bayes_met {ProbBreed} | R Documentation |
Fits a Bayesian multi-environment model using rstan
, the R
interface to Stan
.
bayes_met(
data,
gen,
loc,
repl,
trait,
reg = NULL,
year = NULL,
res.het = FALSE,
iter = 2000,
cores = 2,
chains = 4,
pars = NA,
warmup = floor(iter/2),
thin = 1,
seed = sample.int(.Machine$integer.max, 1),
init = "random",
verbose = FALSE,
algorithm = c("NUTS", "HMC", "Fixed_param"),
control = NULL,
include = TRUE,
show_messages = TRUE,
...
)
data |
A data frame in which to interpret the variables declared in the other arguments. |
gen , loc |
A string. The name of the columns that contain the evaluated candidates and locations (or environments, if you are working with factor combinations), respectively. |
repl |
A string, a vector, or |
trait |
A string. The analysed variable. Currently, only single-trait models are fitted. |
reg |
A string or NULL. The name of the column that contain information on
regions or mega-environments. |
year |
A string or NULL. The name of the column that contain information on
years (or seasons). |
res.het |
Should the model consider heterogeneous residual variances?
Defaults for |
iter |
A positive integer specifying the number of iterations for each chain (including warmup). The default is 2000. |
cores |
Number of cores to use when executing the chains in parallel,
which defaults to 1 but we recommend setting the |
chains |
A positive integer specifying the number of Markov chains. The default is 4. |
pars |
A vector of character strings specifying parameters of interest.
The default is |
warmup |
A positive integer specifying the number of warmup (aka burnin)
iterations per chain. If step-size adaptation is on (which it is by default),
this also controls the number of iterations for which adaptation is run (and
hence these warmup samples should not be used for inference). The number of
warmup iterations should be smaller than |
thin |
A positive integer specifying the period for saving samples. The default is 1, which is usually the recommended value. |
seed |
The seed for random number generation. The default is generated
from 1 to the maximum integer supported by R on the machine. Even if
multiple chains are used, only one seed is needed, with other chains having
seeds derived from that of the first chain to avoid dependent samples.
When a seed is specified by a number, |
init |
Initial values specification. See the detailed documentation for
the init argument in |
verbose |
|
algorithm |
One of sampling algorithms that are implemented in Stan.
Current options are |
control |
A named |
include |
Logical scalar defaulting to |
show_messages |
Either a logical scalar (defaulting to |
... |
Additional arguments can be |
The function has nine available models, which will be fitted according to the options set in the arguments:
Entry-mean model : fitted when repl = NULL
, reg = NULL
and year = NULL
:
y = \mu + g + l + \varepsilon
Where y
is the phenotype, \mu
is the intercept, g
is the genotypic
effect, l
is the location (or environment) effect, and \varepsilon
is
the residue (which contains the genotype-by-location interaction, in this case).
Randomized complete blocks design : fitted when repl
is a single string.
It will fit different models depending if reg
and year
are NULL
:
reg = NULL
and year = NULL
:
y = \mu + g + l + gl + r + \varepsilon
where gl
is the genotype-by-location effect, and r
is the replicate effect.
reg = "reg"
and year = NULL
:
y = \mu + g + m + l + gl + gm + r + \varepsilon
where m
is the region effect, and gm
is the genotype-by-region effect.
reg = NULL
and year = "year"
:
y = \mu + g + t + l + gl + gt + r + \varepsilon
where t
is the year effect, and gt
is the genotype-by-year effect.
reg = "reg"
and year = "year"
:
y = \mu + g + m + t + l + gl + gm + gt + r + \varepsilon
Incomplete blocks design : fitted when repl
is a string vector of size 2.
It will fit different models depending if reg
and year
are NULL
:
reg = NULL
and year = NULL
:
y = \mu + g + l + gl + r + b + \varepsilon
where b
is the block within replicates effect.
reg = "reg"
and year = NULL
:
y = \mu + g + m + l + gl + gm + r + b + \varepsilon
reg = NULL
and year = "year"
:
y = \mu + g + t + l + gl + gt + r + b + \varepsilon
reg = "reg"
and year = "year"
:
y = \mu + g + m + t + l + gl + gm + gt + r + b + \varepsilon
The models described above have predefined priors:
x \sim \mathcal{N} \left( 0, S^{[x]} \right)
\sigma \sim \mathcal{HalfCauchy}\left( 0, S^{[\sigma]} \right)
where x
can be any effect but the error, and \sigma
is the standard
deviation of the likelihood. If res.het = TRUE
, then \sigma_k \sim \mathcal{HalfCauchy}\left( 0, S^{\left[ \sigma_k \right]} \right)
.
The hyperpriors are set as follows:
S^{[x]} \sim \mathcal{HalfCauchy}\left( 0, \phi \right)
where \phi
is the known global hyperparameter defined such as \phi = max(y) \times 10
.
More details about the usage of bayes_met
and other functions of
the ProbBreed
package can be found at https://saulo-chaves.github.io/ProbBreed_site/.
Solutions to convergence or mixing issues can be found at
https://mc-stan.org/misc/warnings.html.
An object of S4 class stanfit
representing
the fitted results. Slot mode
for this object
indicates if the sampling is done or not.
sampling
signature(object = "stanmodel")
Call a sampler (NUTS, HMC, or Fixed_param depending on parameters)
to draw samples from the model defined by S4 class stanmodel
given the data, initial values, etc.
rstan::sampling, rstan::stan, rstan::stanfit
mod = bayes_met(data = soy,
gen = "Gen",
loc = "Loc",
repl = NULL,
year = NULL,
reg = NULL,
res.het = TRUE,
trait = 'Y',
iter = 6000, cores = 4, chains = 4)