Fixes #1248 (#1337)

* bug fix

* update affected test

docs/releases/unreleased.md

@@ -11,11 +11,10 @@ River's mini-batch methods now support pandas v2. In particular, River conforms
 ## clustering
 
 - Add fixes to `cluster.DBSTREAM` algorithm, including:
-  - Addition of the `-` sign before the `fading_factor` in accordance with the algorithm 2 proposed by Hashler and Bolanos (2016) to allow clusters with low weights to be removed. 
-  - The new `micro_cluster` is added with the key derived from the maximum key of the existing micro clusters. If the set of micro clusters is still empty (`len = 0`), a new micro cluster is added with key 0. 
+  - Addition of the `-` sign before the `fading_factor` in accordance with the algorithm 2 proposed by Hashler and Bolanos (2016) to allow clusters with low weights to be removed.
+  - The new `micro_cluster` is added with the key derived from the maximum key of the existing micro clusters. If the set of micro clusters is still empty (`len = 0`), a new micro cluster is added with key 0.
   - `cluster_is_up_to_date` is set to `True` at the end of the `self._recluster()` function.
 
-
 ## datasets
 
 - Added `datasets.WebTraffic`, which is a dataset that counts the occurrences of events on a website. It is a multi-output regression dataset with two outputs.
@@ -31,3 +30,7 @@ River's mini-batch methods now support pandas v2. In particular, River conforms
 ## proba
 
 - Added `_from_state` method to `proba.MultivariateGaussian` to warm start from previous knowledge.
+
+## tree
+
+- Fix a bug in `tree.splitter.NominalSplitterClassif` that generated a mismatch between the number of existing tree branches and the number of tracked branches.

river/ensemble/streaming_random_patches.py

@@ -409,7 +409,7 @@ class SRPClassifier(BaseSRPEnsemble, base.Classifier):
     >>> metric = metrics.Accuracy()
 
     >>> evaluate.progressive_val_score(dataset, model, metric)
-    Accuracy: 72.77%
+    Accuracy: 71.97%
 
     Notes
     -----

river/tree/extremely_fast_decision_tree.py

@@ -432,7 +432,7 @@ def _reevaluate_best_split(self, node, parent, branch_index, **kwargs):
 
                     # update EFDT
                     if parent is None:
-                        # Root case : replace the root node by a new split node
+                        # Root case : replace the root node by the new node
                         self._root = best_split
                     else:
                         parent.children[branch_index] = best_split

river/tree/splitter/nominal_splitter_classif.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 import collections
-import functools
 
 from ..utils import BranchFactory
 from .base import Splitter
@@ -18,9 +17,7 @@ def __init__(self):
         super().__init__()
         self._total_weight_observed = 0.0
         self._missing_weight_observed = 0.0
-        self._att_dist_per_class = collections.defaultdict(
-            functools.partial(collections.defaultdict, float)
-        )
+        self._att_dist_per_class = collections.defaultdict(dict)
         self._att_values = set()
 
     @property
@@ -32,12 +29,20 @@ def update(self, att_val, target_val, sample_weight):
             self._missing_weight_observed += sample_weight
         else:
             self._att_values.add(att_val)
-            self._att_dist_per_class[target_val][att_val] += sample_weight
+
+            try:
+                self._att_dist_per_class[target_val][att_val] += sample_weight
+            except KeyError:
+                self._att_dist_per_class[target_val][att_val] = sample_weight
 
         self._total_weight_observed += sample_weight
 
     def cond_proba(self, att_val, target_val):
         class_dist = self._att_dist_per_class[target_val]
+
+        if att_val not in class_dist:
+            return 0.0
+
         value = class_dist[att_val]
         try:
             return value / sum(class_dist.values())