Merge branch vlad into main

.env.sample

@@ -0,0 +1 @@
+SENTRY_URL=<sentry_url>

CHANGELOG.md

@@ -0,0 +1,18 @@
+
+## [0.0.12](https://gitlab.com/tangibleai/community/mathtext-fastapi/-/tags/0.0.12)
+
+Improve NLU capabilities
+- Improved handling for integers (1), floats (1.0), and text numbers (one)
+- Integrates fuzzy keyword matching for 'easier', 'exit', 'harder', 'hint', 'next', 'stop'
+- Integrates intent classification for user messages
+- Improved conversation management system
+- Created a data-driven quiz prototype
+
+
+## [0.0.0](https://gitlab.com/tangibleai/community/mathtext-fastapi/-/tags/0.0.0)
+
+Initial release
+- Basic text to integer NLU evaluation of user responses
+- Basic sentiment analysis evaluation of user responses
+- Prototype conversation manager using finite state machines
+- Support for logging of user message data

app.py

@@ -4,7 +4,10 @@ or
 `python -m uvicorn app:app --reload --host localhost --port 7860`
 """
 import ast
+import json
 import mathactive.microlessons.num_one as num_one_quiz
+import os
+import sentry_sdk
 from fastapi import FastAPI, Request
 from fastapi.responses import JSONResponse
 from fastapi.staticfiles import StaticFiles
@@ -18,6 +21,22 @@ from mathtext_fastapi.conversation_manager import manage_conversation_response
 from mathtext_fastapi.v2_conversation_manager import manage_conversation_response
 from mathtext_fastapi.nlu import evaluate_message_with_nlu
 from mathtext_fastapi.nlu import run_intent_classification
+from dotenv import load_dotenv
+from sentry_sdk.utils import BadDsn
+
+load_dotenv()
+
+try:
+    sentry_sdk.init(
+        dsn=os.environ.get('SENTRY_DNS'),
+
+        # Set traces_sample_rate to 1.0 to capture 100%
+        # of transactions for performance monitoring.
+        # We recommend adjusting this value in production,
+        traces_sample_rate=0.20,
+    )
+except BadDsn:
+    pass
 
 app = FastAPI()
 
@@ -35,6 +54,11 @@ def home(request: Request):
     return templates.TemplateResponse("home.html", {"request": request})
 
 
+@app.get("/sentry-debug")
+async def trigger_error():
+    division_by_zero = 1 / 0
+
+
 @app.post("/hello")
 def hello(content: Text = None):
     content = {"message": f"Hello {content.content}!"}
@@ -164,7 +188,7 @@ async def num_one(request: Request):
     """
     print("STEP 1")
     data_dict = await request.json()
-    message_data = ast.literal_eval(data_dict.get('message_data', '').get('message_body', ''))
+    message_data = json.loads(data_dict.get('message_data', '').get('message_body', '').replace("'", '"'))
     user_id = message_data['user_id']
     message_text = message_data['message_text']
     print("STEP 2")

mathtext_fastapi/intent_classification.py

@@ -41,9 +41,13 @@ def retrieve_intent_classification_model():
     return model
 
 
+encoder = SentenceTransformer('all-MiniLM-L6-v2')
+# model = retrieve_intent_classification_model()
+DATA_DIR = Path(__file__).parent.parent / "mathtext_fastapi" / "data" / "intent_classification_model.joblib"
+model = load(DATA_DIR)
+
+
 def predict_message_intent(message):
-    encoder = SentenceTransformer('all-MiniLM-L6-v2')
-    model = retrieve_intent_classification_model()
     tokenized_utterance = np.array([list(encoder.encode(message))])
     predicted_label = model.predict(tokenized_utterance)
     predicted_probabilities = model.predict_proba(tokenized_utterance)

mathtext_fastapi/nlu.py

@@ -110,7 +110,10 @@ def run_intent_classification(message_text):
     ]
     
     for command in commands:
-        ratio = fuzz.ratio(command, message_text.lower())
+        try:
+            ratio = fuzz.ratio(command, message_text.lower())
+        except:
+            ratio = 0
         if ratio > 80:
             nlu_response['data'] = command
             nlu_response['confidence'] = ratio / 100
@@ -129,7 +132,7 @@ def evaluate_message_with_nlu(message_data):
     """
     # Keeps system working with two different inputs - full and filtered @event object
     try:
-        message_text = message_data['message_body']
+        message_text = str(message_data['message_body'])
     except KeyError:
         message_data = {
             'author_id': message_data['message']['_vnd']['v1']['chat']['owner'],
@@ -141,11 +144,12 @@ def evaluate_message_with_nlu(message_data):
             'message_inserted_at': message_data['message']['_vnd']['v1']['chat']['inserted_at'],
             'message_updated_at': message_data['message']['_vnd']['v1']['chat']['updated_at'],
         }
-        message_text = message_data['message_body']
+        message_text = str(message_data['message_body'])
 
