Division by zero in HoeffdingAdaptivetree with small memory estimation period

[jsvobo]

river version:
river 0.22.0 (installed with pip)
Python version:
python 3.10
Operating system:
Debian

Describe the bug

When the memory estimation periodis small (can be forced by seting memory_estimate_period=1), and having large ADWIN Boosting classifier number of models (here I had 50), We can randomly run into this error:

File ~/<my env>/lib/python3.10/site-packages/river/ensemble/boosting.py:176, in ADWINBoostingClassifier.learn_one(self, x, y, **kwargs)
    174 for i, model in enumerate(self):
    175     for _ in range(utils.random.poisson(1, self._rng)):
--> [176]  model.learn_one(x, y, **kwargs)
    178     if model.predict_one(x) == y:
    179         self.correct_weight[i] += lambda_poisson

File ~/<my env>/lib/python3.10/site-packages/river/tree/hoeffding_adaptive_tree_classifier.py:234, in HoeffdingAdaptiveTreeClassifier.learn_one(self, x, y, w)
    231 self._root.learn_one(x, y, w=w, tree=self)
    233 if self._train_weight_seen_by_model % self.memory_estimate_period == 0:
--> [234]   self._estimate_model_size()

File ~/<my env>/lib/python3.10/site-packages/river/tree/hoeffding_tree.py:300, in HoeffdingTree._estimate_model_size(self)
    298         total_inactive_size += calculate_object_size(leaf)
    299 if total_active_size > 0:
--> [300]  self._active_leaf_size_estimate = total_active_size / self._n_active_leaves
    301 if total_inactive_size > 0:
    302     self._inactive_leaf_size_estimate = total_inactive_size / self._n_inactive_leaves

Code snippet:

tree.HoeffdingAdaptiveTreeClassifier(grace_period=100,max_size=10,max_depth=15,memory_estimate_period=10000)
model = ensemble.ADWINBoostingClassifier(model = tree_Hoeff,n_models=50, seed=seed)

After this init, i trained learn_one() on my data on dataset that i know works with other models.

Comment:

We should brace this code : "self._active_leaf_size_estimate = total_active_size / self._n_active_leaves" on line 300 in "river/tree/hoeffding_tree.py" by checking is there are any active leaves first.