Added univariate streaming algorithms

pyproject.toml

@@ -8,7 +8,6 @@ license = { text = "Apache-2.0" }
 readme = "README.md"
 
 dependencies = [
-    "river>=0.22.0",
     "scipy>=1.15.3",
 ]
 
@@ -23,12 +22,14 @@ dependencies = [
 
 sklearn = ["scikit-learn>=1.6.1"]
 
-huggingface = ["scipy>=1.15.2", "transformers[torch]>=4.52.3"]
+huggingface = ["transformers[torch]>=4.52.3"]
 
 polars = ["polars>=1.31.0"]
 
 pandas = ["pandas>=2.2.3"]
 
+river = ["river>=0.22.0"]
+
 
 # -------------------------------------------------
 

src/ml3_drift/analysis/analyzer/base.py

@@ -160,7 +160,7 @@ def analyze(
         else:
             categorical_columns_ids = []
 
-        if len(continuous_columns_ids) == 0 and len(categorical_columns_ids) == 0:
+        if not continuous_columns_ids and not categorical_columns_ids:
             raise ValueError(
                 "At least one of continuous_columns or categorical_columns must be provided."
             )

src/ml3_drift/monitoring/algorithms/online/adwin.py

@@ -11,7 +11,12 @@
 from ml3_drift.monitoring.base.online_monitorning_algorithm import (
     OnlineMonitorningAlgorithm,
 )
-from river.drift.adwin import ADWIN as RiverADWIN
+
+RIVER = True
+try:
+    from river.drift.adwin import ADWIN as RiverADWIN
+except ModuleNotFoundError:
+    RIVER = False
 
 
 class ADWIN(OnlineMonitorningAlgorithm, UnivariateMonitoringAlgorithm):
@@ -31,7 +36,13 @@ def __init__(
         max_buckets: int = 5,
         min_window_length: int = 5,
         grace_period: int = 10,
+        *args,
+        **kwargs,
     ) -> None:
+        if not RIVER:
+            raise ModuleNotFoundError(
+                "River library is required for ADWIN algorithm. Please install it using pip install/ uv add ml3-drift[river]"
+            )
         self.p_value = p_value
         self.clock = clock
         self.max_buckets = max_buckets

src/ml3_drift/monitoring/algorithms/online/kswin.py

@@ -11,7 +11,12 @@
 from ml3_drift.monitoring.base.online_monitorning_algorithm import (
     OnlineMonitorningAlgorithm,
 )
-from river.drift.kswin import KSWIN as RiverKSWIN
+
+RIVER = True
+try:
+    from river.drift.kswin import KSWIN as RiverKSWIN
+except ModuleNotFoundError:
+    RIVER = False
 
 
 class KSWIN(OnlineMonitorningAlgorithm, UnivariateMonitoringAlgorithm):
@@ -30,7 +35,13 @@ def __init__(
         window_size: int = 100,
         stat_size: int = 30,
         seed: int | None = None,
+        *args,
+        **kwargs,
     ) -> None:
+        if not RIVER:
+            raise ModuleNotFoundError(
+                "River library is required for KSWIN algorithm. Please install it using pip install/ uv add ml3-drift[river]"
+            )
         self.p_value = p_value
         self.window_size = window_size
         self.stat_size = stat_size

src/ml3_drift/monitoring/algorithms/online/page_hinkley.py

@@ -12,7 +12,12 @@
 from ml3_drift.monitoring.base.online_monitorning_algorithm import (
     OnlineMonitorningAlgorithm,
 )
-from river.drift.page_hinkley import PageHinkley as RiverPageHinkley
+
+RIVER = True
+try:
+    from river.drift.page_hinkley import PageHinkley as RiverPageHinkley
+except ModuleNotFoundError:
+    RIVER = False
 
 
 class PageHinkley(OnlineMonitorningAlgorithm, UnivariateMonitoringAlgorithm):
@@ -33,7 +38,13 @@ def __init__(
         threshold: int = 50,
         mode: str = "both",
         seed: int | None = None,
+        *args,
+        **kwargs,
     ) -> None:
+        if not RIVER:
+            raise ModuleNotFoundError(
+                "River library is required for PageHinkley algorithm. Please install it using pip install/ uv add ml3-drift[river]"
+            )
         self.min_instances = min_instances
         self.delta = delta
         self.alpha = alpha

tests/test_analyzer/test_analyzer.py

@@ -109,7 +109,7 @@ def test_batch_analyzer_null_columns():
         n_samples=n_samples,
         n_cont=n_cont,
         n_cat=n_cat,
-        data_format="pandas",
+        data_format="numpy",
         seed=2,
     )
 

tests/test_monitoring/test_online/test_univariate.py

@@ -1,8 +1,18 @@
+import pytest
+from tests.conftest import is_module_available
 import numpy as np
 from ml3_drift.monitoring.algorithms.online.adwin import ADWIN
 from ml3_drift.monitoring.algorithms.online.kswin import KSWIN
 from ml3_drift.monitoring.algorithms.online.page_hinkley import PageHinkley
 
+if not is_module_available("river"):
+    with pytest.raises(ModuleNotFoundError):
+        ADWIN()
+        KSWIN()
+        PageHinkley()
+    # Prevent tests from running if sklearn is not available
+    pytest.skip(allow_module_level=True)
+
 
 class TestContinuousUnivariateOnlineAlgorithms:
     def test_kswin_univariate_one_drift(self, abrupt_univariate_online_drift_info):