     # Run intent classification only for keywords
     intent_api_response = run_intent_classification(message_text)
     if intent_api_response['data']:
+        prepare_message_data_for_logging(message_data, intent_api_response)
         return intent_api_response
 
     number_api_resp = text2int(message_text.lower())

pyproject.toml

@@ -31,6 +31,7 @@ supabase = "*"
 uvicorn = "0.17.*"
 pandas = "^1.5.3"
 scipy = "^1.10.1"
+sentry_sdk = "^1.19.1"
 
 [tool.poetry.group.dev.dependencies]
 pytest = "^7.2"

requirements.txt

@@ -3,7 +3,7 @@ en-core-web-sm @ https://github.com/explosion/spacy-models/releases/download/en_
 fuzzywuzzy
 jsonpickle
 mathtext @ git+https://gitlab.com/tangibleai/community/mathtext@main
-mathactive @ git+https://gitlab.com/tangibleai/community/mathactive@main
+mathactive @ git+https://gitlab.com/tangibleai/community/mathactive@vlad
 fastapi
 pydantic
 requests
@@ -11,6 +11,7 @@ sentencepiece
 openpyxl
 python-Levenshtein
 sentence-transformers
+sentry-sdk[fastapi]
 supabase
 transitions
 uvicorn

scripts/env_loader.py

@@ -0,0 +1,9 @@
+import dotenv
+import os
+
+dotenv.load_dotenv(".env")
+envs = {}
+
+
+for var in ["SENTRY_URL"]:
+    envs[var] = os.getenv(var)

scripts/logger.py

@@ -0,0 +1,11 @@
+import sentry_sdk
+from . import env_loader
+
+sentry_sdk.init(
+    dsn=env_loader.envs.get("SENTRY_URL"),
+
+    # Set traces_sample_rate to 1.0 to capture 100%
+    # of transactions for performance monitoring.
+    # We recommend adjusting this value in production.
+    traces_sample_rate=1.0
+)

scripts/make_request.py

@@ -66,21 +66,19 @@ def run_simulated_request(endpoint, sample_answer, context=None):
 # run_simulated_request('intent-classification', "I'm not sure")
 # run_simulated_request('sentiment-analysis', 'I reject it')
 # run_simulated_request('text2int', 'seven thousand nine hundred fifty seven')
-# run_simulated_request('nlu', 'test message')
-# run_simulated_request('nlu', 'eight')
-# run_simulated_request('nlu', 'is it 8')
-# run_simulated_request('nlu', 'can I know how its 0.5')
-# run_simulated_request('nlu', 'eight, nine, ten')
-# run_simulated_request('nlu', '8, 9, 10')
-# run_simulated_request('nlu', '8')
-# run_simulated_request('nlu', "I don't know")
-# run_simulated_request('nlu', "I don't know eight")
-# run_simulated_request('nlu', "I don't 9")
-# run_simulated_request('nlu', "0.2")
-# run_simulated_request('nlu', 'Today is a wonderful day')
-# run_simulated_request('nlu', 'IDK 5?')
-# run_simulated_request('v2/manager', '')
-# run_simulated_request('v2/manager', '5')
+run_simulated_request('nlu', 'test message')
+run_simulated_request('nlu', 'eight')
+run_simulated_request('nlu', 'is it 8')
+run_simulated_request('nlu', 'can I know how its 0.5')
+run_simulated_request('nlu', 'eight, nine, ten')
+run_simulated_request('nlu', '8, 9, 10')
+run_simulated_request('nlu', '8')
+run_simulated_request('nlu', "I don't know")
+run_simulated_request('nlu', "I don't know eight")
+run_simulated_request('nlu', "I don't 9")
+run_simulated_request('nlu', "0.2")
+run_simulated_request('nlu', 'Today is a wonderful day')
+run_simulated_request('nlu', 'IDK 5?')
 # run_simulated_request('manager', '')
 # run_simulated_request('manager', 'add')
 # run_simulated_request('manager', 'subtract')