Action fragmentation now uses the network induced by feasible action pairs automatically. Native C++ single-objective construction, multi-objective coefficient vectors and reported values use the same domain.
Multi-objective solves upscale constraint rows whose maximum absolute coefficient is below one, using a positive factor capped at 1e6. The same factor multiplies both the row and its right-hand side in the solver copy; objective units, compiled matrices and postsolve feasibility checks are unchanged.
Fixed multi-objective cost evaluation after locked-out actions are removed from the model. Reported costs and automatic epsilon bounds now use the decoded selections from the solved model.
Epsilon-constraint now automatically reuses successful solutions as warm starts during anchor calculations and subsequent runs with CBC and Gurobi. No new public argument is required. Automatic grids visit secondary bounds from most restrictive to most permissive; manual run order is preserved.
selection_frequency() without changing recurrence counts, and
use pre-rendered maps in the large forest-restoration vignette to reduce
package build and check times.load_meseta() and load_ecosystem_services() alongside
load_sim_multiaction() as the three complete example-data entry points.
Updated the README and integrated-planning vignette; historical data()
and raster-only access remain unchanged for compatibility.Added add_objective_max_effect() and add_objective_min_effect() for signed
changes relative to the reference, including action interactions. Their
arguments are x, actions, features, and alias.
add_objective_max_benefit() is deprecated with lifecycle and delegates to
add_objective_max_effect(), retaining the development branch's signed-net
behavior. The published positive-only behavior changed earlier in this branch.
add_objective_min_loss() is deprecated but retains its original final
negative-part criterion. min_effect() is not an equivalent replacement:
it minimizes signed change and permits compensation. No removal date is set.
Added an educational Meseta Ibérica action-planning tutorial, bundled inputs, and a runnable script covering spatial-cost trade-offs and spatial analyses.
Improved prototype/data-cloning memory use and indexed planning-unit analyses.
Bounded polygon/raster extraction buffers to avoid eager full-raster allocation failures when building large landscape examples.
Fixed single-objective min_cost action scopes so they agree with the MO
cost criterion; planning-unit costs remain global and feasible actions retain
their original indices.
Fixed CBC crashes/short solution vectors caused by trailing empty matrix rows or columns, including fully locked legacy impact models. A redundant bound row preserves dimensions without changing the feasible decisions.
MO solves use bounded positive scalar-objective rescaling, with values returned in their original units. CBC and Gurobi receive precise numerical defaults; explicit solver parameters take precedence. Epsilon bounds are checked against reevaluated selected decisions and checks/effective settings are recorded in solution diagnostics. Binary integrality noise no longer enters objective evaluation or automatic epsilon endpoints.
After an optimal AUGMECON solve, refine normalized slacks at the attained primary level within the remaining time limit. The original augmented criterion and global status/gap are preserved and refinement is recorded.
Native model allocation reserves memory according to the actual core size rather than one million rows/columns for every temporary MO model.
Deprecated add_objective_min_intervention_impact() with lifecycle. It keeps
its original behavior and arguments until removal in a future release. New
workflows express action consequences through effects and net benefit.
The forest-restoration example now uses deficit effects with a common ceiling
per feature and fixed restoration effort, documenting the equivalence and its
limits. Cached plans are migrated by exact affine evaluation of their decisions.
Changed behavior: add_objective_max_benefit() now maximizes signed net
change relative to the reference, including negative individual effects and
interaction corrections. Earlier versions counted positive effects only;
models with negative effects may select different plans. Names and public
arguments are unchanged; no separate net-benefit function is introduced.
add_objective_min_loss() retains its loss-only criterion, with the negative
part computed after combining actions within each planning unit and feature.
Exact loss auxiliaries support mixed signs, overlapping scopes, zero weights,
secondary bounds, and payoff/lexicographic solves in all three MO methods.
Concurrent individual effects are additive unless joint totals are supplied.
Ecological targets and result summaries count the reference once per selected
unit in their action scope, using signed joint changes for final outcomes.
selected_benefit remains the positive-gain metric; selected_net matches
the signed benefit objective when their scopes coincide.
Modern add_effects() tables and rasters accept registered action sets as
total joint outcomes, effects, or relative changes. Original totals and input
tables are preserved separately from sparse signed subset corrections.
Missing interactions/individual effects are explicitly assumed zero.
Exact continuous AND auxiliaries are shared across features, created only
for feasible non-zero corrections, and integrated into single-objective and
all three MO formulations. Inferred negative outcomes are excluded.
Joint solution summaries aggregate signed changes per PU/feature and count
the selected reference once. Signed ecological objectives and concurrent
targets use the final aggregation described above.
Added add_constraint_action_requires() (all/any companions),
add_constraint_action_excludes() (at most one group member), and
add_constraint_action_together() (all or none), scoped separately by PU.
Distinct rules accumulate; duplicate identities and names are errors.
Unavailable actions are treated as zero, including missing pairs, exclusions,
and filtered non-finite costs. Relations use the same compiled feasible set
for single-objective, weighted-sum, epsilon-constraint, and AUGMECON methods.
They do not lift the default action-count limit or introduce joint effects.
Defined a strict repeated-call contract without adding public arguments. Actions, effects (including legacy wrappers), profit, solver settings, MO methods, and unaliased objectives can be defined only once per problem. Second definitions raise an error instead of replacing data or settings. Distinct sets, constraints, spatial names, and objective aliases accumulate. Locks accumulate compatible states and reject inversions in either call order. Targets reject duplicate feature/action scopes immediately, including across absolute and relative calls. Budget identity includes cost components and area identity includes its measure. Compare configurations by deriving each alternative from a common problem before its first definition.
