set_prior(prior, class = "b", coef = "", group = "", nlpar = "",
lb = NULL, ub = NULL)"b" (fixed effects).
See 'Details' for other valid parameter classes.class = "b". Defaults to NULL, that is no restriction.class = "b". Defaults to NULL, that is no restriction.brmsprior to be used in the prior
argument of brm.set_prior is used to define prior distributions for parameters
in c(...),
e.g., c(set_prior(...), set_prior(...)).
b_,
where y is predicted by x1 and x2
(i.e. y ~ x1+x2 in formula syntax).
Then, x1 and x2 have regression parameters
b_x1 and b_x2 respectively.
The default prior for fixed and category specific effects is an
improper flat prior over the reals. Other common options are normal priors
or student-t priors. If we want to have a normal prior with mean 0 and
standard deviation 5 for x1,
and a unit student-t prior with 10 degrees of freedom for x2,
we can specify this via
set_prior("normal(0,5)", class = "b", coef = "x1") and
set_prior("student_t(10,0,1)", class = "b", coef = "x2").
To put the same prior on all fixed effects at once,
we may write as a shortcut set_prior("", class = "b") .
This also leads to faster sampling, because priors can be vectorized in this case.
Both ways of defining priors can be combined using for instance
set_prior("normal(0,2)", class = "b") and
set_prior("normal(0,10)", class = "b", coef = "x1")
at the same time. This will set a normal(0,10) prior on
the fixed effect of x1 and a normal(0,2) prior
on all other fixed effects. However, this will break vectorization and
may slow down the sampling procedure a bit.
In case of the default intercept parameterization
(discussed in the 'Details' section of brm),
the fixed effects intercept has its own parameter class
named "Intercept" and priors can thus be
specified via set_prior("", class = "Intercept") .
Setting a prior on the intercept will not break vectorization
of the other fixed effects.
A special shrinkage prior to be applied on fixed effects is the horseshoe prior.
It is symmetric around zero with fat tails and an infinitely large spike
at zero. This makes it ideal for sparse models that have
many regression coefficients,although only a minority of them is non-zero.
For more details see Carvalho et al. (2009).
The horseshoe prior can be applied on all fixed effects at once
(excluding the intercept) by using set_prior("horseshoe(1)").
The 1 implies that the student-t prior of the local shrinkage
parameters has 1 degrees of freedom. This may, however, lead to an
increased number of divergent transition in 3) may often be a better option, although the prior
no longer resembles a horseshoe in this case.
Generally, models with horseshoe priors a more likely than other models
to have divergent transitions so that increasing adapt_delta
from 0.8 to values closer to 1 will often be necessary.
See the documentation of brm for instructions
on how to increase adapt_delta.
In non-linear models, fixed effects are defined separately for each
non-linear parameter. Accordingly, it is necessary to specify
the corresponding non-linear parameter in set_prior so that priors
we can be assigned correctly.
If, for instance, alpha is the parameter and x the predictor
for which we want to define the prior, we can write
set_prior("", coef = "x", nlpar = "alpha") .
As a shortcut we can use set_prior("", nlpar = "alpha")
to set the same prior on all fixed effects of alpha at once.
If desired, fixed effects parameters can be restricted to fall only
within a certain interval using the lb and ub arguments
of set_prior. This is often required when defining priors
that are not defined everywhere on the real line, such as uniform
or gamma priors. When defining a uniform(2,4) prior,
you should write set_prior("uniform(2,4)", lb = 2, ub = 4).
When using a prior that is defined on the postive reals only
(such as a gamma prior) set lb = 0.
In most situations, it is not useful to restrict fixed effects
parameters through bounded priors, but if you really want to
this is the way to go.
3. Autocorrelation parameters
The autocorrelation parameters currently implemented are named
ar (autoregression), ma (moving average),
and arr (autoregression of the response).
Priors can be defined by set_prior("", class = "ar")
for ar and similar for ma and arr effects.
