.. _settings-chapter: ======== Settings ======== Dataset Source ============== Argonaut provides built-in datasets and a custom upload mode. The tested deployment exposed the following dataset choices: * ``-- Custom Upload Plans --`` * ``Bike Sharing`` * ``Drone Surveillance`` * ``Household Electricity`` * ``Gaussian [Subset]`` * ``Gaussian [Full]`` * ``IoT Data Sharing`` Custom Upload ------------- When ``-- Custom Upload Plans --`` is selected, Argonaut shows an Agent Plan Files area: .. code-block:: text Drop .plans files here or click to browse The file input accepts multiple files. The Run button remains disabled until plan files or a built-in dataset are loaded. Built-In Dataset ---------------- When a built-in dataset is selected, Argonaut shows a loaded dataset confirmation and then exposes agent, plan, and dimension selection. Agent Selection =============== ``Number of agents (N)`` controls how many agents are included in the decision space. Bulk mode Selects the first :math:`N` agents automatically. Subset mode Lets users choose exactly :math:`N` agents manually through checkboxes. The selected count is shown, and ``Done`` is disabled until exactly :math:`N` agents are selected. Plan Selection ============== ``Plans per agent (M)`` controls how many feasible plans are retained for each selected agent. Bulk mode Selects the first :math:`M` plans for each selected agent. Subset mode Lets users manually choose plan subsets when the section is unlocked. In the tested Gaussian subset, each selected agent had up to 16 plans, and the test used the first 2 plans. Dimension Selection =================== ``Number of dimensions (D)`` controls how many plan-vector dimensions are retained. Bulk mode Selects the first :math:`D` dimensions of each plan. Subset mode Lets users manually choose dimensions when the section is unlocked. In the tested Gaussian subset, plans had up to 100 dimensions, and the test used the first 10 dimensions. .. figure:: _static/img/manual-selection.png :alt: Manual agent subset selection :align: center Manual subset mode exposes individual agent/plan/dimension checkboxes and validates the selected count. Plan Editor =========== After agents, plans, and dimensions are locked, Argonaut displays a Plan Editor. It shows selected agents as tabs and displays plan rows with local cost and dimensions as editable numeric cells. The editor can be used to inspect or modify plan values before running an algorithm. .. figure:: _static/img/plan-editor.png :alt: plan editor :align: center Plan editor shows and edits the plan value of each agent. Algorithm Parameters ==================== ``Iterations`` Number of algorithm iterations for Tree-Based and combined modes. The tested run used ``40``. ``Children per node`` Branching factor for the tree topology. ``2`` produces the binary tree layout used in the tested run. ``Simulations`` Number of repeated simulations. The tested run used ``1``. .. figure:: _static/img/parameters.png :alt: Parameters settings in algorithm :align: center Set algorithm parameters. Weighted Total Cost =================== Argonaut represents the total optimization cost as: .. math:: C(s) = (1 - \alpha - \beta) \cdot C_{global}(s) + \alpha \cdot C_{unfairness}(s) + \beta \cdot C_{local}(s) where: and :math:`\alpha, \beta \geq 0`. The tradeoff parameters :math:`\alpha`, :math:`\beta`, and :math:`\gamma` steer the balance between fairness across agents, local discomfort, and global optimality. A higher weight expresses stronger preference toward minimizing that objective. When a weight is 0, the corresponding objective is not optimized. .. figure:: _static/img/objectives.png :alt: Weight adjustement for multi-objectives :align: center Tune weights on different objectives. Cost Functions ============== Global cost functions --------------------- ``VAR - Variance`` Minimizes variance in the global response. It is useful when the goal is to make the aggregate vector more uniform or flat. ``RSS - Residual Sum of Squares`` Measures the residual between the global response and a goal signal. Lower values indicate a better match. ``RMSE - Root Mean Square Error`` Measures dissimilarity between the EPOS response and a scaled incentive signal. ``XCORR - Cross-Correlation`` Measures shape matching through negative cross-correlation. Because EPOS is a minimization algorithm, it minimizes negative cross-correlation. Local cost functions -------------------- ``INDEX - Plan index`` Uses the selected plan index as the local cost signal. ``DISC - Discomfort score`` Uses the plan score as discomfort. ``PREF - Preference`` Uses the plan score as preference. .. figure:: _static/img/cost-functions.png :alt: cost-functions :align: center Choose cost functions in both global cost and local cost.