diff --git a/R/PipeOpNMF.R b/R/PipeOpNMF.R
index e708af7ea..aa2beb689 100644
--- a/R/PipeOpNMF.R
+++ b/R/PipeOpNMF.R
@@ -125,10 +125,8 @@ PipeOpNMF = R6Class("PipeOpNMF",
         track = p_lgl(default = FALSE, tags = c("train", "nmf.options")),
         verbose = p_uty(default = FALSE, tags = c("train", "nmf.options")),
         pbackend = p_uty(tags = c("train", "nmf")),  # .pbackend
-        callback = p_uty(tags = c("train", "nmf"))  # .callback
+        callback = p_uty(tags = c("train", "nmf"), depends = keep.all == TRUE)  # .callback
       )
-      ps$add_dep("keep.all", on = "nrun", cond = CondLarger$new(1))
-      ps$add_dep("callback", on = "keep.all", cond = CondEqual$new(TRUE))
       ps$values = list(rank = 2L, method = "brunet", parallel = FALSE, parallel.required = FALSE)
       super$initialize(id, param_set = ps, param_vals = param_vals, feature_types = c("numeric", "integer"), packages = c("MASS", "NMF"))
     }
@@ -179,10 +177,11 @@ PipeOpNMF = R6Class("PipeOpNMF",
 
 mlr_pipeops$add("nmf", PipeOpNMF)
 
-CondLarger = R6Class("CondLarger", inherit = Condition,
-  public = list(
-    initialize = function(rhs) super$initialize("larger", rhs),
-    test = function(x) !is.na(x) & x > self$rhs,
-    as_string = function(lhs_chr = "x") sprintf("%s > %s", lhs_chr, as.character(self$rhs))
-  )
-)
+# this is just a really bad idea
+## CondLarger = R6Class("CondLarger", inherit = Condition,
+##   public = list(
+##     initialize = function(rhs) super$initialize("larger", rhs),
+##     test = function(x) !is.na(x) & x > self$rhs,
+##     as_string = function(lhs_chr = "x") sprintf("%s > %s", lhs_chr, as.character(self$rhs))
+##   )
+## )
diff --git a/R/PipeOpTextVectorizer.R b/R/PipeOpTextVectorizer.R
index 8c4437fc6..eb1531714 100644
--- a/R/PipeOpTextVectorizer.R
+++ b/R/PipeOpTextVectorizer.R
@@ -187,7 +187,8 @@ PipeOpTextVectorizer = R6Class("PipeOpTextVectorizer",
         skip = p_uty(default = 0, tags = c("train", "predict", "ngrams"), custom_check = curry(check_integerish, min.len = 1, lower = 0, any.missing = FALSE)),
 
         sparsity = p_dbl(lower = 0, upper = 1, default = NULL,
-          tags = c("train", "dfm_trim"), special_vals = list(NULL)),
+          tags = c("train", "dfm_trim"), special_vals = list(NULL),
+          depends = return_type == "bow"),
         termfreq_type = p_fct(default = "count", tags = c("train", "dfm_trim"),
           levels = c("count", "prop", "rank", "quantile")),
         min_termfreq = p_dbl(lower = 0, default = NULL,
@@ -197,29 +198,21 @@ PipeOpTextVectorizer = R6Class("PipeOpTextVectorizer",
 
         scheme_df = p_fct(default = "count", tags = c("train", "docfreq"),
           levels = c("count", "inverse", "inversemax", "inverseprob", "unary")),
-        smoothing_df = p_dbl(lower = 0, default = 0, tags = c("train", "docfreq")),
-        k_df = p_dbl(lower = 0, tags = c("train", "docfreq")),
-        threshold_df = p_dbl(lower = 0, default = 0, tags = c("train", "docfreq")),
-        base_df = p_dbl(lower = 0, default = 10, tags = c("train", "docfreq")),
+        smoothing_df = p_dbl(lower = 0, default = 0, tags = c("train", "docfreq"), depends = scheme_df %in% c("inverse", "inversemax", "inverseprob")),
+        k_df = p_dbl(lower = 0, tags = c("train", "docfreq"), depends = scheme_df %in% c("inverse", "inversemax", "inverseprob")),
+        threshold_df = p_dbl(lower = 0, default = 0, tags = c("train", "docfreq"), depends = scheme_df == "count"),
+        base_df = p_dbl(lower = 0, default = 10, tags = c("train", "docfreq"),
+          depends = scheme_df %in% c("inverse", "inversemax", "inverseprob")),
 
-        scheme_tf = p_fct(default = "count", tags = c("train", "predict", "dfm_weight"),
+        scheme_tf = p_fct(default = "count", tags = c("train", "predict", "dfm_weight", depends = return_type == "bow"),
           levels = c("count", "prop", "propmax", "logcount", "boolean", "augmented", "logave")),
-        k_tf = p_dbl(lower = 0, upper = 1, tags = c("train", "predict", "dfm_weight")),
-        base_tf = p_dbl(lower = 0, default = 10, tags = c("train", "predict", "dfm_weight")),
+        k_tf = p_dbl(lower = 0, upper = 1, tags = c("train", "predict", "dfm_weight"), depends = scheme_tf == "augmented"),
+        base_tf = p_dbl(lower = 0, default = 10, tags = c("train", "predict", "dfm_weight"), depends = scheme_tf %in% c("logcount", "logave")),
 
