Fix keras functional 759

.github/workflows/python-package-conda.yml

@@ -29,7 +29,8 @@ jobs:
       - name: Install test dependencies (conda)
         shell: bash -l {0}
         run: |
-          conda install -y pytest coverage
+          # Pinning numpy and scikit-learn to prevent the issues reported in #982
+          conda install -y pytest coverage "numpy>=1.22,<2.4" "scikit-learn>=1.0" "numba>=0.59.0"
 
       - name: Install package and run tests
         shell: bash -l {0}

.github/workflows/python-package-pip.yml

@@ -24,6 +24,8 @@ jobs:
       - name: Install dependencies (pip)
         run: |
           python -m pip install --upgrade pip
+          # Fixing #982 by pinning numpy to a version compatible with numba
+          pip install "numpy>=1.22,<2.4" 
           pip install pytest coverage
           pip install -e .
 

environment.yml

@@ -3,7 +3,7 @@ channels:
   - conda-forge
 dependencies:
   - flake8>=4.0.1
-  - numpy>=2.3.5
+  - numpy>=1.21.0,<2.4.0
   - pandas>=2.3.3
   - scipy>=1.16.3
   - scikit-learn>=1.8.0
@@ -13,4 +13,4 @@ dependencies:
   - pytest>=6.2.5
   - setuptools>=59.4.0
   - pip:
-      - markdown>=3.3.6
+      - markdown>=3.3.6

mlxtend/evaluate/bias_variance_decomp.py

@@ -25,7 +25,7 @@ def bias_variance_decomp(
     loss="0-1_loss",
     num_rounds=200,
     random_seed=None,
-    **fit_params
+    **fit_params,
 ):
     """
     estimator : object
@@ -106,8 +106,7 @@ def bias_variance_decomp(
         X_boot, y_boot = _draw_bootstrap_sample(rng, X_train, y_train)
 
         # Keras support
-        if estimator.__class__.__name__ in ["Sequential", "Functional"]:
-            # reset model
+        if estimator.__class__.__name__ in ["Sequential", "Functional", "Model"]:
             for ix, layer in enumerate(estimator.layers):
                 if hasattr(estimator.layers[ix], "kernel_initializer") and hasattr(
                     estimator.layers[ix], "bias_initializer"
@@ -128,6 +127,7 @@ def bias_variance_decomp(
             pred = estimator.predict(X_test).reshape(1, -1)
         else:
             pred = estimator.fit(X_boot, y_boot, **fit_params).predict(X_test)
+
         all_pred[i] = pred
 
     if loss == "0-1_loss":