Merge 945e7cd into 5c314fd

RasaHQ · Sep 3, 2020 · 96a529d · 96a529d
2 parents 5c314fd + 945e7cd
commit 96a529d
Show file tree

Hide file tree

Showing 5 changed files with 5 additions and 21 deletions.
diff --git a/changelog/6555.improvement.md b/changelog/6555.improvement.md
@@ -0,0 +1 @@
+Do not set the output dimension of the ´sparse-to-dense´ layers to the same dimension as the dense features.
diff --git a/docs/docs/components/intent-classifiers.mdx b/docs/docs/components/intent-classifiers.mdx
@@ -364,9 +364,6 @@ Intent classifiers assign one of the intents defined in the domain file to incom
     +---------------------------------+------------------+--------------------------------------------------------------+
     | embedding_dimension             | 20               | Dimension size of embedding vectors.                         |
     +---------------------------------+------------------+--------------------------------------------------------------+
-    | dense_dimension                 | text: 512        | Dense dimension for sparse features to use if no dense       |
-    |                                 | label: 20        | features are present.                                        |
-    +---------------------------------+------------------+--------------------------------------------------------------+
     | concat_dimension                | text: 512        | Concat dimension for sequence and sentence features.         |
     |                                 | label: 20        |                                                              |
     +---------------------------------+------------------+--------------------------------------------------------------+

diff --git a/docs/docs/components/selectors.mdx b/docs/docs/components/selectors.mdx
@@ -187,9 +187,6 @@ Selectors predict a bot response from a set of candidate responses.
     +---------------------------------+-------------------+--------------------------------------------------------------+
     | embedding_dimension             | 20                | Dimension size of embedding vectors.                         |
     +---------------------------------+-------------------+--------------------------------------------------------------+
-    | dense_dimension                 | text: 512         | Dense dimension for sparse features to use if no dense       |
-    |                                 | label: 512        | features are present.                                        |
-    +---------------------------------+-------------------+--------------------------------------------------------------+
     | concat_dimension                | text: 512         | Concat dimension for sequence and sentence features.         |
     |                                 | label: 512        |                                                              |
     +---------------------------------+-------------------+--------------------------------------------------------------+

diff --git a/rasa/nlu/classifiers/diet_classifier.py b/rasa/nlu/classifiers/diet_classifier.py
@@ -52,7 +52,6 @@
     EPOCHS,
     RANDOM_SEED,
     LEARNING_RATE,
-    DENSE_DIMENSION,
     RANKING_LENGTH,
     LOSS_TYPE,
     SIMILARITY_TYPE,
@@ -102,6 +101,7 @@
 LABEL_SEQUENCE_LENGTH = f"{LABEL}_{SEQUENCE}_lengths"
 LABEL_IDS = f"{LABEL}_ids"
 TAG_IDS = "tag_ids"
+DEFAULT_DENSE_DIMENSION = 100
 
 POSSIBLE_TAGS = [ENTITY_ATTRIBUTE_TYPE, ENTITY_ATTRIBUTE_ROLE, ENTITY_ATTRIBUTE_GROUP]
 
@@ -172,8 +172,6 @@ def required_components(cls) -> List[Type[Component]]:
         # ## Parameters for embeddings
         # Dimension size of embedding vectors
         EMBEDDING_DIMENSION: 20,
-        # Default dense dimension to use if no dense features are present.
-        DENSE_DIMENSION: {TEXT: 512, LABEL: 20},
         # Default dimension to use for concatenating sequence and sentence features.
         CONCAT_DIMENSION: {TEXT: 512, LABEL: 20},
         # The number of incorrect labels. The algorithm will minimize
@@ -1331,22 +1329,17 @@ def _prepare_layers(self) -> None:
             self._prepare_entity_recognition_layers()
 
     def _prepare_sparse_dense_layers(
-        self,
-        feature_signatures: List[FeatureSignature],
-        name: Text,
-        reg_lambda: float,
-        dense_dim: int,
+        self, feature_signatures: List[FeatureSignature], name: Text, reg_lambda: float
     ) -> None:
         sparse = False
         dense = False
+        dense_dim = DEFAULT_DENSE_DIMENSION
         for is_sparse, feature_dimension in feature_signatures:
             if is_sparse:
                 sparse = True
+                dense_dim = min(dense_dim, feature_dimension)
             else:
                 dense = True
-                # if dense features are present
-                # use the feature dimension of the dense features
-                dense_dim = feature_dimension
 
         if sparse:
             self._tf_layers[f"sparse_to_dense.{name}"] = layers.DenseForSparse(
@@ -1380,7 +1373,6 @@ def _prepare_input_layers(self, name: Text) -> None:
                 self.data_signature[f"{name}_{feature_type}_features"],
                 f"{name}_{feature_type}",
                 self.config[REGULARIZATION_CONSTANT],
-                self.config[DENSE_DIMENSION][name],
             )
             self._tf_layers[f"concat_layer.{name}_{feature_type}"] = layers.Ffnn(
                 [self.config[CONCAT_DIMENSION][name]],

diff --git a/rasa/nlu/selectors/response_selector.py b/rasa/nlu/selectors/response_selector.py
@@ -36,7 +36,6 @@
     EPOCHS,
     RANDOM_SEED,
     LEARNING_RATE,
-    DENSE_DIMENSION,
     RANKING_LENGTH,
     LOSS_TYPE,
     SIMILARITY_TYPE,
@@ -156,8 +155,6 @@ def required_components(cls) -> List[Type[Component]]:
         # ## Parameters for embeddings
         # Dimension size of embedding vectors
         EMBEDDING_DIMENSION: 20,
-        # Default dense dimension to use if no dense features are present.
-        DENSE_DIMENSION: {TEXT: 512, LABEL: 512},
         # Default dimension to use for concatenating sequence and sentence features.
         CONCAT_DIMENSION: {TEXT: 512, LABEL: 512},
         # The number of incorrect labels. The algorithm will minimize
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		Do not set the output dimension of the ´sparse-to-dense´ layers to the same dimension as the dense features.