fix issue #1380

NEWS.md

@@ -8,6 +8,8 @@ specified via `s(..., fx = TRUE)`.
 to the Stan backend in `update` and related methods. (#1373, #1378)
 * Allow to retain unused factors levels via `drop_unused_levels = FALSE` 
 in `brm` and related functions. (#1346)
+* Automatically update old default priors based on new input when 
+when updating models via `update.brmsfit`. (#1380)
 
 ### Other Changes
 

R/update.R

@@ -147,8 +147,10 @@ update.brmsfit <- function(object, formula., newdata = NULL,
     if (!is.brmsprior(dots$prior)) {
       stop2("Argument 'prior' needs to be a 'brmsprior' object.")
     }
-    # update existing priors manually
-    dots$prior <- rbind(dots$prior, object$prior)
+    # update existing priors manually and keep only user-specified ones
+    # default priors are recomputed base on newdata if provided
+    old_user_prior <- subset2(object$prior, source = "user")
+    dots$prior <- rbind(dots$prior, old_user_prior)
     dupl_priors <- duplicated(dots$prior[, rcols_prior()])
     dots$prior <- dots$prior[!dupl_priors, ]
   }
@@ -241,6 +243,10 @@ update.brmsfit <- function(object, formula., newdata = NULL,
       dots$data, bterms = bterms, data2 = object$data2,
       knots = dots$knots, drop_unused_levels = dots$drop_unused_levels
     )
+    object$prior <- .validate_prior(
+      dots$prior, bterms = bterms, data = object$data,
+      sample_prior = dots$sample_prior
+    )
     object$family <- get_element(object$formula, "family")
     object$autocor <- get_element(object$formula, "autocor")
     object$ranef <- tidy_ranef(bterms, data = object$data)