halo
halo copied to clipboard
Running Experiments for ClientGroups
A good starting point: choose a non-empty random % subset of the clients in a group and treat them identically (i.e., all of their profiling data would be treated to as being from the same 'client' to the model).