Switching test models to Keras3 format

.github/workflows/main.yml

@@ -15,7 +15,7 @@ jobs:
 
     strategy:
       matrix:
-        python-version: ["3.8", "3.9", "3.10"]
+        python-version: ["3.9", "3.10", "3.11", "3.12"]
 
     steps:
       - uses: "actions/checkout@v2"
@@ -49,4 +49,4 @@ jobs:
       - name: "Upload coverage to Codecov"
         uses: "codecov/codecov-action@v4"
         with:
-          fail_ci_if_error: true
+          fail_ci_if_error: true

.gitignore

@@ -127,3 +127,4 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+.vscode/settings.json

setup.cfg

@@ -79,7 +79,7 @@ testing =
     linear-tree
     matplotlib
     pandas
-    keras==2.9.0
+    keras>=3.0
     onnx
     onnxruntime
     onnxmltools

tests/io/test_keras_reader.py

@@ -10,7 +10,7 @@
     not keras_available, reason="Test only valid when keras is available"
 )
 def test_keras_reader(datadir):
-    nn = keras.models.load_model(datadir.file("keras_linear_131"), compile=False)
+    nn = keras.models.load_model(datadir.file("keras_linear_131.keras"), compile=False)
     net = load_keras_sequential(nn)
 
     layers = list(net.layers)
@@ -21,7 +21,7 @@ def test_keras_reader(datadir):
     assert layers[2].weights.shape == (3, 1)
 
     nn = keras.models.load_model(
-        datadir.file("keras_linear_131_sigmoid"), compile=False
+        datadir.file("keras_linear_131_sigmoid.keras"), compile=False
     )
     net = load_keras_sequential(nn)
     layers = list(net.layers)
@@ -32,7 +32,7 @@ def test_keras_reader(datadir):
     assert layers[2].weights.shape == (3, 1)
 
     nn = keras.models.load_model(
-        datadir.file("keras_linear_131_sigmoid_output_activation"), compile=False
+        datadir.file("keras_linear_131_sigmoid_output_activation.keras"), compile=False
     )
     net = load_keras_sequential(nn)
     layers = list(net.layers)
@@ -42,7 +42,7 @@ def test_keras_reader(datadir):
     assert layers[1].weights.shape == (1, 3)
     assert layers[2].weights.shape == (3, 1)
 
-    nn = keras.models.load_model(datadir.file("big"), compile=False)
+    nn = keras.models.load_model(datadir.file("big.keras"), compile=False)
     net = load_keras_sequential(nn)
     layers = list(net.layers)
     assert len(layers) == 5

tests/neuralnet/test_keras.py

@@ -108,54 +108,56 @@ def _test_keras_linear_big(keras_fname, reduced_space=False):
 
 @pytest.mark.skipif(not keras_available, reason="Need keras for this test")
 def test_keras_linear_131_full(datadir):
-    _test_keras_linear_131(datadir.file("keras_linear_131"))
-    _test_keras_linear_131(datadir.file("keras_linear_131_sigmoid"))
-    _test_keras_linear_131(datadir.file("keras_linear_131_sigmoid_output_activation"))
+    _test_keras_linear_131(datadir.file("keras_linear_131.keras"))
+    _test_keras_linear_131(datadir.file("keras_linear_131_sigmoid.keras"))
     _test_keras_linear_131(
-        datadir.file("keras_linear_131_sigmoid_softplus_output_activation")
+        datadir.file("keras_linear_131_sigmoid_output_activation.keras")
+    )
+    _test_keras_linear_131(
+        datadir.file("keras_linear_131_sigmoid_softplus_output_activation.keras")
     )
 
 
 @pytest.mark.skipif(not keras_available, reason="Need keras for this test")
 def test_keras_linear_131_reduced(datadir):
-    _test_keras_linear_131(datadir.file("keras_linear_131"), reduced_space=True)
+    _test_keras_linear_131(datadir.file("keras_linear_131.keras"), reduced_space=True)
     _test_keras_linear_131(
-        datadir.file("keras_linear_131_sigmoid"),
+        datadir.file("keras_linear_131_sigmoid.keras"),
         reduced_space=True,
     )
     _test_keras_linear_131(
-        datadir.file("keras_linear_131_sigmoid_output_activation"),
+        datadir.file("keras_linear_131_sigmoid_output_activation.keras"),
         reduced_space=True,
     )
     _test_keras_linear_131(
-        datadir.file("keras_linear_131_sigmoid_softplus_output_activation"),
+        datadir.file("keras_linear_131_sigmoid_softplus_output_activation.keras"),
         reduced_space=True,
     )
 
 
 @pytest.mark.skipif(not keras_available, reason="Need keras for this test")
 def test_keras_linear_131_relu(datadir):
     _test_keras_mip_relu_131(
-        datadir.file("keras_linear_131_relu"),
+        datadir.file("keras_linear_131_relu.keras"),
     )
     _test_keras_complementarity_relu_131(
-        datadir.file("keras_linear_131_relu"),
+        datadir.file("keras_linear_131_relu.keras"),
     )
 
 
 @pytest.mark.skipif(not keras_available, reason="Need keras for this test")
 def test_keras_linear_big(datadir):
-    _test_keras_linear_big(datadir.file("big"), reduced_space=False)
+    _test_keras_linear_big(datadir.file("big.keras"), reduced_space=False)
 
