updated matbench training.

Author: PatReis

changelog.md

@@ -5,6 +5,7 @@ v2.2.3
 * Fix bug in latest ``train_force.py`` of v2.2.2 that forgets to apply inverse scaling to dataset, causing subsequent folds to have wrong labels.
 * HOTFIX: Updated ``MolDynamicsModelPredictor`` to call keras model without very expensive retracing. Alternative mode use `use_predict=True` .
 * Update training results and data subclasses for matbench datasets.
+* Added ``GraphInstanceNormalization`` and `GraphNormalization` to `kgcnn.layers.norm` .
 
 
 v2.2.2

kgcnn/layers/mlp.py

@@ -380,7 +380,7 @@ def __init__(self, units, num_relations: int, num_bases: int = None, num_blocks:
                         **self._conf_mlp_norm_layer_kwargs[i])
                 else:
                     raise NotImplementedError(
-                        "Normalization via %s not supported." % self._conf_normalization_technique[i])
+                        "Normalization via '%s' not supported." % self._conf_normalization_technique[i])
 
     def build(self, input_shape):
         """Build layer."""

kgcnn/layers/norm.py

@@ -406,8 +406,9 @@ def _ragged_mean_std(self, inputs):
         if self.mean_shift:
             mean = mean * tf.expand_dims(self.alpha, axis=0)
         mean = tf.repeat(mean, inputs.row_lengths(), axis=0)
-        diff = values - tf.stop_gradient(mean)
-        square_diff = tf.square(diff)
+        diff = values - mean
+        # Not sure whether to stop gradients for variance if alpha ist used.
+        square_diff = tf.square(diff)  # values - tf.stop_gradient(mean)
         variance = segment_ops_by_name("mean", square_diff, inputs.value_rowids())
         std = tf.sqrt(variance + self._eps)
         std = tf.repeat(std, inputs.row_lengths(), axis=0)