S3params compat (#273)

* tests pass with new paradox * forgot to save * test with new paradox * news * dev cmd check with paradox master
mlr-org · Feb 29, 2024 · be362e0 · be362e0
1 parent a888203
commit be362e0
Show file tree

Hide file tree

Showing 30 changed files with 103 additions and 164 deletions.
diff --git a/.github/workflows/dev-cmd-check.yml b/.github/workflows/dev-cmd-check.yml
@@ -25,6 +25,7 @@ jobs:
       matrix:
         config:
           - {os: ubuntu-latest, r: 'release', dev-package: 'mlr-org/mlr3'}
+          - {os: ubuntu-latest, r: 'release', dev-package: 'mlr-org/paradox'}
 
     steps:
       - uses: actions/checkout@v3

diff --git a/NEWS.md b/NEWS.md
@@ -1,5 +1,7 @@
 # mlr3learners (development version)
 
+* Adaption to new paradox version 1.0.0.
+
 # mlr3learners 0.5.8
 
 * Adaption to memory optimization in mlr3 0.17.1.

diff --git a/R/LearnerClassifCVGlmnet.R b/R/LearnerClassifCVGlmnet.R
@@ -40,7 +40,7 @@ LearnerClassifCVGlmnet = R6Class("LearnerClassifCVGlmnet",
         exmx                 = p_dbl(default = 250.0, tags = "train"),
         fdev                 = p_dbl(0, 1, default = 1.0e-5, tags = "train"),
         foldid               = p_uty(default = NULL, tags = "train"),
-        gamma                = p_uty(tags = "train"),
+        gamma                = p_uty(tags = "train", depends = relax == TRUE),
         grouped              = p_lgl(default = TRUE, tags = "train"),
         intercept            = p_lgl(default = TRUE, tags = "train"),
         keep                 = p_lgl(default = FALSE, tags = "train"),
@@ -72,7 +72,6 @@ LearnerClassifCVGlmnet = R6Class("LearnerClassifCVGlmnet",
         type.multinomial     = p_fct(c("ungrouped", "grouped"), tags = "train"),
         upper.limits         = p_uty(tags = "train")
       )
-      ps$add_dep("gamma", "relax", CondEqual$new(TRUE))
 
       super$initialize(
         id = "classif.cv_glmnet",

diff --git a/R/LearnerClassifGlmnet.R b/R/LearnerClassifGlmnet.R
@@ -55,7 +55,7 @@ LearnerClassifGlmnet = R6Class("LearnerClassifGlmnet",
         exclude              = p_int(1L, tags = "train"),
         exmx                 = p_dbl(default = 250.0, tags = "train"),
         fdev                 = p_dbl(0, 1, default = 1.0e-5, tags = "train"),
-        gamma                = p_dbl(default = 1, tags = "predict"),
+        gamma                = p_dbl(default = 1, tags = "predict", depends = relax == TRUE),
         intercept            = p_lgl(default = TRUE, tags = "train"),
         lambda               = p_uty(tags = "train"),
         lambda.min.ratio     = p_dbl(0, 1, tags = "train"),
@@ -82,7 +82,6 @@ LearnerClassifGlmnet = R6Class("LearnerClassifGlmnet",
         type.multinomial     = p_fct(c("ungrouped", "grouped"), tags = "train"),
         upper.limits         = p_uty(tags = "train")
       )
-      ps$add_dep("gamma", "relax", CondEqual$new(TRUE))
 
       super$initialize(
         id = "classif.glmnet",

diff --git a/R/LearnerClassifLDA.R b/R/LearnerClassifLDA.R
@@ -32,13 +32,12 @@ LearnerClassifLDA = R6Class("LearnerClassifLDA",
       ps = ps(
         dimen          = p_uty(tags = "predict"),
         method         = p_fct(c("moment", "mle", "mve", "t"), default = "moment", tags = "train"),
-        nu             = p_int(tags = "train"),
+        nu             = p_int(tags = "train", depends = method == "t"),
         predict.method = p_fct(c("plug-in", "predictive", "debiased"), default = "plug-in", tags = "predict"),
         predict.prior  = p_uty(tags = "predict"),
         prior          = p_uty(tags = "train"),
         tol            = p_dbl(tags = "train")
       )
-      ps$add_dep("nu", "method", CondEqual$new("t"))
 
