add check that weights have changed during training

tests/test_two_moons/conftest.py

@@ -17,13 +17,6 @@ def batch_size():
  return 128
 
 
-@pytest.fixture()
-def inference_network():
- from bayesflow.networks import CouplingFlow
-
- return CouplingFlow()
-
-
 @pytest.fixture()
 def random_samples(batch_size, simulator):
  return simulator.sample((batch_size,))

tests/test_two_moons/test_two_moons.py

@@ -1,3 +1,4 @@
+import copy
 import keras
 import pytest
 
@@ -19,12 +20,17 @@ def test_fit(approximator, train_dataset, validation_dataset, batch_size):
 
  approximator.build_from_data(train_dataset[0])
 
+ untrained_weights = copy.deepcopy(approximator.weights)
  untrained_metrics = approximator.evaluate(validation_dataset, return_dict=True)
 
  approximator.fit(train_dataset, epochs=20)
 
+ trained_weights = approximator.weights
  trained_metrics = approximator.evaluate(validation_dataset, return_dict=True)
 
+ # check weights have changed during training
+ assert any([keras.ops.any(~keras.ops.isclose(u, t)) for u, t in zip(untrained_weights, trained_weights)])
+
  assert isinstance(untrained_metrics, dict)
  assert isinstance(trained_metrics, dict)