[tmva][sofie] Improve Keras test robustness and diagnostics

  * Skip training for models without trainable weights to avoid spurious
    Keras warnings about "no trainable weights"

  * Suppress NumPy 2.0 __array__ DeprecationWarning during model.save()
    (upstream TF/Keras compatibility issue) using a scoped warnings context

  * Fix Keras inference call for single-input models by unwrapping
    singleton input lists (Keras expects a tensor, not [tensor])

  * Remove redundant load_weights() after load_model()

  * Replace custom is_accurate() with np.testing.assert_allclose()
    for clearer diagnostics and standardized comparison
    * use atol=1e-2 and rtol=0 to preserve previous absolute tolerance behavior

  * Update LeakyReLU argument (alpha -> negative_slope) to avoid
    deprecation warnings

These changes make the tests more robust, reduce noise from external
dependencies, and significantly improve failure diagnostics.

Author: guitargeek

bindings/pyroot/pythonizations/test/generate_keras_functional.py

@@ -1,6 +1,9 @@
+import warnings
+
 def generate_keras_functional(dst_dir):
 
     import numpy as np
+    import keras
     from keras import layers, models
     from parser_test_function import is_channels_first_supported
 
@@ -16,8 +19,14 @@ def train_and_save(model, name):
 
         model.compile(optimizer='adam', loss='mean_squared_error', metrics=['mae'])
         model.summary()
-        model.fit(x_train, y_train, epochs=1, verbose=0)
-        model.save(f"{dst_dir}/Functional_{name}_test.keras")
+        if len(model.trainable_weights) > 0:
+            model.fit(x_train, y_train, epochs=1, verbose=0)
+
+        with warnings.catch_warnings():
+            # Some object inside TensorFlow/Keras has an outdated __array__ implementation
+            warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*__array__.*copy keyword.*")
+            model.save(f"{dst_dir}/Functional_{name}_test.keras")
+
         print("generated and saved functional model",name)
 
 
@@ -211,7 +220,11 @@ def train_and_save(model, name):
     sub = layers.Subtract()([d1, d2])
     mul = layers.Multiply()([d1, d2])
     merged = layers.Concatenate()([add, sub, mul])
-    merged = layers.LeakyReLU(alpha=0.1)(merged)
+    # `alpha` was renamed to `negative_slope` in Keras 3
+    if keras.__version__ >= "3.0":
+        merged = layers.LeakyReLU(negative_slope=0.1)(merged)
+    else:
+        merged = layers.LeakyReLU(alpha=0.1)(merged)
     out = layers.Dense(4, activation="softmax")(merged)
     model = models.Model([inp1, inp2], out)
     train_and_save(model, "Layer_Combination_3")

bindings/pyroot/pythonizations/test/generate_keras_sequential.py

@@ -1,5 +1,8 @@
+import warnings
+
 def generate_keras_sequential(dst_dir):
 
+    import keras
     import numpy as np
     from keras import layers, models
     from parser_test_function import is_channels_first_supported
@@ -9,10 +12,15 @@ def train_and_save(model, name):
         x_train = np.random.rand(32, *model.input_shape[1:])
         y_train = np.random.rand(32, *model.output_shape[1:])
         model.compile(optimizer='adam', loss='mean_squared_error', metrics=['mae'])
-        model.fit(x_train, y_train, epochs=1, verbose=0)
+        if len(model.trainable_weights) > 0:
+            model.fit(x_train, y_train, epochs=1, verbose=0)
         model.summary()
         print("fitting sequential model",name)
-        model.save(f"{dst_dir}/Sequential_{name}_test.keras")
+
+        with warnings.catch_warnings():
+            # Some object inside TensorFlow/Keras has an outdated __array__ implementation
+            warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*__array__.*copy keyword.*")
+            model.save(f"{dst_dir}/Sequential_{name}_test.keras")
 
 
     # Binary Ops: Add, Subtract, Multiply are not typical in Sequential - skipping those
@@ -183,6 +191,9 @@ def train_and_save(model, name):
     ])
     train_and_save(modelA, "Layer_Combination_1")
 
