It's neat to see training/evaluating/serving ML models get commoditized (eg Amazon's new service). But the hard part is feature generation.
-
-
(BTW if building infrastructure to automate more of the feature engineering is interesting to you, get in touch.)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@avibryant I would love to hear more about useful ways of converting an m dimensional event stream into a n dimensional feature vector -
@avibryant for example are there any techniques that could find features like "count of past events x where a_x/b_x > 10" -
@snoble I think the first problem is just to make it trivial to *express* features like that, produce training sets, and track the values. -
@avibryant I think that's right. Once you have a grammar you can navigate the space way better -
@avibryant so we have ways to convert m dimensions to scalers (random forests, logistic regression)... -
@avibryant ... and a way to roll up streams of scalers (algebird). Just need a way to optimize combinations
End of conversation
New conversation -
-
-
@avibryant yep: (semi-)automate the damn feature engineering and you start providing real value -
@beaucronin@avibryant Turns out this is not super easy, esp in real-time with potential drift -
@mgershoff@avibryant and thus, the value - End of conversation
New conversation -
-
-
@avibryant We'll all be replaced by a deep learning network soon enough.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@avibryant I'd think making it cost effective is the hard part.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.