Fix temporary directory race condition in sklearn estimators (#529)

Caution : for the tests, MPI must be disabled otherwise a segfault would occur because of a huge use of resources

Author: tramora

CHANGELOG.md

@@ -11,6 +11,7 @@
 ### Fixed
 - (General) Automatic Credentials Discovery-based credential retrieval for
   Google cloud storage (GCS).
+- (`sklearn`) Temporary directory race condition in estimators.
 
 ## 11.0.0.0 - 2025-12-19
 

khiops/sklearn/estimators.py

@@ -362,8 +362,6 @@ def fit(self, X, y=None, **kwargs):
 
         # Create temporary directory and tables
         computation_dir = self._create_computation_dir("fit")
-        initial_runner_temp_dir = kh.get_runner().root_temp_dir
-        kh.get_runner().root_temp_dir = computation_dir
 
         # Create the dataset, fit the model and reset in case of any failure
         try:
@@ -377,7 +375,6 @@ def fit(self, X, y=None, **kwargs):
         # Cleanup and restore the runner's temporary dir
         finally:
             self._cleanup_computation_dir(computation_dir)
-            kh.get_runner().root_temp_dir = initial_runner_temp_dir
 
         # If on "fitted" state then:
         # - self.model_ must be a DictionaryDomain
@@ -963,15 +960,13 @@ def _simplify(
         computation_dir = self._create_computation_dir("simplify")
         output_dir = self._get_output_dir(computation_dir)
         simplify_log_file_path = fs.get_child_path(output_dir, "khiops_simplify_cc.log")
-        initial_runner_temp_dir = kh.get_runner().root_temp_dir
         full_coclustering_file_path = fs.get_child_path(
             output_dir, "FullCoclustering.khcj"
         )
         simplified_coclustering_file_path = fs.get_child_path(
             output_dir, "Coclustering.khcj"
         )
         self.model_report_.write_khiops_json_file(full_coclustering_file_path)
-        kh.get_runner().root_temp_dir = computation_dir
         try:
             # - simplify_coclustering, then
             # - prepare_coclustering_deployment
@@ -1040,7 +1035,6 @@ def _simplify(
             )
         finally:
             self._cleanup_computation_dir(computation_dir)
-            kh.get_runner().root_temp_dir = initial_runner_temp_dir
         return simplified_cc
 
     def simplify(
@@ -1101,8 +1095,6 @@ def predict(self, X):
         """
         # Create temporary directory
         computation_dir = self._create_computation_dir("predict")
-        initial_runner_temp_dir = kh.get_runner().root_temp_dir
-        kh.get_runner().root_temp_dir = computation_dir
 
         # Create the input dataset
         ds = Dataset(X)
@@ -1119,7 +1111,6 @@ def predict(self, X):
         # Cleanup and restore the runner's temporary dir
         finally:
             self._cleanup_computation_dir(computation_dir)
-            kh.get_runner().root_temp_dir = initial_runner_temp_dir
 
         # Transform to numpy.array
         y_pred = y_pred.to_numpy()
@@ -1557,8 +1548,6 @@ def predict(self, X):
         """
         # Create temporary directory
         computation_dir = self._create_computation_dir("predict")
-        initial_runner_temp_dir = kh.get_runner().root_temp_dir
-        kh.get_runner().root_temp_dir = computation_dir
 
         try:
             # Create the input dataset
@@ -1575,10 +1564,6 @@ def predict(self, X):
         # Cleanup and restore the runner's temporary dir
         finally:
             self._cleanup_computation_dir(computation_dir)
-            kh.get_runner().root_temp_dir = initial_runner_temp_dir
-
-        # Restore the runner's temporary dir
-        kh.get_runner().root_temp_dir = initial_runner_temp_dir
 
         # Return pd.Series in the monotable + pandas case
         assert isinstance(y_pred, (str, pd.DataFrame)), "Expected str or DataFrame"
@@ -1997,8 +1982,6 @@ def predict_proba(self, X):
         """
         # Create temporary directory and tables
         computation_dir = self._create_computation_dir("predict_proba")
-        initial_runner_temp_dir = kh.get_runner().root_temp_dir
-        kh.get_runner().root_temp_dir = computation_dir
 
         # Create the input dataset
 
@@ -2015,7 +1998,6 @@ def predict_proba(self, X):
         # Cleanup and restore the runner's temporary dir
         finally:
             self._cleanup_computation_dir(computation_dir)
-            kh.get_runner().root_temp_dir = initial_runner_temp_dir
 
         # - Reorder the columns to that of self.classes_
         # - Transform to np.ndarray
@@ -2649,8 +2631,6 @@ def transform(self, X):
         """
         # Create temporary directory
         computation_dir = self._create_computation_dir("transform")
-        initial_runner_temp_dir = kh.get_runner().root_temp_dir
-        kh.get_runner().root_temp_dir = computation_dir
 