+    # `alpha` was renamed to `negative_slope` in Keras 3
+    negative_slope_key = "negative_slope" if keras.__version__ >= "3.0" else "alpha"
+
     modelB = models.Sequential([
         layers.Input(shape=(32,32,3)),
         layers.Conv2D(8, (3,3), padding='valid', data_format='channels_last', activation='relu'),
@@ -193,7 +204,7 @@ def train_and_save(model, name):
         layers.Permute((2, 1)),
         layers.Flatten(),
         layers.Dense(32),
-        layers.LeakyReLU(alpha=0.1),
+        layers.LeakyReLU(**{negative_slope_key : 0.1}),
         layers.Dense(10, activation='softmax'),
     ])
     train_and_save(modelB, "Layer_Combination_2")
@@ -210,4 +221,4 @@ def train_and_save(model, name):
         layers.Dense(8, activation='swish'),
         layers.Dense(3, activation='softmax'),
     ])
-    train_and_save(modelC, "Layer_Combination_3")
+    train_and_save(modelC, "Layer_Combination_3")

bindings/pyroot/pythonizations/test/parser_test_function.py

@@ -2,10 +2,6 @@
 
 '''
 The test file contains two types of functions:
-    is_accurate:
-        - This function checks whether the inference results from SOFIE and Keras are accurate within a specified
-          tolerance. Since the inference result from Keras is not flattened, the function flattens both tensors before
-          performing the comparison.
 
     generate_and_test_inference:
         - This function accepts the following inputs:
@@ -29,7 +25,7 @@
           shape from the model object.
         - Convert the inference results to NumPy arrays:
           The SOFIE result is of type vector<float>, and the Keras result is a TensorFlow tensor. Both are converted to
-          NumPy arrays before being passed to the is_accurate function for comparison.
+          NumPy arrays before being passed to the np.testing.assert_allclose function for comparison.
 
 '''
 def is_channels_first_supported() :
@@ -42,16 +38,6 @@ def is_channels_first_supported() :
 
       return True
 
-def is_accurate(tensor_a, tensor_b, tolerance=1e-2):
-    tensor_a = tensor_a.flatten()
-    tensor_b = tensor_b.flatten()
-    for i in range(len(tensor_a)):
-        difference = abs(tensor_a[i] - tensor_b[i])
-        if difference > tolerance:
-            print(tensor_a[i], tensor_b[i])
-            return False
-    return True
-
 def generate_and_test_inference(model_file_path: str, generated_header_file_dir: str = None, batch_size=1):
 
     import keras
@@ -81,7 +67,6 @@ def generate_and_test_inference(model_file_path: str, generated_header_file_dir:
     sofie_model_namespace = getattr(ROOT, "TMVA_SOFIE_" + model_name)
     inference_session = sofie_model_namespace.Session(generated_header_file_path.removesuffix(".hxx") + ".dat")
     keras_model = keras.models.load_model(model_file_path)
-    keras_model.load_weights(model_file_path)
 
     input_tensors = []
     for model_input in keras_model.inputs:
@@ -91,11 +76,17 @@ def generate_and_test_inference(model_file_path: str, generated_header_file_dir:
     sofie_inference_result = inference_session.infer(*input_tensors)
     sofie_output_tensor_shape = list(rmodel.GetTensorShape(rmodel.GetOutputTensorNames()[0]))   # get output shape
                                                                                                 # from SOFIE
-    keras_inference_result = keras_model(input_tensors)
+    # Keras explicitly forbids input tensor lists of size 1
+    if len(keras_model.inputs) == 1:
+        keras_inference_result = keras_model(input_tensors[0])
+    else:
+        keras_inference_result = keras_model(input_tensors)
     if sofie_output_tensor_shape != list(keras_inference_result.shape):
         raise AssertionError("Output tensor dimensions from SOFIE and Keras do not match")
-    sofie_inference_result = np.asarray(sofie_inference_result)
-    keras_inference_result = np.asarray(keras_inference_result)
-    is_inference_accurate = is_accurate(sofie_inference_result, keras_inference_result)
-    if not is_inference_accurate:
-        raise AssertionError("Inference results from SOFIE and Keras do not match")
+
+    np.testing.assert_allclose(
+        np.asarray(sofie_inference_result).flatten(),
+        np.asarray(keras_inference_result).flatten(),
+        atol=1e-2,
+        rtol=0.  # explicitly disable relative tolerance (NumPy uses |a - b| <= atol + rtol * |b|)
+    )