Updated nlp.py script to account for the repackaging of the riva.clie…

…nt Python module

Signed-off-by: Sven Chilton <[email protected]>

virtual-assistant-rasa/rasa-weatherbot/riva_local/nlp/nlp.py

@@ -5,16 +5,20 @@
 # README.md file.
 # ==============================================================================
 
-import riva_api.riva_nlp_pb2 as rnlp
-import riva_api.riva_nlp_pb2_grpc as rnlp_srv
+import riva.client
+from riva.client.proto.riva_nlp_pb2 import (
+    AnalyzeIntentResponse,
+    NaturalQueryResponse,
+    TokenClassResponse
+)
 
 import grpc
 from config import riva_config, rivanlp_config
 import requests
 import json
 
-channel = grpc.insecure_channel(riva_config["RIVA_SPEECH_API_URL"])
-riva_nlp = rnlp_srv.RivaLanguageUnderstandingStub(channel)
+auth = riva.client.Auth(uri=riva_config["RIVA_SPEECH_API_URL"])
+riva_nlp = riva.client.NLPService(auth)
 
 # The default intent object for specifying the city
 SPECIFY_CITY_INTENT = { 'value': "specify_city", 'confidence': 1 }
@@ -28,7 +32,7 @@
 nlu_fallback_threshold = rivanlp_config["NLU_FALLBACK_THRESHOLD"] if "NLU_FALLBACK_THRESHOLD" in rivanlp_config else NLU_FALLBACK_THRESHOLD
 
 
-def get_intent_rnlp(resp, result):
+def get_intent_old(resp, result):
     if hasattr(resp, 'intent') and (hasattr(resp, 'domain') and resp.domain.class_name != "nomatch.none"):
         result['intent'] = { 'value': resp.intent.class_name, 'confidence': resp.intent.score }
         parentclassintent_index = result['intent']['value'].find(".")
@@ -43,7 +47,7 @@ def set_nlu_fallback_intent(result):
             print("[Riva NLU] Intent set to ", NLU_FALLBACK_INTENT)
 
 
-def get_entities_rnlp(resp, result):
+def get_entities(resp, result):
     location_found_flag = False
     all_entities_class = {}
     all_entities = []
@@ -72,7 +76,7 @@ def get_entities_rnlp(resp, result):
                 result['entities'].append(entity)
                 break
     return location_found_flag
-                
+
 
 def get_entities_rcnlp(resp, result):
     location_found_flag = False
@@ -89,54 +93,71 @@ def get_entities_rcnlp(resp, result):
 
 
 def get_riva_output(text):
-    # Submit an AnalyzeIntentRequest. We do not provide a domain with the query, so a domain
+    # Submit an AnalyzeIntent request. We do not provide a domain with the query, so a domain
     # classifier is run first, and based on the inferred value from the domain classifier,
     # the query is run through the appropriate intent/slot classifier
     # Note: the detected domain is also returned in the response.
-    result = {'intent': None, 'entities': []}
     try:
-        req = rnlp.AnalyzeIntentRequest()
-        req.query = str(text)
         # The <domain_name> is appended to "riva_intent_" to look for a model "riva_intent_<domain_name>"
         # So the model "riva_intent_<domain_name>" needs to be preloaded in riva server.
         # In this case the domain is weather and the model being used is "riva_intent_weather-misc".
-        req.options.domain = "weather"
-        resp = riva_nlp.AnalyzeIntent(req)
+        options = riva.client.AnalyzeIntentOptions(lang='en-US', domain='weather')
+
+        resp: AnalyzeIntentResponse = riva_nlp.analyze_intent(text, options)
+
     except Exception as inst:
         # An exception occurred
-        print("[Riva NLU] ERROR: Error during NLU request: " + str(inst))
-        return {"error": 'Riva_NLU_Error', "error_message": str(inst)}
-    get_intent_rnlp(resp, result)
-    location_found_flag = get_entities_rnlp(resp, result)
-    if not location_found_flag:
+        print("[Riva NLU] Error during NLU request")
+        return {'riva_error': 'riva_error'}
+    entities = {}
+    get_intent(resp, entities)
+    get_slots(resp, entities)
+    if 'location' not in entities:
         if verbose:
             print(f"[Riva NLU] Did not find any location in the string: {text}\n"
                     "[Riva NLU] Checking again using NER model")
         try:
-            req = rnlp.TokenClassRequest()
-            req.model.model_name = "riva_ner"
-            req.text.append(text)
-            resp_ner = riva_nlp.ClassifyTokens(req)
+            model_name = "riva_ner"
+            resp_ner: TokenClassResponse = riva_nlp.classify_tokens(text, model_name)
         except Exception as inst:
             # An exception occurred
-            print("[Riva NLU] ERROR: Error during NLU request (riva_ner): " + str(inst))
-            return {"error": 'Riva_NLU_Error', "error_message": str(inst)}
+            print("[Riva NLU] Error during NLU request (riva_ner)")
+            return {'riva_error': 'riva_error'}
+
         if verbose:
             print(f"[Riva NLU] NER response results: \n {resp_ner.results[0].results}\n")
             print("[Riva NLU] Location Entities:")
-        location_found_flag = get_entities_rcnlp(resp_ner, result)
-        if location_found_flag:
-            if not result['intent']:
-                result['intent'] = SPECIFY_CITY_INTENT
+        loc_count = 0
+        for result in resp_ner.results[0].results:
+            if result.label[0].class_name == "LOC":
                 if verbose:
-                    print("[Riva NLU] Intent set to ", SPECIFY_CITY_INTENT)
-        else:
+                    print(f"[Riva NLU] Location found: {result.token}") # Flow unhandled for multiple location input
+                loc_count += 1
+                entities['location'] = result.token
+        if loc_count == 0:
             if verbose:
                 print("[Riva NLU] No location found in string using NER LOC")
-    set_nlu_fallback_intent(result)
+                print("[Riva NLU] Checking response domain")
+            if resp.domain.class_name == "nomatch.none":
+                # as a final resort try QA API
+                if enable_qa == "true":
+                    if verbose:
+                        print("[Riva NLU] Checking using QA API")
+                    riva_misty_profile = requests.get(nlp_config["RIVA_MISTY_PROFILE"]).text # Live pull from Cloud
+                    qa_resp = get_qa_answer(riva_misty_profile, text, p_threshold)
+                    if not qa_resp['result'] == '':
+                        if verbose:
+                            print("[Riva NLU] received qa result")
+                        entities['intent'] = 'qa_answer'
+                        entities['answer_span'] = qa_resp['result']
+                        entities['query'] = text
+                    else:
+                        entities['intent'] = 'riva_error'
+                else:
+                    entities['intent'] = 'riva_error'
     if verbose:
-        print("[Riva NLU] Final Riva NLU Output: ", result)
-    return result
+        print("[Riva NLU] This is what entities contain: ", entities)
+    return entities
 
 
 def get_intent_and_entities(text):