       super$initialize(
         id = "classif.lda",

diff --git a/R/LearnerClassifQDA.R b/R/LearnerClassifQDA.R
@@ -31,12 +31,11 @@ LearnerClassifQDA = R6Class("LearnerClassifQDA",
     initialize = function() {
       ps = ps(
         method         = p_fct(c("moment", "mle", "mve", "t"), default = "moment", tags = "train"),
-        nu             = p_int(tags = "train"),
+        nu             = p_int(tags = "train", depends = method == "t"),
         predict.method = p_fct(c("plug-in", "predictive", "debiased"), default = "plug-in", tags = "predict"),
         predict.prior  = p_uty(tags = "predict"),
         prior          = p_uty(tags = "train")
       )
-      ps$add_dep("nu", "method", CondEqual$new("t"))
 
       super$initialize(
         id = "classif.qda",

diff --git a/R/LearnerClassifRanger.R b/R/LearnerClassifRanger.R
@@ -49,8 +49,8 @@ LearnerClassifRanger = R6Class("LearnerClassifRanger",
         minprop                      = p_dbl(default = 0.1, tags = "train"),
         mtry                         = p_int(lower = 1L, special_vals = list(NULL), tags = "train"),
         mtry.ratio                   = p_dbl(lower = 0, upper = 1, tags = "train"),
+        num.random.splits            = p_int(1L, default = 1L, tags = "train", depends = splitrule == "extratrees"),
         node.stats                   = p_lgl(default = FALSE, tags = "train"),
-        num.random.splits            = p_int(1L, default = 1L, tags = "train"),
         num.threads                  = p_int(1L, default = 1L, tags = c("train", "predict", "threads")),
         num.trees                    = p_int(1L, default = 500L, tags = c("train", "predict", "hotstart")),
         oob.error                    = p_lgl(default = TRUE, tags = "train"),
@@ -60,7 +60,7 @@ LearnerClassifRanger = R6Class("LearnerClassifRanger",
         respect.unordered.factors    = p_fct(c("ignore", "order", "partition"), default = "ignore", tags = "train"),
         sample.fraction              = p_dbl(0L, 1L, tags = "train"),
         save.memory                  = p_lgl(default = FALSE, tags = "train"),
-        scale.permutation.importance = p_lgl(default = FALSE, tags = "train"),
+        scale.permutation.importance = p_lgl(default = FALSE, tags = "train", depends = importance == "permutation"),
         se.method                    = p_fct(c("jack", "infjack"), default = "infjack", tags = "predict"),
         seed                         = p_int(default = NULL, special_vals = list(NULL), tags = c("train", "predict")),
         split.select.weights         = p_uty(default = NULL, tags = "train"),
@@ -71,9 +71,6 @@ LearnerClassifRanger = R6Class("LearnerClassifRanger",
 
       ps$values = list(num.threads = 1L)
 
-      ps$add_dep("num.random.splits", "splitrule", CondEqual$new("extratrees"))
-      ps$add_dep("scale.permutation.importance", "importance", CondEqual$new("permutation"))
-
       super$initialize(
         id = "classif.ranger",
         param_set = ps,

diff --git a/R/LearnerClassifSVM.R b/R/LearnerClassifSVM.R
@@ -26,26 +26,21 @@ LearnerClassifSVM = R6Class("LearnerClassifSVM",
       ps = ps(
         cachesize       = p_dbl(default = 40L, tags = "train"),
         class.weights   = p_uty(default = NULL, tags = "train"),
-        coef0           = p_dbl(default = 0, tags = "train"),
-        cost            = p_dbl(0, default = 1, tags = "train"),
+        coef0           = p_dbl(default = 0, tags = "train", depends = kernel %in% c("polynomial", "sigmoid")),
+        cost            = p_dbl(0, default = 1, tags = "train", depends = type == "C-classification"),
         cross           = p_int(0L, default = 0L, tags = "train"),
         decision.values = p_lgl(default = FALSE, tags = "predict"),
-        degree          = p_int(1L, default = 3L, tags = "train"),
+        degree          = p_int(1L, default = 3L, tags = "train", depends = kernel == "polynomial"),
         epsilon         = p_dbl(0, default = 0.1, tags = "train"),
         fitted          = p_lgl(default = TRUE, tags = "train"),
-        gamma           = p_dbl(0, tags = "train"),
+        gamma           = p_dbl(0, tags = "train", depends = kernel %in% c("polynomial", "radial", "sigmoid")),
         kernel          = p_fct(c("linear", "polynomial", "radial", "sigmoid"), default = "radial", tags = "train"),
-        nu              = p_dbl(default = 0.5, tags = "train"),
+        nu              = p_dbl(default = 0.5, tags = "train", depends = type == "nu-classification"),
         scale           = p_uty(default = TRUE, tags = "train"),
         shrinking       = p_lgl(default = TRUE, tags = "train"),
         tolerance       = p_dbl(0, default = 0.001, tags = "train"),
         type            = p_fct(c("C-classification", "nu-classification"), default = "C-classification", tags = "train")
       )
-      ps$add_dep("cost", "type", CondEqual$new("C-classification"))
-      ps$add_dep("nu", "type", CondEqual$new("nu-classification"))
-      ps$add_dep("degree", "kernel", CondEqual$new("polynomial"))
-      ps$add_dep("coef0", "kernel", CondAnyOf$new(c("polynomial", "sigmoid")))
-      ps$add_dep("gamma", "kernel", CondAnyOf$new(c("polynomial", "radial", "sigmoid")))
 
