seed on test normalizers

tests/test_layers.py

@@ -192,7 +192,7 @@ def train_k_lip_model(
     # define logging features
     logdir = os.path.join("logs", uft.LIP_LAYERS, "%s" % layer_type.__name__)
     os.makedirs(logdir, exist_ok=True)
-    
+
     callback_list = (
         []
     )  # [callbacks.TensorBoard(logdir), hp.KerasCallback(logdir, hparams)]

tests/test_losses.py

@@ -92,14 +92,14 @@ def check_serialization(nb_class, loss):
 
 
 def binary_data(x):
-    """Return a Framework float32 tensor of shape [N, 1] 
-        from a list/np.array of shape [N]"""
+    """Return a Framework float32 tensor of shape [N, 1]
+    from a list/np.array of shape [N]"""
     return np.expand_dims(np.array(x, dtype=np.float32), axis=-1)
 
 
 def one_hot_data(x, n_class):
-    """Return a Framework float32 tensor of shape [N, n_class] 
-        from a list/np.array of shape [N]"""
+    """Return a Framework float32 tensor of shape [N, n_class]
+    from a list/np.array of shape [N]"""
     return np.eye(n_class)[x]
 
 
@@ -840,8 +840,8 @@ def test_minibatches_binary_loss_generic(
 def test_multilabel_loss_generic(loss_instance, loss_params, rtol):
     """
     Assert binary losses with multilabels.
-    Three losses are tested (KRLoss, HingeMarginLoss and HKRLoss). 
-    We compare losses with three separate binary classification and 
+    Three losses are tested (KRLoss, HingeMarginLoss and HKRLoss).
+    We compare losses with three separate binary classification and
     the corresponding multilabel problem.
     """
     # Create predictions and labels for 3 binary problems and the concatenated

tests/test_metrics.py

@@ -40,8 +40,8 @@
 
 
 def one_hot_data(x, n_class):
-    """Return a Framework float32 tensor of shape [N, n_class] 
-        from a list/np.array of shape [N]"""
+    """Return a Framework float32 tensor of shape [N, n_class]
+    from a list/np.array of shape [N]"""
     return np.eye(n_class)[x]
 
 

tests/test_normalizers.py

@@ -149,6 +149,8 @@ def test_kernel_conv_svd(kernel_shape, strides):
     if hasattr(_padding_circular, "unavailable_class"):
         pytest.skip("_padding_circular not implemented")
 
+    np.random.seed(42)
+
     kernel = np.random.normal(size=kernel_shape).astype("float32")
     spectral_input_shape, RO_case, cPad = set_spectral_input_shape(kernel, strides)
 
@@ -220,6 +222,8 @@ def test_kernel_conv_svd(kernel_shape, strides):
     ],
 )
 def test_bjorck_normalization(kernel_shape):
+    np.random.seed(42)
+
     kernel = np.random.normal(size=kernel_shape).astype("float32")
     """Compare max singular value using power iteration and tf.linalg.svd"""
     sigmas_svd = np.linalg.svd(
@@ -265,6 +269,7 @@ def test_bjorck_normalization(kernel_shape):
 def test_reshaped_kernel_orthogonalization(kernel_shape):
     if hasattr(reshaped_kernel_orthogonalization, "unavailable_class"):
         pytest.skip("reshaped_kernel_orthogonalization not implemented")
+    np.random.seed(42)
 
     kernel = np.random.normal(size=kernel_shape).astype("float32")
     """Compare max singular value using power iteration and tf.linalg.svd"""
@@ -312,6 +317,8 @@ def test_bjorck_norm():
     """
     test bjorck_norm parametrization implementation
     """
+    np.random.seed(42)
+
     m = uft.get_instance_framework(
         tLinear, {"in_features": 2, "out_features": 2}
     )  # torch.nn.Linear(2, 2)
@@ -351,6 +358,8 @@ def test_frobenius_norm():
     """
     test frobenius_norm parametrization implementation
     """
+    np.random.seed(42)
+
     m = uft.get_instance_framework(
         tLinear, {"in_features": 2, "out_features": 2}
     )  # torch.nn.Linear(2, 2)
@@ -386,6 +395,8 @@ def test_frobenius_norm_disjoint_neurons():
     """
     Test `disjoint_neurons=True` argument in frobenius_norm parametrization
     """
+    np.random.seed(42)
+
     params = {"in_features": 5, "out_features": 3}
     m = uft.get_instance_framework(tLinear, params)
 
@@ -410,6 +421,8 @@ def test_lconv_norm():
     """
     test lconv_norm parametrization implementation
     """
+    np.random.seed(42)
+
     params = {
         "in_channels": 1,
         "out_channels": 2,

tests/utils_framework.py

@@ -224,6 +224,7 @@ def get_instance_withcheck(
     ),
 }
 
+
 def get_instance_framework(instance_type, inst_params):
     if instance_type not in getters_dict:
         instance = get_instance_generic(instance_type, inst_params)