Configuration Parameters

EPOS has the following configuration parameters in epos.properties file within conf directory:

General parameters:

• dataset: Dataset name to be used. This is the name of the directory located inside datasets directory that contains all data.
• numIterations: Number of iterations. Default is 40.
• numChildren: Number of children of each inner node. Default is 2.
• numAgents: Number of participating agents. Default is 100.
• numPlans: The maximum number of possible plans per agent. Lower than this is possible. If more exist in the file, only first numPlans rows are read. Default is 16.
• planDim: Dimension of every possible plan of every agent. Must be equal across the agents and must correspond to the dataset.
• logLevel: The log level. Options are (case-sensitive): SEVERE, ALL, INFO, WARNING, FINE, FINER, FINEST. The output is printed on the console.

Repetitions and randomness parameters:

• numSim: Number of repetitions, each with different relative position of the agents in the tree. Default is 1.
• shuffle: The number of agent-shuffling to make before assigning them to the tree hierarchy. More info on Tree Topology. Default is 0.
• shuffleFile: The path to a file containing already shuffled agents in one column, no header. Default is null.
• reorganizationSeed: The seed on which the Random generator for shuffling is based on. Default is 0. For more information, see Tree Reorganization.

Parameters controlling the Objectives:

• alpha: Weight of unfairness in multi-objective function. 0<= alpha <= 1. Default is 0. For more information, see Multi-Objective Optimization.
• beta: Weight of local cost in multi-objective function. 0<= beta <= 1. Default is 0. For more information, see Multi-Objective Optimization.
• globalCostFunction: The global cost function to be used. Options are (case-sensitive): VAR for Variance, XCORR for Cross-Correlation, RSS for Residual Sum of Squares and RMSE for Root Mean Square Error. Default is VAR. XCORR uses Standard Normalization by definition, RMSE has its own way of scaling and RSS uses standard normalization by default, but can use others. For more information, see Global Cost Function.
• scaling: The scaling technique to be used with RSS function, ignored otherwise. Options are (case-sensitive): STD for Standard Normalization, MIN-MAX for Min-Max Scaling and UNIT-LENGTH for Unit-Length Scaling. Default is STD. For more information, see Scaling Techniques.
• goalSignalPath: The reference signal, paired with RSS, XCORR or RMSE global cost function, otherwise ignored. A path to a file with the signal in one column. The length of the signal from the file must be of size planDim. If shorter, it will be padded in a Round-Robyn manner, if longer, it will be cropped to the size given in planDim. For more information, see Incentive Signals.
• localCostFunction: The local cost function. Options are (case-sensitive): DISC for Discomfort-based Local Cost Function, PREF for Preference-based Local Cost Function, which is converted to DISC by 1 - PREF, INDEX for plan indices to be used as costs (see Index-based Local Cost Function). Default is DISC. For more information, see Local Cost Function.

Dynamic and on-the-fly reorganization parameters:

• strategy: The strategy for triggering reorganizations on-the-fly. The options include: never to disable reorganizations, periodically for Reorganization Criterion: Periodically, convergence for Reorganization Criterion: Convergence and globalCostReduction for Reorganization Criterion: Global Cost Reduction.
• periodically.reorganizationPeriod: Sets the period of the reorganizations when Reorganization Criterion: Periodically strategy is used.
• convergence.memorizationOffset: Sets the memorization offset when Reorganization Criterion: Convergence strategy is used.
• globalCost.reductionThreshold: Sets the threshold when Reorganization Criterion: Global Cost Reduction strategy is used.
• strategy.reorganizationSeed: The random generator seed controlling the randomness in reorganizations. Default is 0.