By default, ar and ma are bounded between -1
and 1 and arr is unbounded (you may change this
by using the arguments lb and ub). The default
prior is flat over the definition area.
4. Standard deviations of random effects
Each random effect of each grouping factor has a standard deviation named
sd__ . Consider, for instance, the formula
y ~ x1+x2+(1+x1|g).
We see that the intercept as well as x1 are random effects
nested in the grouping factor g.
The corresponding standard deviation parameters are named as
sd_g_Intercept and sd_g_x1 respectively.
These parameters are restriced to be non-negative and, by default,
have a half student-t prior with 3 degrees of freedom and a
scale parameter that depends on the standard deviation of the response
after applying the link function. Minimally, the scale parameter is 5.
To define a prior distribution only for standard deviations
of a specific grouping factor,
use
set_prior("", class = "sd", group = "") .
To define a prior distribution only for a specific standard deviation
of a specific grouping factor, you may write
set_prior("", class = "sd", group = "", coef = "") .
Recommendations on useful prior distributions for
standard deviations are given in Gelman (2006).
When defining priors on random effects parameters in non-linear models,
please make sure to specify the corresponding non-linear parameter
through the nlpar argument in the same way as for fixed effects.
5. Correlations of random effects
If there is more than one random effect per grouping factor,
the correlations between those random
effects have to be estimated.
The prior "lkj_corr_cholesky(eta)" or in short
"lkj(eta)" with eta > 0
is essentially the only prior for (choelsky factors) of correlation matrices.
If eta = 1 (the default) all correlations matrices
are equally likely a priori. If eta > 1, extreme correlations
become less likely, whereas 0 < eta < 1 results in
higher probabilities for extreme correlations.
Correlation matrix parameters in brms models are named as
cor_(group), (e.g., cor_g if g is the grouping factor).
To set the same prior on every correlation matrix,
use for instance set_prior("lkj(2)", class = "cor").
6. Parameters for specific families
Some families need additional parameters to be estimated.
Families gaussian, student, and cauchy
need the parameter sigma
to account for the residual standard deviation.
By default, sigma has a half student-t prior that scales
in the same way as the random effects standard deviations.
Furthermore, family student needs the parameter
nu representing the degrees of freedom of students t distribution.
By default, nu has prior "gamma(2,0.1)"
and a fixed lower bound of 1.
Families gamma, weibull, inverse.gaussian, and
negbinomial need a shape parameter that has a
"student_t(3,0,5)" prior by default.
For families cumulative, cratio, sratio,
and acat, and only if threshold = "equidistant",
the parameter delta is used to model the distance between
two adjacent thresholds.
By default, delta has an improper flat prior over the reals.
Every family specific parameter has its own prior class, so that
set_prior("", class = "") it the right way to go.
Often, it may not be immediately clear,
which parameters are present in the model.
To get a full list of parameters and parameter classes for which
priors can be specified (depending on the model)
use function get_prior.get_prior## check which parameters can have priors
get_prior(rating ~ treat + period + carry + (1|subject),
data = inhaler, family = sratio(),
threshold = "equidistant")
## define some priors
prior <- c(set_prior("normal(0,10)", class = "b"),
set_prior("normal(1,2)", class = "b", coef = "treat"),
set_prior("cauchy(0,2)", class = "sd",
group = "subject", coef = "Intercept"),
set_prior("uniform(-5,5)", class = "delta"))
## verify that the priors indeed found their way into Stan's model code
make_stancode(rating ~ period + carry + (1|subject),
data = inhaler, family = sratio(),
partial = ~ treat, threshold = "equidistant",
prior = prior)
## use horseshoe priors to model sparsity in fixed effects parameters
make_stancode(count ~ log_Age_c + log_Base4_c * Trt_c,
data = epilepsy, family = poisson(),
prior = set_prior("horseshoe(3)"))Run the code above in your browser using DataLab