Class RoundMetadata
Contains detailed metrics for a single federated learning round.
public class RoundMetadata
- Inheritance
-
RoundMetadata
- Inherited Members
Remarks
For Beginners: Information about what happened in one specific training round.
Properties
AverageLocalLoss
Gets or sets the average local loss across selected clients.
public double AverageLocalLoss { get; set; }
Property Value
Remarks
For Beginners: The average loss that clients achieved on their local data before sending updates. Comparing this to global loss can reveal overfitting.
CommunicationMB
Gets or sets the communication cost for this round in megabytes.
public double CommunicationMB { get; set; }
Property Value
GlobalAccuracy
Gets or sets the global model accuracy after this round.
public double GlobalAccuracy { get; set; }
Property Value
GlobalLoss
Gets or sets the global model loss after this round.
public double GlobalLoss { get; set; }
Property Value
PrivacyBudgetConsumed
Gets or sets the privacy budget consumed in this round.
public double PrivacyBudgetConsumed { get; set; }
Property Value
RoundNumber
Gets or sets the round number (0-indexed).
public int RoundNumber { get; set; }
Property Value
RoundTimeSeconds
Gets or sets the time taken for this round in seconds.
public double RoundTimeSeconds { get; set; }
Property Value
SelectedClientIds
Gets or sets the IDs of clients selected for this round.
public List<int> SelectedClientIds { get; set; }
Property Value
UploadCompressionRatio
Gets or sets the effective upload compression ratio for this round (1.0 means no compression).
public double UploadCompressionRatio { get; set; }
Property Value
Remarks
For Beginners: This is the fraction of upload bandwidth used compared to sending a full update. For example, 0.1 means uploads were about 10% of the uncompressed size.