Mine networks for establishing the link between genome and g-matrix. The output from this function is required to run individual-based simulations in the rest of the package. The key input to this function, 'gmatrix', is a (square) covariance matrix, with each row and column representing a trait for the individual-based model. This function will run an evolutionary algorithm to try to find a network that produces traits with the covariance structure of gmatrix from a set of random standard normal values. The network from loci values to trait values goes through a number of linked nodes to achieve this, and each generation tests the stress of the resulting network in terms of expected squared deviation of trait covariances from the input gmatrix. Simulations can take minutes to hours or longer, depending on parameters chosen and the number of traits. See vignettes for a more comprehensive explanation for what this function is doing.
mine_gmatrix(
loci = 18,
layers = 6,
indivs = 1000,
npsize = 2000,
mu_pr = 0.05,
mu_sd = 0.01,
max_gen = 1000,
pr_cross = 0.05,
sampleK = 40,
chooseK = 4,
term_cri = -5.3,
sd_ini = 0.1,
use_cor = FALSE,
prnt_out = TRUE,
gmatrix
)The number of loci for an individual. Simulations can allow for both haploid and diploid individuals. Allele values at each loci affect trait values through a network of intermediary nodes.
The number of hidden layers in the network linking loci to traits.
The number of individuals initialised in each generation of the evolutionary algorithm to test among-individual trait correlations. Individuals are initialised with allele values drawn from a standard normal distribution.
The size of the population of networks in each generation of the evolutionary algorithm. Each network is a discrete individual in the population.
The probability that a value in the network will mutate in a generation. Mutation events change the existing value by adding a new value drawn from a normal distribution with a mean of 0 and standard deviation of mu_sd.
The standard deviation of the random normal value mean centered at 0 that is added to the existing value of the network when a mutation event occurs.
The maximum number of generations that the evolutionary algorithm is allowed to run before terminating (regardless of how well the evolved covariance structure matches the pre-specified gmatrix).
The probability that a focal network in the population will initiate a crossover of a subset of its values with a randomly selected second network (note that any given network might therefore be included in more than one crossover event in a generation). The size of the subset is determined randomly.
During a round of selection, the number of random networks chosen to compete in a tournament. A single generation will include as many tournaments as necessary to create a new network population of size npsize.
During a round of selection tournament, the number of networks within the sampleK random subset of the tournament that have the highest fitness will be selected to populate the next generation of networks
The criteria for terminating the evolutionary algorithm. The algorithm will terminate if a network is found in which the mean squared deviation of the covariance matrix elements from gmatrix is less than exp(term_crit).
The standard deviation of initialised network values at the start of the evolutionary algorithm. All network values are initialised by randomly sampling from a normal distribution with a mean of 0 and a standard deviation of sd_ini
Should the gmatrix be treated as a correlation matrix rather than a covariance matrix when calculating fitness?
Should the function print out progress showing the stress for each generation
The pre-specified trait covariance matrix. This will define what the covariance will be between each trait when allele values are drawn from a standard normal distribution.
A list of eight elements that includes the following: (1) A vector of input parameters, (2) the pre-specified covariance matrix, (3) matrix defining the effects of loci values on the first layer of the network, (4) a three dimensional array link the first network layer to trait values, (5) a matrix of the marginal effect of each locus on each trait, (6) the mined covariance structure, (7) all network values to be inserted into individual genomes, and (8) the log stress of the mined matrix against the pre-specified matrix.