Merge pull request #5184 from RasaHQ/bug-multi-intent

Fix bug with multi-intents
RasaHQ · Feb 7, 2020 · a3549d1 · a3549d1
2 parents f95c5fa + 2c7a4a5
commit a3549d1
Show file tree

Hide file tree

Showing 6 changed files with 88 additions and 8 deletions.
diff --git a/changelog/5162.bugfix.rst b/changelog/5162.bugfix.rst
@@ -0,0 +1,3 @@
+Fix problem with multi-intents.
+Training with multi-intents using the ``CountVectorsFeaturizer`` together with ``EmbeddingIntentClassifier`` is
+working again.
diff --git a/data/examples/rasa/demo-rasa-multi-intent.md b/data/examples/rasa/demo-rasa-multi-intent.md
@@ -0,0 +1,71 @@
+## intent:affirm
+- yes
+- yep
+- yeah
+- indeed
+- that's right
+- ok
+- great
+- right, thank you
+- correct
+- great choice
+- sounds really good
+
+## intent:goodbye
+- bye
+- goodbye
+- good bye
+- stop
+- end
+- farewell
+- Bye bye
+- have a good one
+
+## intent:greet
+- hey
+- howdy
+- hey there
+- hello
+- hi
+- good morning
+- good evening
+- dear sir
+
+## intent:chitchat+ask_name
+- What's your name?
+- What can I call you?
+
+## intent:chitchat+ask_weather
+- How's the weather?
+- Is it too hot outside?
+
+## intent:restaurant_search
+- i'm looking for a place to eat
+- I want to grab lunch
+- I am searching for a dinner spot
+- i'm looking for a place in the [north](location) of town
+- show me [chinese](cuisine) restaurants
+- show me [chines](cuisine:chinese) restaurants in the [north](location)
+- show me a [mexican](cuisine) place in the [centre](location)
+- i am looking for an [indian](cuisine) spot called olaolaolaolaolaola
+- search for restaurants
+- anywhere in the [west](location)
+- anywhere near [18328](location)
+- I am looking for [asian fusion](cuisine) food
+- I am looking a restaurant in [29432](location)
+- I am looking for [mexican indian fusion](cuisine)
+- [central](location) [indian](cuisine) restaurant
+
+## synonym:chinese
++ Chines
+* Chinese
+
+## synonym:vegetarian
+- vegg
+- veggie
+
+## regex:zipcode
+- [0-9]{5}
+
+## regex:greet
+- hey[^\s]*
diff --git a/data/examples/rasa/demo-rasa.json b/data/examples/rasa/demo-rasa.json
@@ -23,17 +23,17 @@
     "common_examples": [
       {
         "text": "hey",
-        "intent": "greet", 
+        "intent": "greet",
         "entities": []
       }, 
       {
         "text": "howdy",
-        "intent": "greet", 
+        "intent": "greet",
         "entities": []
       }, 
       {
         "text": "hey there",
-        "intent": "greet", 
+        "intent": "greet",
         "entities": []
       }, 
       {

diff --git a/rasa/nlu/classifiers/embedding_intent_classifier.py b/rasa/nlu/classifiers/embedding_intent_classifier.py
@@ -543,7 +543,8 @@ def _combine_sparse_dense_features(
                 dense_features.append(f)
 
         output = tf.concat(dense_features, axis=-1) * mask
-        output = tf.squeeze(output, axis=1)
+        # reduce dimensionality of output
+        output = tf.reduce_sum(output, axis=1)
 
         return output
 
@@ -985,6 +986,6 @@ def load(
         else:
             raise_warning(
                 f"Failed to load nlu model. "
-                f"Maybe the path '{os.path.abspath(model_dir)}' doesn't exist?",
+                f"Maybe the path '{os.path.abspath(model_dir)}' doesn't exist?"
             )
             return cls(component_config=meta)
diff --git a/tests/core/test_data.py b/tests/core/test_data.py
@@ -148,6 +148,7 @@ def test_same_file_names_get_resolved(tmpdir):
                 "data/examples/rasa/demo-rasa.json",
                 "data/examples/rasa/demo-rasa.md",
                 "data/examples/rasa/demo-rasa-responses.md",
+                "data/examples/rasa/demo-rasa-multi-intent.md",
             },
         ),
         ("wit", {"data/examples/wit/demo-flights.json"}),

diff --git a/tests/nlu/classifiers/test_embedding_intent_classifier.py b/tests/nlu/classifiers/test_embedding_intent_classifier.py
@@ -116,7 +116,11 @@ def test_check_labels_features_exist(messages, expected):
 
 async def test_train(component_builder, tmpdir):
     pipeline = [
-        {"name": "ConveRTTokenizer"},
+        {
+            "name": "ConveRTTokenizer",
+            "intent_tokenization_flag": True,
+            "intent_split_symbol": "+",
+        },
         {"name": "CountVectorsFeaturizer"},
         {"name": "ConveRTFeaturizer"},
         {"name": "EmbeddingIntentClassifier"},
@@ -127,7 +131,7 @@ async def test_train(component_builder, tmpdir):
     (trained, _, persisted_path) = await train(
         _config,
         path=tmpdir.strpath,
-        data=DEFAULT_DATA_PATH,
+        data="data/examples/rasa/demo-rasa-multi-intent.md",
         component_builder=component_builder,
     )
 
@@ -243,7 +247,7 @@ async def test_softmax_normalization(
     [({"loss_type": "margin", "random_seed": 42}, LABEL_RANKING_LENGTH)],
 )
 async def test_margin_loss_is_not_normalized(
-    monkeypatch, component_builder, tmpdir, classifier_params, output_length,
+    monkeypatch, component_builder, tmpdir, classifier_params, output_length
 ):
     pipeline = as_pipeline(
         "WhitespaceTokenizer", "CountVectorsFeaturizer", "EmbeddingIntentClassifier"