool for mobile apps
3 points by andrewxhill 19 days ago | 7 comments
Hi DataTau!

We are about (couple weeks) to launch a new service for automated feature extraction for data coming from mobile apps. The service works by running feature extraction directly on device and providing a the full stream of high quality features for you to model, mine, etc.

We are all data scientists and engineers, not marketers, so would love your feedback on the service. Anyone out there that would use such a service? I'd love to chat with you if you'd be up for it and pick your brain.

Keep in mind the web design, text etc is prelim and will be getting polished up over the next weeks. Still, I'd love to know if the points resonate or if there are other areas you think we should expand on more. Let me know!

https://www.textile.io/

Thanks!





3 points by guangningyu 18 days ago | link

What kind of feature can you provide? device level or app level?

reply

2 points by andrewxhill 18 days ago | link

We are building out features that fall into,

user features: e.g. 'at home'

basic app features: e.g. 'daily active app user' or 'is active'

custom app features: e.g. 'item in cart'

device features: e.g. lots of regional, temporal, and connectivity breakdowns

The above highlight a number of the semantic features, there are also a host of temporal characteristics based on long and short term time-series.

finally we are working on profiles/segmentation, both general and real-time. e.g. '9-5er' 'workaholic' etc.

reply

2 points by croatiankp 15 days ago | link

Have you considered overlaying demographic information and regional weather info? Device connection speed? etc

reply

1 point by andrewxhill 15 days ago | link

Totally! One of my previous projects was helping build out the Data Observatory at Carto (https://carto.com/data-observatory/, https://cartodb.github.io/bigmetadata/) so for sure this has been on my mind. Our focus right now is exposing a lot of the real-time/rolling segmentation (e.g. currently is daily active [driver, app user, buyer], normally a weekly active [driver, app user, buyer]) that is possible, but soon mixing in some of those external variables.

Speed and connectivity etc are already being used to generate features.

Thanks for the q!

reply

2 points by andrewxhill 19 days ago | link

https://www.textile.io/ <- clickable link

reply

1 point by agawronski 18 days ago | link

What is the benefit of doing feature engineering on device?

reply

1 point by andrewxhill 18 days ago | link

The two biggest benefits are that we can build very nuanced data models at the individual level that support many of our features being extracted. But the primary motivation was to provide features that could be modeled offline by data scientists and models produced could be put back into production, on the mobile device.

That said, there are features that will require server side, and those will come down the road.

reply




RSS | Announcements