Fit a Hawkes process or HawkesN process model on one or many event cascades and learn model parameters.

Fit a Hawkes process or HawkesN process model on one or many event cascades and learn model parameters.

fit_series(
  data,
  model_type,
  cores = 1,
  init_pars = NULL,
  .init_no = NULL,
  observation_time = NULL,
  lower_bound = NULL,
  upper_bound = NULL,
  model_vars = NULL,
  parallel_type = "PSOCK",
  ...
)

Arguments

data	A list of data.frame(s) where each data.frame is an event cascade with event tims and event magnitudes (optional)
model_type	A string representing the model type, e.g. EXP for Hawkes processes with an exponential kernel function
cores	The number of cores used for parallel fitting, defaults to 1 (non-parallel)
init_pars	A data.frame of initial parameters passed to the fitting program. Parameters should be aligned with required ones for the corresponding "model_type". The default initial parameters will be used if not provided.
.init_no	If initi_pars is not provided, currently 10 random starting parameters are generated for fitting. This controls which random points are used. Defaults to NULL
observation_time	The event cascades observation time(s). This can either be a single number indicating a common observation time for all cascades or a vector of observation times which has the same length as the number of cascades.
lower_bound	Model parameter lower bounds. A named vector where names are model parameters and values are the lowest possible values.
upper_bound	Model parameter upper bounds. A named vector where names are model parameters and values are the largest possible values.
model_vars	A named list of extra variables provided to hawkes objects
parallel_type	One of "PSOCK" or "FORK". Default to "PSOCK". See "Details" in makeCluster parallel.
...	Further arguments passed to ampl

Value

A model object where the [par] is fitted on [data]. [convergence] indicates the fitting convergence status and [value] is the negative log-likelihood value of the fitted model on [data].

Examples

if (FALSE) {
data <- generate_series(model_type = 'EXP',
                                     par = c(K = 0.9, theta = 1),
                                     sim_no = 10, Tmax = Inf)
fitted <- fit_series(data, 'EXP', observation_time = Inf)
fitted$par # fitted parameters
fitted$convergence # convergence status
fitted$value # negative log-likelihood value
}