 
 @pytest.mark.skip("Skip - this test is too big for now")
 def test_keras_linear_big_reduced_space(datadir):
-    _test_keras_linear_big("./models/big", reduced_space=True)
+    _test_keras_linear_big("./models/big.keras", reduced_space=True)
 
 
 @pytest.mark.skipif(not keras_available, reason="Need keras for this test")
 def test_scaling_NN_block(datadir):
-    NN = keras.models.load_model(datadir.file("keras_linear_131_relu"))
+    NN = keras.models.load_model(datadir.file("keras_linear_131_relu.keras"))
 
     model = pyo.ConcreteModel()
 
@@ -186,7 +188,7 @@ def obj(mdl):
         result = pyo.SolverFactory("cbc").solve(model, tee=False)
 
         x_s = (x - scale_x[0]) / scale_x[1]
-        y_s = NN.predict(x=[x_s])
+        y_s = NN.predict([np.array((x_s,))])
         y = y_s * scale_y[1] + scale_y[0]
 
         assert y - pyo.value(model.nn.outputs[0]) <= 1e-3

tests/neuralnet/train_keras_models.py

@@ -1,11 +1,11 @@
 import pytest
-import tensorflow.keras as keras
+import keras
 
 # from conftest import get_neural_network_data
 from keras.layers import Conv2D, Dense
 from keras.models import Model, Sequential
 from pyomo.common.fileutils import this_file_dir
-from tensorflow.keras.optimizers import Adamax
+from keras.optimizers import Adamax
 
 from omlt.io import write_onnx_model_with_bounds
 
@@ -40,7 +40,7 @@ def train_models():
     history = nn.fit(
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
-    nn.save(this_file_dir() + "/models/keras_linear_131")
+    nn.save(this_file_dir() + "/models/keras_linear_131.keras")
 
     x, y, x_test = get_neural_network_data("131")
     nn = Sequential(name="keras_linear_131_sigmoid")
@@ -72,7 +72,7 @@ def train_models():
     history = nn.fit(
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
-    nn.save(this_file_dir() + "/models/keras_linear_131_sigmoid")
+    nn.save(this_file_dir() + "/models/keras_linear_131_sigmoid.keras")
 
     x, y, x_test = get_neural_network_data("131")
     nn = Sequential(name="keras_linear_131_sigmoid_output_activation")
@@ -105,7 +105,9 @@ def train_models():
     history = nn.fit(
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
-    nn.save(this_file_dir() + "/models/keras_linear_131_sigmoid_output_activation")
+    nn.save(
+        this_file_dir() + "/models/keras_linear_131_sigmoid_output_activation.keras"
+    )
 
     x, y, x_test = get_neural_network_data("131")
     nn = Sequential(name="keras_linear_131_relu")
@@ -137,7 +139,7 @@ def train_models():
     history = nn.fit(
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
-    nn.save(this_file_dir() + "/models/keras_linear_131_relu")
+    nn.save(this_file_dir() + "/models/keras_linear_131_relu.keras")
 
     x, y, x_test = get_neural_network_data("131")
     nn = Sequential(name="keras_linear_131_relu_output_activation")
@@ -170,7 +172,7 @@ def train_models():
     history = nn.fit(
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
-    nn.save(this_file_dir() + "/models/keras_linear_131_relu_output_activation")
+    nn.save(this_file_dir() + "/models/keras_linear_131_relu_output_activation.keras")
 
     x, y, x_test = get_neural_network_data("131")
     nn = Sequential(name="keras_linear_131_sigmoid_softplus_output_activation")
@@ -204,7 +206,8 @@ def train_models():
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
     nn.save(
-        this_file_dir() + "/models/keras_linear_131_sigmoid_softplus_output_activation"
+        this_file_dir()
+        + "/models/keras_linear_131_sigmoid_softplus_output_activation.keras"
     )
 
     x, y, x_test = get_neural_network_data("131")
@@ -263,7 +266,7 @@ def train_models():
     history = nn.fit(
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
-    nn.save(this_file_dir() + "/models/big")
+    nn.save(this_file_dir() + "/models/big.keras")
 
     x, y, x_test = get_neural_network_data("2353")
     nn = Sequential(name="keras_linear_2353")
@@ -306,7 +309,7 @@ def train_models():
         x=x, y=y, validation_split=0.2, batch_size=16, verbose=1, epochs=15
     )
 
-    nn.save(this_file_dir() + "/models/keras_linear_2353")
+    nn.save(this_file_dir() + "/models/keras_linear_2353.keras")
 
 
 def train_conv():

tox.ini

@@ -4,15 +4,17 @@
 
 [tox]
 minversion = 3.15
-envlist = py36, py37, py38, py39, py310, lint
+envlist = py36, py37, py38, py39, py310, py311, py312, lint
 
 [gh-actions]
 python =
     3.6: py36
     3.7: py37
-    3.8: lint, py38
-    3.9: py39
+    3.8: py38
+    3.9: lint, py39
     3.10: py310
+    3.11: py311
+    3.12: py312
 
 [testenv]
 deps = pytest