       super$initialize(
         id = "classif.svm",

diff --git a/R/LearnerClassifXgboost.R b/R/LearnerClassifXgboost.R
@@ -92,74 +92,57 @@ LearnerClassifXgboost = R6Class("LearnerClassifXgboost",
         early_stopping_set          = p_fct(c("none", "train", "test"), default = "none", tags = "train"),
         eta                         = p_dbl(0, 1, default = 0.3, tags = c("train", "control")),
         eval_metric                 = p_uty(tags = "train"),
-        feature_selector            = p_fct(c("cyclic", "shuffle", "random", "greedy", "thrifty"), default = "cyclic", tags = "train"),
+        feature_selector            = p_fct(c("cyclic", "shuffle", "random", "greedy", "thrifty"), default = "cyclic", tags = "train", depends = booster == "gblinear"),
         feval                       = p_uty(default = NULL, tags = "train"),
         gamma                       = p_dbl(0, default = 0, tags = c("train", "control")),
-        grow_policy                 = p_fct(c("depthwise", "lossguide"), default = "depthwise", tags = "train"),
+        grow_policy                 = p_fct(c("depthwise", "lossguide"), default = "depthwise", tags = "train", depends = tree_method == "hist"),
         interaction_constraints     = p_uty(tags = "train"),
         iterationrange              = p_uty(tags = "predict"),
         lambda                      = p_dbl(0, default = 1, tags = "train"),
-        lambda_bias                 = p_dbl(0, default = 0, tags = "train"),
-        max_bin                     = p_int(2L, default = 256L, tags = "train"),
+        lambda_bias                 = p_dbl(0, default = 0, tags = "train", depends = booster == "gblinear"),
+        max_bin                     = p_int(2L, default = 256L, tags = "train", depends = tree_method == "hist"),
         max_delta_step              = p_dbl(0, default = 0, tags = "train"),
         max_depth                   = p_int(0L, default = 6L, tags = c("train", "control")),
-        max_leaves                  = p_int(0L, default = 0L, tags = "train"),
+        max_leaves                  = p_int(0L, default = 0L, tags = "train", depends = grow_policy == "lossguide"),
         maximize                    = p_lgl(default = NULL, special_vals = list(NULL), tags = "train"),
         min_child_weight            = p_dbl(0, default = 1, tags = c("train", "control")),
         missing                     = p_dbl(default = NA, tags = c("train", "predict"), special_vals = list(NA, NA_real_, NULL)),
         monotone_constraints        = p_uty(default = 0, tags = c("train", "control"), custom_check = crate(function(x) { checkmate::check_integerish(x, lower = -1, upper = 1, any.missing = FALSE) })), # nolint
-        normalize_type              = p_fct(c("tree", "forest"), default = "tree", tags = "train"),
+        normalize_type              = p_fct(c("tree", "forest"), default = "tree", tags = "train", depends = booster == "dart"),
         nrounds                     = p_int(1L, tags = c("train", "hotstart")),
         nthread                     = p_int(1L, default = 1L, tags = c("train", "control", "threads")),
         ntreelimit                  = p_int(1L, default = NULL, special_vals = list(NULL), tags = "predict"),
         num_parallel_tree           = p_int(1L, default = 1L, tags = c("train", "control")),
         objective                   = p_uty(default = "binary:logistic", tags = c("train", "predict", "control")),
-        one_drop                    = p_lgl(default = FALSE, tags = "train"),
+        one_drop                    = p_lgl(default = FALSE, tags = "train", depends = booster == "dart"),
         outputmargin                = p_lgl(default = FALSE, tags = "predict"),
         predcontrib                 = p_lgl(default = FALSE, tags = "predict"),
         predinteraction             = p_lgl(default = FALSE, tags = "predict"),
         predleaf                    = p_lgl(default = FALSE, tags = "predict"),
-        print_every_n               = p_int(1L, default = 1L, tags = "train"),
+        print_every_n               = p_int(1L, default = 1L, tags = "train", depends = verbose == 1L),
         process_type                = p_fct(c("default", "update"), default = "default", tags = "train"),
-        rate_drop                   = p_dbl(0, 1, default = 0, tags = "train"),
+        rate_drop                   = p_dbl(0, 1, default = 0, tags = "train", depends = booster == "dart"),
         refresh_leaf                = p_lgl(default = TRUE, tags = "train"),
         reshape                     = p_lgl(default = FALSE, tags = "predict"),
         seed_per_iteration          = p_lgl(default = FALSE, tags = "train"),
-        sampling_method             = p_fct(c("uniform", "gradient_based"), default = "uniform", tags = "train"),
-        sample_type                 = p_fct(c("uniform", "weighted"), default = "uniform", tags = "train"),
+        sampling_method             = p_fct(c("uniform", "gradient_based"), default = "uniform", tags = "train", depends = booster == "gbtree"),
+        sample_type                 = p_fct(c("uniform", "weighted"), default = "uniform", tags = "train", depends = booster == "dart"),
         save_name                   = p_uty(default = NULL, tags = "train"),
         save_period                 = p_int(0, default = NULL, special_vals = list(NULL), tags = "train"),
         scale_pos_weight            = p_dbl(default = 1, tags = "train"),
-        skip_drop                   = p_dbl(0, 1, default = 0, tags = "train"),
+        skip_drop                   = p_dbl(0, 1, default = 0, tags = "train", depends = booster == "dart"),
         strict_shape                = p_lgl(default = FALSE, tags = "predict"),
         subsample                   = p_dbl(0, 1, default = 1, tags = c("train", "control")),
-        top_k                       = p_int(0, default = 0, tags = "train"),
+        top_k                       = p_int(0, default = 0, tags = "train", depends = feature_selector %in% c("greedy", "thrifty") && booster == "gblinear"),
         training                    = p_lgl(default = FALSE, tags = "predict"),
-        tree_method                 = p_fct(c("auto", "exact", "approx", "hist", "gpu_hist"), default = "auto", tags = "train"),
-        tweedie_variance_power      = p_dbl(1, 2, default = 1.5, tags = "train"),
+        tree_method                 = p_fct(c("auto", "exact", "approx", "hist", "gpu_hist"), default = "auto", tags = "train", depends = booster %in% c("gbtree", "dart")),
+        tweedie_variance_power      = p_dbl(1, 2, default = 1.5, tags = "train", depends = objective == "reg:tweedie"),
         updater                     = p_uty(tags = "train"), # Default depends on the selected booster
         verbose                     = p_int(0L, 2L, default = 1L, tags = "train"),
         watchlist                   = p_uty(default = NULL, tags = "train"),
         xgb_model                   = p_uty(default = NULL, tags = "train")
 
       )
-      # param deps
-      ps$add_dep("tweedie_variance_power", "objective", CondEqual$new("reg:tweedie"))
-      ps$add_dep("print_every_n", "verbose", CondEqual$new(1L))
-      ps$add_dep("sampling_method", "booster", CondEqual$new("gbtree"))
-      ps$add_dep("normalize_type", "booster", CondEqual$new("dart"))
-      ps$add_dep("rate_drop", "booster", CondEqual$new("dart"))
-      ps$add_dep("skip_drop", "booster", CondEqual$new("dart"))
-      ps$add_dep("one_drop", "booster", CondEqual$new("dart"))
-      ps$add_dep("sample_type", "booster", CondEqual$new("dart"))
-      ps$add_dep("tree_method", "booster", CondAnyOf$new(c("gbtree", "dart")))
-      ps$add_dep("grow_policy", "tree_method", CondEqual$new("hist"))
-      ps$add_dep("max_leaves", "grow_policy", CondEqual$new("lossguide"))
-      ps$add_dep("max_bin", "tree_method", CondEqual$new("hist"))
-      ps$add_dep("feature_selector", "booster", CondEqual$new("gblinear"))
-      ps$add_dep("top_k", "booster", CondEqual$new("gblinear"))
-      ps$add_dep("top_k", "feature_selector", CondAnyOf$new(c("greedy", "thrifty")))
-      ps$add_dep("lambda_bias", "booster", CondEqual$new("gblinear"))
 
       # custom defaults
       ps$values = list(nrounds = 1L, nthread = 1L, verbose = 0L, early_stopping_set = "none")

diff --git a/R/LearnerRegrCVGlmnet.R b/R/LearnerRegrCVGlmnet.R
@@ -38,7 +38,7 @@ LearnerRegrCVGlmnet = R6Class("LearnerRegrCVGlmnet",
         family               = p_fct(c("gaussian", "poisson"), default = "gaussian", tags = "train"),
         fdev                 = p_dbl(0, 1, default = 1.0e-5, tags = "train"),
         foldid               = p_uty(default = NULL, tags = "train"),
-        gamma                = p_uty(tags = "train"),
+        gamma                = p_uty(tags = "train", depends = relax == TRUE),
         grouped              = p_lgl(default = TRUE, tags = "train"),
         intercept            = p_lgl(default = TRUE, tags = "train"),
         keep                 = p_lgl(default = FALSE, tags = "train"),
@@ -64,14 +64,12 @@ LearnerRegrCVGlmnet = R6Class("LearnerRegrCVGlmnet",
         standardize.response = p_lgl(default = FALSE, tags = "train"),
         thresh               = p_dbl(0, default = 1e-07, tags = "train"),
         trace.it             = p_int(0, 1, default = 0, tags = "train"),
-        type.gaussian        = p_fct(c("covariance", "naive"), tags = "train"),
+        type.gaussian        = p_fct(c("covariance", "naive"), tags = "train", depends = family == "gaussian"),
         type.logistic        = p_fct(c("Newton", "modified.Newton"), tags = "train"),
         type.measure         = p_fct(c("deviance", "class", "auc", "mse", "mae"), default = "deviance", tags = "train"),
         type.multinomial     = p_fct(c("ungrouped", "grouped"), tags = "train"),
         upper.limits         = p_uty(tags = "train")
       )
-      ps$add_dep("gamma", "relax", CondEqual$new(TRUE))
-      ps$add_dep("type.gaussian", "family", CondEqual$new("gaussian"))
 
       ps$values = list(family = "gaussian")
 

diff --git a/R/LearnerRegrGlmnet.R b/R/LearnerRegrGlmnet.R
@@ -40,7 +40,7 @@ LearnerRegrGlmnet = R6Class("LearnerRegrGlmnet",
         exmx                  = p_dbl(default = 250.0, tags = "train"),
         family                = p_fct(c("gaussian", "poisson"), default = "gaussian", tags = "train"),
         fdev                  = p_dbl(0, 1, default = 1.0e-5, tags = "train"),
-        gamma                 = p_dbl(default = 1, tags = "train"),
+        gamma                 = p_dbl(default = 1, tags = "train", depends = relax == TRUE),
         grouped               = p_lgl(default = TRUE, tags = "train"),
         intercept             = p_lgl(default = TRUE, tags = "train"),
         keep                  = p_lgl(default = FALSE, tags = "train"),
@@ -65,13 +65,11 @@ LearnerRegrGlmnet = R6Class("LearnerRegrGlmnet",
         standardize.response  = p_lgl(default = FALSE, tags = "train"),
         thresh                = p_dbl(0, default = 1e-07, tags = "train"),
         trace.it              = p_int(0, 1, default = 0, tags = "train"),
-        type.gaussian         = p_fct(c("covariance", "naive"), tags = "train"),
+        type.gaussian         = p_fct(c("covariance", "naive"), tags = "train", depends = family == "gaussian"),
         type.logistic         = p_fct(c("Newton", "modified.Newton"), tags = "train"),
         type.multinomial      = p_fct(c("ungrouped", "grouped"), tags = "train"),
         upper.limits          = p_uty(tags = "train")
       )
-      ps$add_dep("gamma", "relax", CondEqual$new(TRUE))
-      ps$add_dep("type.gaussian", "family", CondEqual$new("gaussian"))
 
       ps$values = list(family = "gaussian")