Fix BanditClassifier (online-ml#1262)

river/bandit/thompson.py

@@ -29,7 +29,7 @@ class ThompsonSampling(bandit.base.Policy):
 
  Parameters
  ----------
- dist
+ reward_obj
  A distribution to sample from.
  burn_in
  The number of steps to use for the burn-in phase. Each arm is given the chance to be pulled
@@ -51,7 +51,7 @@ class ThompsonSampling(bandit.base.Policy):
  >>> _ = env.reset(seed=42)
  >>> _ = env.action_space.seed(123)
 
- >>> policy = bandit.ThompsonSampling(dist=proba.Beta(), seed=101)
+ >>> policy = bandit.ThompsonSampling(reward_obj=proba.Beta(), seed=101)
 
  >>> metric = stats.Sum()
  >>> while True:
@@ -71,22 +71,20 @@ class ThompsonSampling(bandit.base.Policy):
 
  """
 
- def __init__(self, dist: proba.base.Distribution, burn_in=0, seed: int | None = None):
- super().__init__(reward_obj=dist, burn_in=burn_in)
+ def __init__(
+ self, reward_obj: proba.base.Distribution = None, burn_in=0, seed: int | None = None
+ ):
+ super().__init__(reward_obj=reward_obj, burn_in=burn_in)
  self.seed = seed
  self._rng = random.Random(seed)
- self._rewards.default_factory = self._clone_dist_with_seed
+ self._rewards.default_factory = self._clone_reward_obj_with_seed
 
- def _clone_dist_with_seed(self):
- return self.dist.clone({"seed": self._rng.randint(0, 2**32)})
-
- @property
- def dist(self):
- return self.reward_obj
+ def _clone_reward_obj_with_seed(self):
+ return self.reward_obj.clone({"seed": self._rng.randint(0, 2**32)})
 
  def _pull(self, arm_ids):
  return max(arm_ids, key=lambda arm_id: self._rewards[arm_id].sample())
 
  @classmethod
  def _unit_test_params(cls):
- yield {"dist": proba.Beta()}
+ yield {"reward_obj": proba.Beta()}

river/model_selection/bandit.py

@@ -206,7 +206,7 @@ def learn_one(self, x, y):
  y_pred = (
  model.predict_one(x) if self.metric.requires_labels else model.predict_proba_one(x)
  )
- self.policy.update(arm_id, y_true=y, y_pred=y_pred)
+ self.policy.update(arm_id, y, y_pred)
  model.learn_one(x, y)
 
  return self

river/model_selection/test_bandit.py

@@ -0,0 +1,87 @@
+from __future__ import annotations
+
+import importlib
+import inspect
+
+import pytest
+
+from river import (
+ bandit,
+ datasets,
+ evaluate,
+ linear_model,
+ metrics,
+ model_selection,
+ optim,
+ preprocessing,
+)
+
+
+def test_1259():
+ """
+
+ https://github.com/online-ml/river/issues/1259
+
+ >>> from river import bandit
+ >>> from river import datasets
+ >>> from river import evaluate
+ >>> from river import linear_model
+ >>> from river import metrics
+ >>> from river import model_selection
+ >>> from river import optim
+ >>> from river import preprocessing
+
+ >>> models = [
+ ... linear_model.LogisticRegression(optimizer=optim.SGD(lr=lr))
+ ... for lr in [0.0001, 0.001, 1e-05, 0.01]
+ ... ]
+
+ >>> dataset = datasets.Phishing()
+ >>> model = (
+ ... preprocessing.StandardScaler() |
+ ... model_selection.BanditClassifier(
+ ... models,
+ ... metric=metrics.Accuracy(),
+ ... policy=bandit.Exp3(
+ ... gamma=0.5,
+ ... seed=42
+ ... )
+ ... )
+ ... )
+ >>> metric = metrics.Accuracy()
+
+ >>> evaluate.progressive_val_score(dataset, model, metric)
+ Accuracy: 87.20%
+
+ """
+
+
+@pytest.mark.parametrize(
+ "policy",
+ [
+ pytest.param(
+ policy(**params),
+ id=f"{policy.__name__}",
+ )
+ for _, policy in inspect.getmembers(
+ importlib.import_module("river.bandit"),
+ lambda obj: inspect.isclass(obj) and issubclass(obj, bandit.base.Policy),
+ )
+ for params in policy._unit_test_params()
+ if policy.__name__ != "ThompsonSampling"
+ ],
+)
+def test_bandit_classifier_with_each_policy(policy):
+ models = [
+ linear_model.LogisticRegression(optimizer=optim.SGD(lr=lr))
+ for lr in [0.0001, 0.001, 1e-05, 0.01]
+ ]
+
+ dataset = datasets.Phishing()
+ model = preprocessing.StandardScaler() | model_selection.BanditClassifier(
+ models, metric=metrics.Accuracy(), policy=policy
+ )
+ metric = metrics.Accuracy()
+
+ score = evaluate.progressive_val_score(dataset, model, metric)
+ assert score.get() > 0.5