Skip to contents

An R6 Class object, a collection of model for population of BiSBM networks

bisbmpop.Rd

This object contain all the logic and methods to estimate a colBiSBM model

Public fields

n

A list with two dimensions, each of size M for the rows and cols

A

List of incidence Matrix of size n[[1]][m]xn[[2]][m]

M

Number of networks

mask

List of M masks, indicating NAs in the matrices. 1 for NA, 0 else

distribution

Emission distribution either : "poisson" or "bernoulli"

net_id

A vector containing the "ids" or names of the networks (if none given, they are set to their number in A list)

model_list

A list of size Q1max * Q2max containing the best models

discarded_model_list

A list of size Q1max * Q2max * (nb_models - 1) containing the discarded models.

global_opts

A list of options for the model space exploration. See details for more information on which options are available.

fit_opts

A list of options specifically for fitting the models.

separated_inits

A nested list : Q1 init containing Q2 init with each entry containing list of size M storing the separated inits for the class.

exploration_order_list

A list used to store the path taken in the state space.

Z_init

A list of initializations for the Z memberships.

free_mixture_row

A boolean signaling if there is free mixture on the row blocks

free_mixture_col

A boolean signaling if there is free mixture on the columns blocks

ICL

A matrix of size Q1*Q2 storing the best ICL found for each value of Q1, Q2.

sep_BiSBM

A named list containing attributes for each sep BiSBM.

BICL

A matrix of size Q1*Q2 storing the best BICL found for each value of Q1, Q2.

vbound

A matrix of size Q1*Q2 storing the best vbound found for each value of Q1, Q2.

best_fit

A fitBipartiteSBMPop object changing regularly to store the current best fit.

adjusted_fit

Defining a field in case the user wants to adjust a prefitted bisbmpop.

logfactA

A quantity used with the Poisson probability distribution

improved

A field use at each step to check if it has improved.

moving_window_coordinates

A list of size two containing the coordinates of the bottom left and top right points of the square

old_moving_window_coordinates

A list containing the previous coordinates of the moving window.

joint_modelisation_preferred

A boolean to store the preferred modelisation.

Methods

Public methods

Method new()

Create a new instance of the bisbmpop object

This class is generally called via the user function estimate_colBiSBM

Usage

bisbmpop$new(
  netlist = NULL,
  net_id = NULL,
  distribution = NULL,
  free_mixture_row = FALSE,
  free_mixture_col = FALSE,
  Z_init = NULL,
  global_opts = list(),
  fit_opts = list()
)

Arguments

netlist

The list of M networks

net_id

A list of name for the networks, defaults to 1 to M if not provided

distribution

The emission distribution either "bernoulli" or "poisson"

free_mixture_row

A boolean indicating if there is free mixture on the row blocks

free_mixture_col

A boolean indicating if there is free mixture on the column blocks

Z_init

A bidimensional list providing a clustering of the row and column nodes

global_opts

A list of global options used by the algorithm. See details of the user function for more information.

fit_opts

A list of fit options used by the algorithm. See details of the user function for more information.

Returns

A new 'bisbmpop' object. A method to perform the splitting of the clusters

Method split_clustering()

Usage

bisbmpop$split_clustering(origin_model, axis = "row")

Arguments

origin_model

a model (fitBipartite object) to split from.

axis

a string to indicate if this is a column split or a row split. "row" or "col".

Returns

best of the possible models tested A method to perform the merging of the clusters

Method merge_clustering()

Usage

bisbmpop$merge_clustering(origin_model, axis = "row")

Arguments

origin_model

a model (fitBipartite object) to merge from.

axis

a string to indicate if this is a "row", "col" or "both" merge

Returns

best of the possible models tested Method to greedily explore state of space looking for the mode and storing the models discovered along the way

Method greedy_exploration()

Usage

bisbmpop$greedy_exploration(
  starting_point,
  max_iter = self$fit_opts$greedy_exploration_max_steps,
  max_step_without_improvement =
    self$fit_opts$greedy_exploration_max_steps_without_improvement
)

Arguments

starting_point

A vector of the two coordinates c(Q1,Q2) which are the starting point.

max_iter

The maximum number of iteration of greedy exploration to perform, defaults to 12.

max_step_without_improvement

defaults to 2, the number of steps to try improving before stopping the search.

Returns

c(Q1_mode, Q2_mode) which indicates the Q1 and Q2 for which the BICL was maximal. Optimization from a given Z_init

The method takes no parameters

Method optimize_from_zinit()

Usage

bisbmpop$optimize_from_zinit()

Returns

nothing but stores the models in the model list Burn-in method to start exploring the state of space

The method takes no parameters but modify the object

Method burn_in()

Usage

bisbmpop$burn_in()

Returns

nothing; but stores the values The optimization method

Method optimize()

This method performs the burn in (ie initialization + greedy exploration) and the steps of moving window with cluster splitting and merging around the mode found.

Usage

bisbmpop$optimize()

Returns

nothing; but stores the values The moving window application

Method moving_window()

This method is a moving windows over the Q1xQ2 space for the number of clusters

Usage

bisbmpop$moving_window(center, depth = 1)

Arguments

center

is the coordinates (Q1, Q2) in the model list of the mode

depth

is how far away from the center the method should be applied in a grid style going from center - (depth,depth) to center + (depth, depth)

Returns

nothing; but updates the object by adding new models To truncate the discarder model list

Method truncate_discarded_model_list()

This method remove the worst models regarding the BICL criterion.

Usage

bisbmpop$truncate_discarded_model_list()

Returns

nothing Check if the point is in the limits

Method point_is_in_limits()

Usage

bisbmpop$point_is_in_limits(point)

Arguments

point

A vector of size 2, containing the first coordinate Q1, the number of row blocks and Q2 the number of column blocks.

Returns

A boolean if the point if in the limits of [0,Q1_max]x[0,Q2_max] Store the criteria and best fit

Method store_criteria_and_best_fit()

This method stores the criteria (vbound, ICL, BICL) of the models in model_list attribute and at the end choose the best model using the BICL criterion.

Usage

bisbmpop$store_criteria_and_best_fit()

Returns

nothing; modifies the object Computation of the separated Bi SBM

Method compute_sep_BiSBM_BICL()

This method performs the computation of BiSBM for each of the network in the netlist.

Usage

bisbmpop$compute_sep_BiSBM_BICL()

Returns

nothing; stores the values Method to choose the collection modelisation or a separated one

Method choose_joint_or_separated()

Using the BICL criterion, the best model fitted after the procedure and the separated Bi SBM this method choose the one that maximizes the BICL criterion.

Usage

bisbmpop$choose_joint_or_separated()

Returns

nothing; stores a boolean Print the vbound, the ICL and the BICL

Method print_metrics()

Usage

bisbmpop$print_metrics()

Returns

nothing The message printed when one prints the object

Method show()

Usage

bisbmpop$show(
  type = "Fitted Collection of Bipartite SBM\nwith all fitted models"
)

Arguments

type

The title above the message. The print method

Method print()

Usage

bisbmpop$print()

Returns

nothing; prints to console

Method clone()

The objects of this class are cloneable with this method.

Usage

bisbmpop$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.