         # Create and transform the dataset
         try:
@@ -2665,7 +2645,6 @@ def transform(self, X):
         # Cleanup and restore the runner's temporary dir
         finally:
             self._cleanup_computation_dir(computation_dir)
-            kh.get_runner().root_temp_dir = initial_runner_temp_dir
         return X_transformed.to_numpy(copy=False)
 
     def _transform_prepare_deployment_for_transform(self, ds):

tests/test_sklearn.py

@@ -6,18 +6,24 @@
 ######################################################################################
 """Tests parameter transfer between Khiops sklearn and core APIs"""
 import contextlib
+import io
 import os
 import shutil
 import unittest
 import warnings
+from concurrent.futures import as_completed
+from concurrent.futures.thread import ThreadPoolExecutor
 from itertools import product
 
 import numpy as np
+import pandas as pd
 from sklearn.exceptions import NotFittedError
 from sklearn.utils.estimator_checks import check_estimator
 from sklearn.utils.validation import check_is_fitted
 
 import khiops.core as kh
+import khiops.core.internals.filesystems as fs
+from khiops.core.internals.runner import KhiopsLocalRunner
 from khiops.sklearn.estimators import (
     KhiopsClassifier,
     KhiopsCoclustering,
@@ -1773,6 +1779,27 @@ def test_sklearn_check_estimator(self):
                     print("Done")
 
 
+def no_mpi(func):
+    """Disable MPI (in setting proc number to 1) to save resources"""
+
+    def inner(self):
+        # set up a single cpu runner
+        initial_runner = kh.get_runner()
+        initial_proc_number = os.environ["KHIOPS_PROC_NUMBER"]
+        os.environ["KHIOPS_PROC_NUMBER"] = "1"
+        single_cpu_runner = KhiopsLocalRunner()
+        kh.set_runner(single_cpu_runner)
+
+        # call the initial test function
+        func(self)
+
+        # restore the runner
+        os.environ["KHIOPS_PROC_NUMBER"] = initial_proc_number
+        kh.set_runner(initial_runner)
+
+    return inner
+
+
 class KhiopsSklearnVariousTests(unittest.TestCase):
     """Miscelanous sklearn classes tests"""
 
@@ -1827,3 +1854,89 @@ def test_export_operations_raise_when_not_fitted(self):
             with self.subTest(export_operation=export_operation, estimator=estimator):
                 with self.assertRaises(NotFittedError):
                     getattr(estimator, export_operation)("report.khj")
+
+    @no_mpi
+    def test_concurrency_safe_operations(self):
+        """Ensure no race condition occurs when running concurrent operations"""
+
+        # Define all the function calls that will be submitted to the threads
+        def predict_func(clf, X):
+            return clf.predict(X)
+
+        def predict_proba_func(clf, X):
+            return clf.predict_proba(X)
+
+        def encoder_fit_transform_func(khe, X, y):
+            return khe.fit_transform(X, y)
+
+        def estimator_fit_func(khcc, X, id_column):
+            return khcc.fit(X, id_column=id_column)
+
+        def coclustering_simplify_func(khcc):
+            return khcc.simplify()
+
+        def coclustering_predict_func(khcc, X):
+            return khcc.predict(X)
+
+        clf = KhiopsClassifier(n_trees=0)
+        adult_df = pd.read_csv(
+            f"{kh.get_samples_dir()}/Adult/Adult.txt", sep="\t", header=0
+        )
+        X = adult_df.drop("class", axis=1)
+        clf.fit(X, adult_df["class"])
+
+        # Test `predict`, `predict_proba` of `KhiopsPredictor` and its children
+        # (`KhiopsClassifier` and `KhiopsRegressor`)
+        with ThreadPoolExecutor(max_workers=5) as executor:
+            futures = {executor.submit(predict_func, clf, X): i for i in range(5)}
+            for future in as_completed(futures):
+                print(future.result())
+            futures = {executor.submit(predict_proba_func, clf, X): i for i in range(5)}
+            for future in as_completed(futures):
+                print(future.result())
+
+        # Test `transform` of `KhiopsEncoder`
+        khe = KhiopsEncoder()
+
+        y = adult_df["class"]
+        with ThreadPoolExecutor(max_workers=5) as executor:
+            futures = {
+                executor.submit(encoder_fit_transform_func, khe, X, y): i
+                for i in range(5)
+            }
+            for future in as_completed(futures):
+                print(future.result())
+
+        # Test `fit`, `simplify` and `predict` of
+        # `KhiopsCoclustering` and `KhiopsEstimator`
+        splice_data_dir = fs.get_child_path(
+            kh.get_runner().samples_dir, "SpliceJunction"
+        )
+        splice_data_file_path = fs.get_child_path(
+            splice_data_dir, "SpliceJunctionDNA.txt"
+        )
+
+        # Read the splice junction secondary datatable
+        with io.BytesIO(fs.read(splice_data_file_path)) as splice_data_file:
+            splice_df = pd.read_csv(splice_data_file, sep="\t")
+
+        khcc = KhiopsCoclustering()
+
+        with ThreadPoolExecutor(max_workers=5) as executor:
+            futures = {
+                executor.submit(estimator_fit_func, khcc, splice_df, "SampleId"): i
+                for i in range(5)
+            }
+            for future in as_completed(futures):
+                print(future.result())
+            futures = {
+                executor.submit(coclustering_simplify_func, khcc): i for i in range(5)
+            }
+            for future in as_completed(futures):
+                print(future.result())
+            futures = {
+                executor.submit(coclustering_predict_func, khcc, splice_df): i
+                for i in range(5)
+            }
+            for future in as_completed(futures):
+                print(future.result())