Format code with black 23

examples/introduction_development.ipynb

@@ -596,7 +596,6 @@
     "        return ibackend.gradients(Xs, Ys, reversed_Ys)\n",
     "\n",
     "    def _create_analysis(self, *args, **kwargs):\n",
-    "\n",
     "        self._add_conditional_reverse_mapping(\n",
     "            # Apply to all layers that contain a relu activation\n",
     "            lambda layer: kchecks.contains_activation(layer, \"relu\"),\n",
@@ -706,7 +705,6 @@
     "        return X\n",
     "\n",
     "    def _create_analysis(self, *args, **kwargs):\n",
-    "\n",
     "        self._add_conditional_reverse_mapping(\n",
     "            # Apply to all layers that contain a kernel\n",
     "            lambda layer: kchecks.contains_kernel(layer),\n",

examples/mnist_perturbation.ipynb

@@ -230,6 +230,7 @@
     "steps = 15\n",
     "regions_per_step = 1  # Perturbate 1 region per step\n",
     "\n",
+    "\n",
     "# Scale to [0, 1] range for plotting.\n",
     "def input_postprocessing(X):\n",
     "    return revert_preprocessing(X) / 255\n",

examples/sentiment_analysis.ipynb

@@ -934,7 +934,6 @@
     "\n",
     "\n",
     "def train_model(model, batch_size=128, epochs=20):\n",
-    "\n",
     "    x_train = DATASETS[\"training\"][\"x4d\"]\n",
     "    y_train = to_one_hot(DATASETS[\"training\"][\"y\"])\n",
     "\n",
@@ -1088,7 +1087,6 @@
     "analysis = np.zeros([len(test_sample_indices), len(analyzers), 1, MAX_SEQ_LENGTH])\n",
     "\n",
     "for i, ridx in enumerate(test_sample_indices):\n",
-    "\n",
     "    x, y = DATASETS[\"testing\"][\"x4d\"][ridx], DATASETS[\"testing\"][\"y\"][ridx]\n",
     "\n",
     "    t_start = time.time()\n",
@@ -1100,7 +1098,6 @@
     "    test_sample_preds[i] = y_hat\n",
     "\n",
     "    for aidx, analyzer in enumerate(analyzers):\n",
-    "\n",
     "        a = np.squeeze(analyzer.analyze(x))\n",
     "        a = np.sum(a, axis=1)\n",
     "\n",
@@ -1478,7 +1475,6 @@
    "source": [
     "# Traverse over the analysis results and visualize them.\n",
     "for i, idx in enumerate(test_sample_indices):\n",
-    "\n",
     "    words = [decoder[t] for t in list(DATASETS[\"testing\"][\"encoded_reviews\"][idx])]\n",
     "\n",
     "    print(\"Review(id=%d): %s\" % (idx, \" \".join(words)))\n",

src/innvestigate/analyzer/deeptaylor.py

@@ -39,7 +39,6 @@ def __init__(self, model: Model, *args, **kwargs) -> None:
         self._do_model_checks()
 
     def _create_analysis(self, *args: Any, **kwargs: Any):
-
         # Kernel layers.
         self._add_conditional_reverse_mapping(
             lambda l: (ichecks.contains_kernel(l) and ichecks.contains_activation(l)),
@@ -176,7 +175,6 @@ def __init__(self, model, low=None, high=None, **kwargs):
         self._bounds_high = high
 
     def _create_analysis(self, *args, **kwargs):
-
         low, high = self._bounds_low, self._bounds_high
 
         class BoundedProxyRule(lrp_rules.BoundedRule):

src/innvestigate/analyzer/gradient_based.py

@@ -146,7 +146,6 @@ class InputTimesGradient(Gradient):
     """
 
     def __init__(self, model, **kwargs):
-
         super().__init__(model, **kwargs)
 
     def _create_analysis(self, model, stop_analysis_at_tensors=None):
@@ -207,7 +206,6 @@ def __init__(self, model, **kwargs):
         self._do_model_checks()
 
     def _create_analysis(self, *args, **kwargs):
-
         self._add_conditional_reverse_mapping(
             lambda layer: ichecks.contains_activation(layer, "relu"),
             DeconvnetReverseReLULayer,
@@ -247,7 +245,6 @@ def __init__(self, model, **kwargs):
         self._do_model_checks()
 
     def _create_analysis(self, *args, **kwargs):
-
         self._add_conditional_reverse_mapping(
             lambda layer: ichecks.contains_activation(layer, "relu"),
             guided_backprop_reverse_relu_layer,
@@ -321,7 +318,6 @@ def __init__(
             subanalyzer = kwargs.pop("subanalyzer")
         # If initialized normally:
         else:
-
             subanalyzer = Gradient(
                 model,
                 neuron_selection_mode=neuron_selection_mode,

src/innvestigate/analyzer/relevance_based/relevance_rule.py

@@ -54,7 +54,6 @@ def apply(
         Rs: list[Tensor],
         _reverse_state,
     ) -> list[Tensor]:
-
         # Get activations.
         Zs = ibackend.apply(self._layer_wo_act, Xs)
         # Divide incoming relevance by the activations.

src/innvestigate/analyzer/reverse_base.py

@@ -254,7 +254,6 @@ def _reverse_model(
     def _create_analysis(
         self, model: Model, stop_analysis_at_tensors: list[Tensor] = None
     ):
-
         if stop_analysis_at_tensors is None:
             stop_analysis_at_tensors = []
 
@@ -320,7 +319,6 @@ def _create_analysis(
         return ret
 
     def _handle_debug_output(self, debug_values):
-
         if self._reverse_check_min_max_values:
             indices = self._debug_tensors_indices["min"]
             tmp = debug_values[indices[0] : indices[1]]

src/innvestigate/analyzer/wrapper.py

@@ -84,7 +84,6 @@ def __init__(
         neuron_selection_mode="max_activation",
         **kwargs,
     ):
-
         if neuron_selection_mode == "max_activation":
             # TODO: find a more transparent way.
             #

src/innvestigate/backend/graph.py

@@ -542,7 +542,6 @@ def actual_mapping(
                 reversed_Ys: list[Tensor],
                 reverse_state,
             ):
-
                 X2s = ibackend.apply(surrogate_layer1, Xs)
                 # Apply first mapping
                 # TODO (alber): update reverse state
@@ -788,7 +787,6 @@ def get_model_execution_trace(
 
     model_execution_trace = []
     for nid, l, Xs, Ys in id_execution_trace:
-
         if isinstance(l, klayers.InputLayer):
             # The nids that created or receive the tensors.
             Xs_nids = []  # Input layer does not receive.

tests/backend/test_layer.py

@@ -69,7 +69,6 @@ def test_reducemean_layer():
 @pytest.mark.fast
 @pytest.mark.precommit
 def test_fast_one_layer():
-
     model = kmodels.Sequential([klayers.Dense(2, input_shape=(3,), use_bias=False)])
     weights = [np.asarray(((1, 2), (3, 4), (5, 6)))]
     model.set_weights(weights)
@@ -90,7 +89,6 @@ def test_fast_one_layer():
 @pytest.mark.fast
 @pytest.mark.precommit
 def test_fast_two_layers():
-
     model = kmodels.Sequential(
         [
             klayers.Dense(2, input_shape=(3,), use_bias=False),