createAlgorithm function

createAlgorithm

Specifies the algorithm used in the estimation based on characteristics of the state and the effects.

createAlgorithm(
  state,
  effects,
  multinomialProposal = FALSE,
  burnInN1 = NULL,
  thinningN1 = NULL,
  iterationsN1 = NULL,
  gainN1 = 0.1,
  burnInN2 = NULL,
  thinningN2 = NULL,
  initialIterationsN2 = 50,
  nsubN2 = 4,
  initGain = 0.6,
  burnInN3 = NULL,
  thinningN3 = NULL,
  iterationsN3 = 500,
  allowLoops = NULL
)

monanAlgorithmCreate(
  state,
  effects,
  multinomialProposal = FALSE,
  burnInN1 = NULL,
  thinningN1 = NULL,
  iterationsN1 = NULL,
  gainN1 = 0.1,
  burnInN2 = NULL,
  thinningN2 = NULL,
  initialIterationsN2 = 50,
  nsubN2 = 4,
  initGain = 0.6,
  burnInN3 = NULL,
  thinningN3 = NULL,
  iterationsN3 = 500,
  allowLoops = NULL
)

Arguments

• state: An object of class "processState.monan" that contains all relevant information about the outcome in the form of an edgelist, the nodesets, and covariates.
• effects: An object of class "effectsList.monan" that specifies the model.
• multinomialProposal: How should the next possible outcome in the simulation chains be sampled? If TRUE, fewer simulation steps are needed, but each simulation step takes considerably longer. Defaults to FALSE.
• burnInN1: The number of simulation steps before the first draw in Phase 1. A recommended value is at least n_Individuals * n_locations if multinomialProposal = FALSE, and at least n_Individuals if multinomialProposal = TRUE which is set as default.
• thinningN1: The number of simulation steps between two draws in Phase 1. A recommended value is at least 0.5 * n_Individuals * n_locations if multinomialProposal = FALSE, and at least n_Individuals if multinomialProposal = TRUE which is set as default.
• iterationsN1: The number of draws taken in Phase 1. A recommended value is at least 4 * n_effects which is set as default. If the value is too low, there will be an error in Phase 1.
• gainN1: The size of the updating step after Phase 1. A conservative value is 0, values higher than 0.25 are courageous. Defaults to 0.1.
• burnInN2: The number of simulation steps before the first draw in Phase 1. A recommended value is at least n_Individuals * n_locations if multinomialProposal = FALSE, and at least n_Individuals if multinomialProposal = TRUE which is set as default.
• thinningN2: The number of simulation steps between two draws in Phase 2. A recommended value is at least 0.5 * n_Individuals * n_locations if multinomialProposal = FALSE, and at least n_Individuals if multinomialProposal = TRUE which is set as default.
• initialIterationsN2: The number of draws taken in subphase 1 of Phase 2. For first estimations, a recommended value is around 50 (default to 50). Note that in later subphases, the number of iterations increases. If this is a further estimation to improve convergence, higher values (100+) are recommended.
• nsubN2: Number of subphases in Phase 2. In case this is the first estimation, 4 subphases are recommended and set as default. If convergence in a previous estimation was close, then 1-2 subphases should be enough.
• initGain: The magnitude of parameter updates in the first subphase of Phase 2. Values of around 0.2 (default) are recommended.
• burnInN3: The number of simulation steps before the first draw in Phase 3. A recommended value is at least 3 * n_Individuals * n_locations if multinomialProposal = FALSE, and at least 3 * n_Individuals if multinomialProposal = TRUE which is set as default.
• thinningN3: The number of simulation steps between two draws in Phase 3. A recommended value is at least n_Individuals * n_locations if multinomialProposal = FALSE, and at least 2 * n_Individuals if multinomialProposal = TRUE which is set as default. In case this value is too low, the outcome might erroneously indicate a lack of convergence.
• iterationsN3: Number of draws in Phase 3. Recommended are at the very least 500 (default). In case this value is too low, the outcome might erroneously indicate a lack of convergence.
• allowLoops: Logical: can individuals/resources stay in their origin?

Returns

An object of class "algorithm.monan".

Examples

# define algorithm based on state and effects characteristics
myAlg <- createAlgorithm(myState, myEffects, multinomialProposal = FALSE)

See Also

createProcessState(), createEffectsObject(), estimateMobilityNetwork()