This follows up on a long-standing discussion from the machine learning working group: a desire for some way to attach properties to an ML graph. The various ML backends implementing this have mechanisms for tweaking a model's behavior based on properties; this string-based approach seems the most flexible way to access this.
This follows up on a long-standing discussion from the machine learning working group: a desire for some way to attach properties to an ML
graph
. The various ML backends implementing this have mechanisms for tweaking a model's behavior based on properties; thisstring
-based approach seems the most flexible way to access this.