Added add_constraint_action_cardinality() for per-unit minimum, maximum,
or exact action counts. Repeated calls support distinct spatial and action
scopes; explicit total maxima/equalities replace the implicit one-action
limit only in their covered units. Rules are compiled for single-objective
and multi-objective methods, including concurrent economic and ecological
workflows with final feature aggregation.
Added add_action_sets() for named lists or long membership tables and
get_action_sets() for inspecting definitions. Actions can belong to several
sets. Definitions are independent of action feasibility, selection rules,
costs, and effects; registering them does not enable simultaneous actions.
Action-set registration validates members and identifier conflicts, adds new sets across calls, and preserves the original problem. Action catalogs are fixed after their first definition, protecting dependent sets and indices.
add_effects() now accepts exactly one of effect, outcome, or
relative_change. Tables without pu expand over feasible action pairs.reference_amount, action_outcome, and signed effect,
while retaining amount_after, benefit, and loss for compatibility.raster_aggregation and raster_type for spatial effect inputs.add_effects() and removed legacy syntax from implicit reserve-model setup.add_spatial_relations() now supports allow_self = TRUE together with directed = TRUE. Diagonal entries (i, i) in directed relations are interpreted as unary planning-unit terms rather than as directed self-dependencies.frontier_neighbors(), linkage_distances(), linkage_transition(),
linkage_turnover(), and linkage_contrasts() for objective--decision
linkage analysis.get_solution_states() and selection_consistency() for summarizing
spatial decisions across solutions.gap_limit precision.load_sim_multiaction() and its associated simulated planning units, features, action costs, and action-specific effects for examples, tests, and introductory documentation.restoration action, while retaining fixed conservation commitments as spatial context.SolutionSet to support reproducible vignette and pkgdown builds without requiring Gurobi or rerunning the full optimization.solution_id is shown as the first column.run_id from user-facing solution extractors and frontier outputs; it is now kept for get_runs() and internal matching only.internal_id from planning-unit outputs.get_targets() so it consistently returns solution_id instead of run_id.frontier_extremes() and frontier_distances() to follow the same solution_id-based output convention.get_runs().Renamed planning-unit functions and arguments to use planning_units consistently instead of pu.
Deprecated older function aliases kept for backwards compatibility:
get_pu() in favour of get_planning_units();plot_spatial_pu() in favour of plot_spatial_planning_units();add_constraint_locked_pu() in favour of the planning-unit naming convention;add_objective_min_fragmentation_pu() in favour of add_objective_min_fragmentation_planning_units().Deprecated older multi-objective run helpers:
run_grid() in favour of set_runs_grid();run_manual() in favour of set_runs_manual();mo_control() in favour of set_runs_control().solution_id as the main identifier for stored solutions.run_id primarily for get_runs() and internal diagnostics, reducing ambiguity in user-facing summary outputs.get_objectives() to return cleaner user-facing tables while preserving run-level identifiers internally for frontier and dominance calculations.solution/solutions terminology consistently.normalize_weights = FALSE.feasible_only handling from objective extraction workflows.solution_id-based public API.run terminology across getters, plotting, and frontier utilities.solve() consistently returns a SolutionSet, including for single-objective problems.Solution class from the public API and documentation. Individual run-level solutions remain available only as internal components of a SolutionSet.solution_id identifiers to distinguish stored solutions from attempted runs identified by run_id.run_id and solution_id consistently across extraction and analysis functions.SolutionSet printing and documentation to reflect the distinction between run attempts and stored solutions.dist_features.amount_after to represent the feature amount obtained when a planning unit is conserved.run_grid() and run_manual().mo_control().SolutionSet run table.solution_id only to runs that produced a stored solution.get_runs() to extract run-level status, runtime, gap, design parameters, solution identifiers, and objective values.get_objectives() to extract objective values in long or wide format.get_objective_specs() to extract registered objective aliases, objective types, model types, optimization senses, and creation metadata.run_id and solution_id.SolutionSet architecture.solution_filter() to return a coherently filtered SolutionSet.solution_filter() can filter by run_id, solution_id, solver status, or feasibility.moocore.SolutionSet objects before modification to prevent reference-based mutation of the original object.solution_append() to combine compatible SolutionSet objects generated from the same planning problem.solution_append() verifies compatibility of planning units, features, actions, effects, targets, constraints, locks, spatial relations, and objective definitions.run_id and solution_id values.solution_unique() to retain one representative from groups of equivalent solutions.solution_unique() can identify repeated solutions using either complete decision vectors or objective values.frontier_extremes() to identify the observed minimum and maximum values of each objective.best or worst according to each objective's optimization sense.frontier_distances() to calculate normalized distances to observed ideal and nadir points.SolutionSet.selection_frequency() to calculate how frequently each planning-unit/action assignment is selected across stored solutions.conservation action.selection_similarity() to quantify structural similarity among solutions.Solution class.SolutionSet terminology.run_id, solution_id, objective extraction, filtering, frontier analysis, and selection analysis.moocore as an optional dependency for non-dominance filtering.multiscape.Problem, Solution, and SolutionSet classes. The public result architecture was later unified around SolutionSet in
version 1.1.0.create_problem(), add_*(), set_*(), and solve().