You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There are different ways of sampling clients. Currently we pass the sampling scheme and rate directly as algorithm arguments. It is better to make client sampler a seperate class and an instance is made and passed to the algorithm as argument instead.
The text was updated successfully, but these errors were encountered:
An alternate solution is to create a ClientManager class and move the _client memory definition there. It would be more interesting if DataManager also supported assigning clients to partition indices (Mapping). The assignment could be defined by a ClientManager class's public API method. Thus, if the number of clients is less than the number of partitions, the selected partitions can be drawn in any predetermined order.
There are different ways of sampling clients. Currently we pass the sampling scheme and rate directly as algorithm arguments. It is better to make client sampler a seperate class and an instance is made and passed to the algorithm as argument instead.
The text was updated successfully, but these errors were encountered: