Remove verbose descriptor magic

This is now just a regular attribute, we convert to a `level_enum` when
needed to forward to the `libcuml` API. In the future we'll remove this
forwarding entirely in favor of configuring the logger contextually
outside of the libcuml calls.

Author: jcrist

python/cuml/cuml/cluster/dbscan.pyx

@@ -386,7 +386,7 @@ class DBSCAN(Base,
 
         cdef float eps = self.eps
         cdef int min_samples = self.min_samples
-        cdef level_enum verbose = self.verbose
+        cdef level_enum verbose = self._verbose_level
         cdef bool multi_gpu = self._multi_gpu
         cdef size_t max_mbytes_per_batch = self.max_mbytes_per_batch or 0
 

python/cuml/cuml/cluster/kmeans.pyx

@@ -25,7 +25,6 @@ from libcpp cimport bool
 from pylibraft.common.handle cimport handle_t
 
 cimport cuml.cluster.cpp.kmeans as lib
-from cuml.internals.logger cimport level_enum
 from cuml.metrics.distance_type cimport DistanceType
 
 
@@ -36,7 +35,7 @@ cdef _kmeans_init_params(kmeans, lib.KMeansParams& params):
     params.n_clusters = kmeans.n_clusters
     params.max_iter = kmeans.max_iter
     params.tol = kmeans.tol
-    params.verbosity = <level_enum>(<int>kmeans.verbose)
+    params.verbosity = kmeans._verbose_level
     params.metric = DistanceType.L2Expanded
     params.batch_samples = int(kmeans.max_samples_per_batch)
     params.oversampling_factor = kmeans.oversampling_factor

python/cuml/cuml/ensemble/randomforest_common.pyx

@@ -187,7 +187,6 @@ class BaseRandomForestModel(Base, InteropMixin):
             "min_impurity_decrease",
             "max_batch_size",
             "random_state",
-            "criterion",
             "n_streams",
             "oob_score",
             "handle",
@@ -421,7 +420,7 @@ class BaseRandomForestModel(Base, InteropMixin):
         cdef uintptr_t y_ptr = y.ptr
         cdef int n_rows = X.shape[0]
         cdef int n_cols = X.shape[1]
-        cdef level_enum verbose = <level_enum> self.verbose
+        cdef level_enum verbose = <level_enum> self._verbose_level
         cdef int n_classes = self.n_classes_ if is_classifier else 0
 
         if self.max_depth <= 0:

python/cuml/cuml/experimental/linear_model/lars.pyx

@@ -262,13 +262,13 @@ class Lars(Base, RegressorMixin):
             larsFit(handle_[0], <float*> X_ptr, n_rows, <int> self.n_cols,
                     <float*> y_ptr, <float*> beta_ptr, <int*> active_idx_ptr,
                     <float*> alphas_ptr, &n_active, <float*> Gram_ptr,
-                    max_iter, <float*> coef_path_ptr, self.verbose, ld_X,
+                    max_iter, <float*> coef_path_ptr, self._verbose_level, ld_X,
                     ld_G, <float> self.eps)
         else:
             larsFit(handle_[0], <double*> X_ptr, n_rows, <int> self.n_cols,
                     <double*> y_ptr, <double*> beta_ptr, <int*> active_idx_ptr,
                     <double*> alphas_ptr, &n_active, <double*> Gram_ptr,
-                    max_iter, <double*> coef_path_ptr, self.verbose,
+                    max_iter, <double*> coef_path_ptr, self._verbose_level,
                     ld_X, ld_G, <double> self.eps)
         self.n_active = n_active
         self.n_iter_ = n_active

python/cuml/cuml/internals/base.py

@@ -14,8 +14,6 @@
 import cuml.internals.input_utils
 import cuml.internals.logger as logger
 import cuml.internals.nvtx as nvtx
-from cuml.internals import api_context_managers
-from cuml.internals.global_settings import GlobalSettings
 from cuml.internals.input_utils import determine_array_type
 from cuml.internals.mixins import TagsMixin
 from cuml.internals.output_type import (
@@ -24,40 +22,6 @@
 )
 
