Update to use RandomState to avoid compatibility issues with train_te…

…st_split

pipeline_lib/core/pipeline.py

@@ -8,7 +8,7 @@
 
 from pipeline_lib.core.data_container import DataContainer
 from pipeline_lib.core.model_registry import ModelRegistry
-from pipeline_lib.core.random_generator import initialize_generator
+from pipeline_lib.core.random_state_generator import initialize_random_state
 from pipeline_lib.core.step_registry import StepRegistry
 from pipeline_lib.core.steps import PipelineStep
 
@@ -89,7 +89,7 @@ def from_json(cls, path: str) -> Pipeline:
             seed = config["pipeline"]["seed"]
         else:
             seed = 42
-        initialize_generator(seed)
+        initialize_random_state(seed)
 
         steps = []
 

pipeline_lib/core/random_state_generator.py

@@ -0,0 +1,28 @@
+from typing import Optional
+
+import numpy as np
+from numpy.random import RandomState
+
+_random_state = None
+
+
+def get_random_state() -> Optional[RandomState]:
+    """
+    Get the global random state object.
+
+    Returns:
+        RandomState or None: The global random state object if initialized, else None.
+    """
+    global _random_state
+    return _random_state
+
+
+def initialize_random_state(seed: int):
+    """
+    Initialize the global random state object with the provided seed.
+
+    Args:
+        seed (int): The seed value to initialize the random state object.
+    """
+    global _random_state
+    _random_state = np.random.RandomState(seed)

pipeline_lib/core/steps/explainer_dashboard.py

@@ -1,7 +1,7 @@
 from explainerdashboard import RegressionExplainer
 
 from pipeline_lib.core import DataContainer
-from pipeline_lib.core.random_generator import get_random_generator
+from pipeline_lib.core.random_state_generator import get_random_state
 from pipeline_lib.core.steps.base import PipelineStep
 
 
@@ -38,7 +38,7 @@ def execute(self, data: DataContainer) -> DataContainer:
                 f" {self.max_samples}."
             )
             self.logger.info(f"Sampling {self.max_samples} data points from the dataset.")
-            df = df.sample(n=self.max_samples, random_state=get_random_generator().integers(0, 100))
+            df = df.sample(n=self.max_samples, random_state=get_random_state())
 
         drop_columns = data._drop_columns
         if drop_columns:

pipeline_lib/core/steps/tabular_split.py

@@ -3,7 +3,7 @@
 from sklearn.model_selection import train_test_split
 
 from pipeline_lib.core import DataContainer
-from pipeline_lib.core.random_generator import get_random_generator
+from pipeline_lib.core.random_state_generator import get_random_state
 from pipeline_lib.core.steps.base import PipelineStep
 
 
@@ -61,19 +61,19 @@ def execute(self, data: DataContainer) -> DataContainer:
             train_val_df, test_df = train_test_split(
                 df,
                 test_size=self.test_percentage,
-                random_state=get_random_generator().integers(0, 100),
+                random_state=get_random_state(),
             )
             train_df, validation_df = train_test_split(
                 train_val_df,
                 train_size=self.train_percentage
                 / (self.train_percentage + self.validation_percentage),
-                random_state=get_random_generator().integers(0, 100),
+                random_state=get_random_state(),
             )
         else:
             train_df, validation_df = train_test_split(
                 df,
                 train_size=self.train_percentage,
-                random_state=get_random_generator().integers(0, 100),
+                random_state=get_random_state(),
             )
             test_df = None