Logging & Output¶

The logging functions in EPOS are extensive and they track, record, and report information about the system at each iteration and also at the end / termination of the simulation. The class implementations can be found at agent.logging package in the code. In order to initialise and add different loggers to the agent, they need to be added in the experiment.MultiObjectiveIEPOSExperiment class:

LoggingProvider<MultiObjectiveIEPOSAgent<Vector>> loggingProvider = new LoggingProvider<MultiObjectiveIEPOSAgent<Vector>>();
GlobalCostLogger<Vector>              GCLogger  = new GlobalCostLogger<Vector>(Configuration.getGlobalCostPath());
LocalCostMultiObjectiveLogger<Vector> LCLogger  = new LocalCostMultiObjectiveLogger<Vector>(Configuration.getLocalCostPath());
TerminationLogger<Vector>             TLogger   = new TerminationLogger<Vector>(Configuration.getTerminationPath());
SelectedPlanLogger<Vector>            SPLogger  = new SelectedPlanLogger<Vector>(Configuration.getSelectedPlansPath(), config.numAgents);
GlobalResponseVectorLogger<Vector>    GRVLogger = new GlobalResponseVectorLogger<Vector>(Configuration.getGlobalResponsePath());
PlanFrequencyLogger<Vector>           DstLogger = new PlanFrequencyLogger<Vector>(Configuration.getDistributionPath());
UnfairnessLogger<Vector>              ULogger   = new UnfairnessLogger<Vector>(Configuration.getUnfairnessPath());
GlobalComplexCostLogger<Vector>       GCXLogger = new GlobalComplexCostLogger<Vector>(Configuration.getGlobalComplexCostPath());
WeightsLogger<Vector>                 WLogger   = new WeightsLogger<Vector>(Configuration.getWeightsPath());
ReorganizationLogger<Vector>          RLogger   = new ReorganizationLogger<Vector>(Configuration.getReorganizationPath());

GCLogger.setRun(Configuration.permutationID);
LCLogger.setRun(Configuration.permutationID);
TLogger.setRun(Configuration.permutationID);
SPLogger.setRun(Configuration.permutationID);
GRVLogger.setRun(Configuration.permutationID);
DstLogger.setRun(Configuration.permutationID);
ULogger.setRun(Configuration.permutationID);
GCXLogger.setRun(Configuration.permutationID);
WLogger.setRun(Configuration.permutationID);
RLogger.setRun(Configuration.permutationID);


loggingProvider.add(GCLogger);
loggingProvider.add(LCLogger);
loggingProvider.add(TLogger);
loggingProvider.add(SPLogger);
loggingProvider.add(GRVLogger);
loggingProvider.add(DstLogger);
loggingProvider.add(ULogger);
loggingProvider.add(GCXLogger);
loggingProvider.add(WLogger);
loggingProvider.add(RLogger);

In order to add / develop new loggers to EPOS, the GlobalCostLogger template is recommended. In addition to this, each logger, when their print function is called, they write their data on output files with address defined in config.Configuration class.

Global Complex Cost Logger¶

The global complex cost logger is implemented in agent.logging.GlobalComplexCostLogger.java and by default, it record the complex cost defined in Multi-Objective Optimization. The output file is named global-complex-cost.csv and can be found in this path: ~/output/. The .csv file is an matrix showing complex cost of the whole system, at each iteration, and across multiple runs. In addition, it shows the mean and standard deviation of the cost across different runs. By default, the complex cost (Multi-Objective Optimization) is defined as:

\[\gamma \cdot G^{(t)} + \alpha \cdot U^{(t)} + \beta \cdot L^{(t)}\]

Global Cost Logger¶

The global cost logger is implemented in agent.logging.GlobalCostLogger.java and by default, it records the global cost defined in Multi-Objective Optimization. The output file is named global-cost.csv and can be found in this path: ~/output/. The .csv file is a matrix showing the global cost of the whole system, at each iteration, and across multiple runs. In addition, it shows the mean and the standard deviation of the global cost across different runs. If no specific global cost logger is defined at the config.Configuration then EPOS assigns GlobalComplexCostLogger<Vector> GCXLogger as the global cost logger.

Global Response Logger¶

The global response logger is implemented in agent.logging.GlobalResponseCostLogger.java and by default, it records the global response defined in Global Cost Function. The output file is named global-response-cost.csv and can be found in this path: ~/output/. The .csv file is a matrix showing global response of the system, i.e. the sum of all selected plans, for every EPOS repetition and in every iteration.

Indexed Histogram Logger¶

The index histogram logger is implemented in agent.logging.PlanFrequencyLogger.java and by default, it records the plan frequency defined in Plans. The output file is named indexes-histogram.csv and can be found in this path: ~/output/. The .csv file is an matrix showing how many times each plan was selected at the end of each run / simulation.

Local Cost Logger¶

The local cost logger is implemented in agent.logging.LocalCostLogger.java and by default, it records the local cost defined in Local Cost Function. The output file is named local-cost.csv and can be found in this path: ~/output/. The .csv file is a matrix showing local cost of all agents at each iteration and for every EPOS repetition. In addition, it shows the mean and the standard deviation of the cost across different runs.

Selected Plans¶

The selected plan logger is implemented in agent.logging.SelectedPlanLogger.java and by default, it records the selected plan index defined in Plans. The output file is named selected-plan.csv and can be found in this path: ~/output/. The .csv file is a matrix showing selected plan index of each agent at each iteration and in every repetition.

Termination Logger¶

The termination logger is implemented in agent.logging.TerminationLogger.java and by default, it records the terminal iteration defined in Termination Criteria. The output file is named termination.csv and can be found in this path: ~/output/. The .csv file is a matrix showing at which iteration EPOS terminated depending on the termination criteria.

Unfairness Logger¶

The unfairness logger is implemented in agent.logging.UnfairnessCostLogger.java and by default, it record the unfairness defined in Multi-Objective Optimization. The output file is named unfairness.csv and can be found in this path: ~/output/. The .csv file is a matrix showing the unfairness of the whole system, at each iteration, and across multiple runs. In addition, it shows the mean and standard deviation of the cost across different runs.

Reorganization Logger¶

This logger is implemented in agent.logging.ReorganizationLogger.java. For each EPOS run, the logger creates a new column that contains the cumulative number of reorganizations until the iteration index of the first column. More regarding reorganization and tree topology of EPOS can be found at Tree Reorganization.

Weights, Alpha, Beta Logger¶

The weights logger is implemented in agent.logging.WeightsLogger.java and by default, it records the weights defined in Multi-Objective Optimization. The output file is named weights-alpha-beta.csv and can be found in this path: ~/output/. The .csv file is a matrix showing \(\alpha\), \(\beta\) and \(\gamma = 1 - \alpha - \beta\) for every iteration.