 
-class VerbosityDescriptor:
-    """Descriptor for ensuring correct type is used for verbosity
-
-    This descriptor ensures that when the 'verbose' attribute of a cuML
-    estimator is accessed external to the cuML API, an integer is returned
-    (consistent with Scikit-Learn's API for verbosity). Internal to the API, an
-    enum is used. Scikit-Learn's numerical values for verbosity are the inverse
-    of those used by spdlog, so the numerical value is also inverted internal
-    to the cuML API. This ensures that cuML code treats verbosity values as
-    expected for an spdlog-based codebase.
-    """
-
-    def __get__(self, obj, cls=None):
-        if api_context_managers.in_internal_api():
-            return logger._verbose_to_level(obj._verbose)
-        else:
-            return obj._verbose
-
-    def __set__(self, obj, value):
-        if api_context_managers.in_internal_api():
-            assert isinstance(value, logger.level_enum), (
-                "The log level should always be provided as a level_enum, "
-                "not an integer"
-            )
-            obj._verbose = logger._verbose_from_level(value)
-        else:
-            if isinstance(value, logger.level_enum):
-                raise ValueError(
-                    "The log level should always be provided as an integer, "
-                    "not using the enum"
-                )
-            obj._verbose = value
-
-
 class Base(TagsMixin, metaclass=cuml.internals.BaseMetaClass):
     """
     Base class for all the ML algos. It handles some of the common operations
@@ -191,28 +155,10 @@ def __init__(
         verbose=False,
         output_type=None,
     ):
-        """
-        Constructor. All children must call init method of this base class.
-
-        """
         self.handle = (
             pylibraft.common.handle.Handle() if handle is None else handle
         )
-
-        # The following manipulation of the root_cm ensures that the verbose
-        # descriptor sees any set or get of the verbose attribute as happening
-        # internal to the cuML API. Currently, __init__ calls do not take place
-        # within an api context manager, so setting "verbose" here would
-        # otherwise appear to be external to the cuML API. This behavior will
-        # be corrected with the update of cuML's API context manager
-        # infrastructure in https://github.com/rapidsai/cuml/pull/6189.
-        GlobalSettings().prev_root_cm = GlobalSettings().root_cm
-        GlobalSettings().root_cm = True
-        self.verbose = logger._verbose_to_level(verbose)
-        # Please see above note on manipulation of the root_cm. This should be
-        # rendered unnecessary with https://github.com/rapidsai/cuml/pull/6189.
-        GlobalSettings().root_cm = GlobalSettings().prev_root_cm
-
+        self.verbose = verbose
         self.output_type = _check_output_type_str(
             cuml.global_settings.output_type
             if output_type is None
@@ -224,8 +170,6 @@ def __init__(
         if nvtx_benchmark and nvtx_benchmark.lower() == "true":
             self.set_nvtx_annotations()
 
-    verbose = VerbosityDescriptor()
-
     def __repr__(self):
         """
         Pretty prints the arguments of a class using Scikit-learn standard :)
@@ -250,6 +194,11 @@ def __repr__(self):
             output += " <sk_model_ attribute used>"
         return output
 
+    @property
+    def _verbose_level(self):
+        """The current `verbose` setting as a `logger.level_enum`"""
+        return logger._verbose_to_level(self.verbose)
+
     @classmethod
     def _get_param_names(cls):
         """
@@ -267,20 +216,7 @@ def get_params(self, deep=True):
         need anything other than what is there in this method, then it doesn't
         have to override this method
         """
-        params = dict()
-        variables = self._get_param_names()
-        for key in variables:
-            var_value = getattr(self, key, None)
-            # We are currently internal to the cuML API, but the value we
-            # return will immediately be returned external to the API, so we
-            # must perform the translation from enum to integer before
-            # returning the value. Ordinarily, this is handled by
-            # VerbosityDescriptor for direct access to the verbose
-            # attribute.
-            if key == "verbose":
-                var_value = logger._verbose_from_level(var_value)
-            params[key] = var_value
-        return params
+        return {name: getattr(self, name) for name in self._get_param_names()}
 
     def set_params(self, **params):
         """
@@ -291,15 +227,13 @@ def set_params(self, **params):
         """
         if not params:
             return self
-        variables = self._get_param_names()
+        valid_params = self._get_param_names()
         for key, value in params.items():
-            if key not in variables:
-                raise ValueError("Bad param '%s' passed to set_params" % key)
-            else:
-                # Switch verbose to enum since we are now internal to cuML API
-                if key == "verbose":
-                    value = logger._verbose_to_level(value)
-                setattr(self, key, value)
+            if key not in valid_params:
+                raise ValueError(
+                    f"Invalid parameter {key!r} for `{type(self).__name__}`"
+                )
+            setattr(self, key, value)
         return self
 
     def _set_output_type(self, inp):

python/cuml/cuml/internals/logger.pyx

@@ -34,7 +34,7 @@ def _verbose_to_level(verbose: bool | int) -> level_enum:
     elif verbose is False:
         return level_enum.info
     else:
-        return level_enum(max(6 - verbose, 0))
+        return level_enum(min(max(6 - verbose, 0), 6))
 
 
 cdef class LogLevelSetter:

python/cuml/cuml/linear_model/elastic_net.py

@@ -265,7 +265,7 @@ def fit(
                 max_iter=self.max_iter,
                 tol=self.tol,
                 penalty_normalized=False,
-                verbose=self.verbose,
+                verbose=self._verbose_level,
                 handle=self.handle,
             )
             coef = CumlArray(data=coef.to_output("cupy").flatten())

python/cuml/cuml/linear_model/logistic_regression.py

@@ -312,7 +312,7 @@ def fit(
             max_iter=self.max_iter,
             tol=self.tol,
             linesearch_max_iter=self.linesearch_max_iter,
-            verbose=self.verbose,
+            verbose=self._verbose_level,
             handle=self.handle,
             lbfgs_memory=self.lbfgs_memory,
             penalty_normalized=self.penalty_normalized,

python/cuml/cuml/linear_model/logistic_regression_mg.pyx

@@ -186,7 +186,7 @@ class LogisticRegressionMG(MGFitMixin, LogisticRegression):
             linesearch_max_iter=self.linesearch_max_iter,
             lbfgs_memory=self.lbfgs_memory,
             penalty_normalized=self.penalty_normalized,
-            verbose=self.verbose,
+            verbose=self._verbose_level,
         )
 
         # Allocate outputs

python/cuml/cuml/manifold/t_sne.pyx

@@ -253,7 +253,7 @@ cdef _init_params(self, int n_samples, TSNEParams &params):
     params.pre_momentum = pre_momentum
     params.post_momentum = post_momentum
     params.random_state = seed
-    params.verbosity = self.verbose
+    params.verbosity = self._verbose_level
     params.square_distances = self.square_distances
     params.algorithm = algo
     params.init = init