         return_type = p_fct(levels = c("bow", "integer_sequence", "factor_sequence"), tags = c("train", "predict")),
-        sequence_length = p_int(default = 0, lower = 0, upper = Inf, tags = c("train", "predict", "integer_sequence"))
-      )$
-        add_dep("base_df", "scheme_df", CondAnyOf$new(c("inverse", "inversemax", "inverseprob")))$
-        add_dep("smoothing_df", "scheme_df", CondAnyOf$new(c("inverse", "inversemax", "inverseprob")))$
-        add_dep("k_df", "scheme_df", CondAnyOf$new(c("inverse", "inversemax", "inverseprob")))$
-        add_dep("base_df", "scheme_df", CondAnyOf$new(c("inverse", "inversemax", "inverseprob")))$
-        add_dep("threshold_df", "scheme_df", CondEqual$new("count"))$
-        add_dep("k_tf", "scheme_tf", CondEqual$new("augmented"))$
-        add_dep("base_tf", "scheme_tf", CondAnyOf$new(c("logcount", "logave")))$
-        add_dep("scheme_tf", "return_type", CondEqual$new("bow"))$
-        add_dep("sparsity", "return_type", CondEqual$new("bow"))$
-        add_dep("sequence_length", "return_type", CondAnyOf$new(c("integer_sequence", "factor_sequence")))
+        sequence_length = p_int(default = 0, lower = 0, upper = Inf, tags = c("train", "predict", "integer_sequence"),
+          depends = return_type %in% c("integer_sequence", "factor_sequence"))
+      )
 
       ps$values = list(stopwords_language = "smart", extra_stopwords = character(0), n = 1, scheme_df = "unary", return_type = "bow")
       super$initialize(id = id, param_set = ps, param_vals = param_vals, packages = c("quanteda", "stopwords"), feature_types = "character")
diff --git a/R/PipeOpTuneThreshold.R b/R/PipeOpTuneThreshold.R
index 440e2b6df..e4891eb6f 100644
--- a/R/PipeOpTuneThreshold.R
+++ b/R/PipeOpTuneThreshold.R
@@ -120,7 +120,7 @@ PipeOpTuneThreshold = R6Class("PipeOpTuneThreshold",
       ps = private$.make_param_set(pred)
       measure = self$param_set$values$measure
       if (is.character(measure)) measure = msr(measure) else measure
-      codomain = do.call(ps, structure(list(p_dbl(tags = ifelse(measure$minimize, "minimize", "maximize"))), names = measure$id))
+      codomain = do.call(paradox::ps, structure(list(p_dbl(tags = ifelse(measure$minimize, "minimize", "maximize"))), names = measure$id))
 
       objfun = bbotk::ObjectiveRFun$new(
         fun = function(xs) private$.objfun(xs, pred = pred, measure = measure),
diff --git a/R/PipeOpVtreat.R b/R/PipeOpVtreat.R
index 473723ebd..8a87e9ba5 100644
--- a/R/PipeOpVtreat.R
+++ b/R/PipeOpVtreat.R
@@ -136,7 +136,7 @@ PipeOpVtreat = R6Class("PipeOpVtreat",
         smFactor = p_dbl(lower = 0, upper = Inf, default = 0, tags = c("train", "regression", "classification", "multinomial")),
         rareCount = p_int(lower = 0L, upper = Inf, default = 0, tags = c("train", "regression", "classification", "multinomial")),
         rareSig = p_dbl(lower = 0, upper = 1, special_vals = list(NULL), tags = c("train", "regression", "classification", "multinomial")),  # default NULL for regression, classification, 1 for multinomial
-        collarProb = p_dbl(lower = 0, upper = 1, default = 0, tags = c("train", "regression", "classification", "multinomial")),
+        collarProb = p_dbl(lower = 0, upper = 1, default = 0, tags = c("train", "regression", "classification", "multinomial"), depends = doCollar == TRUE),
         doCollar = p_lgl(default = FALSE, tags = c("train", "regression", "classification", "multinomial")),
         codeRestriction = p_uty(default = NULL, custom_check = function(x) checkmate::check_character(x, any.missing = FALSE, null.ok = TRUE),
           tags = c("train", "regression", "classification", "multinomial")),
@@ -160,7 +160,6 @@ PipeOpVtreat = R6Class("PipeOpVtreat",
         imputation_map = p_uty(default = NULL, custom_check = function(x) checkmate::check_list(x, null.ok = TRUE), tags = c("train", "predict"))
         # NOTE: parallelCluster missing intentionally and will be set to NULL
       )
-      ps$add_dep("collarProb", on = "doCollar", cond = CondEqual$new(TRUE))
       ps$values = list(recommended = TRUE, cols_to_copy = selector_none())
       super$initialize(id, param_set = ps, param_vals = param_vals, packages = "vtreat", tags = c("encode", "missings"))
     }
diff --git a/tests/testthat/helper_test_pipeops.R b/tests/testthat/helper_test_pipeops.R
index 0589a0791..5265177bb 100644
--- a/tests/testthat/helper_test_pipeops.R
+++ b/tests/testthat/helper_test_pipeops.R
@@ -45,7 +45,7 @@ PipeOpDebugMulti = R6Class("PipeOpDebugMulti",
       p = ps(par = p_int(lower = 0, upper = 10, default = 0, tags = c("train", "predict")))
       self$nin = length(inputs)
       self$nout = length(outputs)
-      super$initialize(id, p),
+      super$initialize(id, param_set = p,
         input = data.table(name = inputs, train = "*", predict = "*"),
         output = data.table(name = outputs, train = "*", predict = "*"))
     }),