Spaces:
Build error
Build error
Merge branch vlad into main
Browse files- .env.sample +1 -0
- CHANGELOG.md +18 -0
- app.py +25 -1
- mathtext_fastapi/intent_classification.py +6 -2
- mathtext_fastapi/nlu.py +7 -3
- pyproject.toml +1 -0
- requirements.txt +2 -1
- scripts/env_loader.py +9 -0
- scripts/logger.py +11 -0
- scripts/make_request.py +13 -15
.env.sample
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
SENTRY_URL=<sentry_url>
|
CHANGELOG.md
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
## [0.0.12](https://gitlab.com/tangibleai/community/mathtext-fastapi/-/tags/0.0.12)
|
3 |
+
|
4 |
+
Improve NLU capabilities
|
5 |
+
- Improved handling for integers (1), floats (1.0), and text numbers (one)
|
6 |
+
- Integrates fuzzy keyword matching for 'easier', 'exit', 'harder', 'hint', 'next', 'stop'
|
7 |
+
- Integrates intent classification for user messages
|
8 |
+
- Improved conversation management system
|
9 |
+
- Created a data-driven quiz prototype
|
10 |
+
|
11 |
+
|
12 |
+
## [0.0.0](https://gitlab.com/tangibleai/community/mathtext-fastapi/-/tags/0.0.0)
|
13 |
+
|
14 |
+
Initial release
|
15 |
+
- Basic text to integer NLU evaluation of user responses
|
16 |
+
- Basic sentiment analysis evaluation of user responses
|
17 |
+
- Prototype conversation manager using finite state machines
|
18 |
+
- Support for logging of user message data
|
app.py
CHANGED
@@ -4,7 +4,10 @@ or
|
|
4 |
`python -m uvicorn app:app --reload --host localhost --port 7860`
|
5 |
"""
|
6 |
import ast
|
|
|
7 |
import mathactive.microlessons.num_one as num_one_quiz
|
|
|
|
|
8 |
from fastapi import FastAPI, Request
|
9 |
from fastapi.responses import JSONResponse
|
10 |
from fastapi.staticfiles import StaticFiles
|
@@ -18,6 +21,22 @@ from mathtext_fastapi.conversation_manager import manage_conversation_response
|
|
18 |
from mathtext_fastapi.v2_conversation_manager import manage_conversation_response
|
19 |
from mathtext_fastapi.nlu import evaluate_message_with_nlu
|
20 |
from mathtext_fastapi.nlu import run_intent_classification
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
app = FastAPI()
|
23 |
|
@@ -35,6 +54,11 @@ def home(request: Request):
|
|
35 |
return templates.TemplateResponse("home.html", {"request": request})
|
36 |
|
37 |
|
|
|
|
|
|
|
|
|
|
|
38 |
@app.post("/hello")
|
39 |
def hello(content: Text = None):
|
40 |
content = {"message": f"Hello {content.content}!"}
|
@@ -164,7 +188,7 @@ async def num_one(request: Request):
|
|
164 |
"""
|
165 |
print("STEP 1")
|
166 |
data_dict = await request.json()
|
167 |
-
message_data =
|
168 |
user_id = message_data['user_id']
|
169 |
message_text = message_data['message_text']
|
170 |
print("STEP 2")
|
|
|
4 |
`python -m uvicorn app:app --reload --host localhost --port 7860`
|
5 |
"""
|
6 |
import ast
|
7 |
+
import json
|
8 |
import mathactive.microlessons.num_one as num_one_quiz
|
9 |
+
import os
|
10 |
+
import sentry_sdk
|
11 |
from fastapi import FastAPI, Request
|
12 |
from fastapi.responses import JSONResponse
|
13 |
from fastapi.staticfiles import StaticFiles
|
|
|
21 |
from mathtext_fastapi.v2_conversation_manager import manage_conversation_response
|
22 |
from mathtext_fastapi.nlu import evaluate_message_with_nlu
|
23 |
from mathtext_fastapi.nlu import run_intent_classification
|
24 |
+
from dotenv import load_dotenv
|
25 |
+
from sentry_sdk.utils import BadDsn
|
26 |
+
|
27 |
+
load_dotenv()
|
28 |
+
|
29 |
+
try:
|
30 |
+
sentry_sdk.init(
|
31 |
+
dsn=os.environ.get('SENTRY_DNS'),
|
32 |
+
|
33 |
+
# Set traces_sample_rate to 1.0 to capture 100%
|
34 |
+
# of transactions for performance monitoring.
|
35 |
+
# We recommend adjusting this value in production,
|
36 |
+
traces_sample_rate=0.20,
|
37 |
+
)
|
38 |
+
except BadDsn:
|
39 |
+
pass
|
40 |
|
41 |
app = FastAPI()
|
42 |
|
|
|
54 |
return templates.TemplateResponse("home.html", {"request": request})
|
55 |
|
56 |
|
57 |
+
@app.get("/sentry-debug")
|
58 |
+
async def trigger_error():
|
59 |
+
division_by_zero = 1 / 0
|
60 |
+
|
61 |
+
|
62 |
@app.post("/hello")
|
63 |
def hello(content: Text = None):
|
64 |
content = {"message": f"Hello {content.content}!"}
|
|
|
188 |
"""
|
189 |
print("STEP 1")
|
190 |
data_dict = await request.json()
|
191 |
+
message_data = json.loads(data_dict.get('message_data', '').get('message_body', '').replace("'", '"'))
|
192 |
user_id = message_data['user_id']
|
193 |
message_text = message_data['message_text']
|
194 |
print("STEP 2")
|
mathtext_fastapi/intent_classification.py
CHANGED
@@ -41,9 +41,13 @@ def retrieve_intent_classification_model():
|
|
41 |
return model
|
42 |
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
def predict_message_intent(message):
|
45 |
-
encoder = SentenceTransformer('all-MiniLM-L6-v2')
|
46 |
-
model = retrieve_intent_classification_model()
|
47 |
tokenized_utterance = np.array([list(encoder.encode(message))])
|
48 |
predicted_label = model.predict(tokenized_utterance)
|
49 |
predicted_probabilities = model.predict_proba(tokenized_utterance)
|
|
|
41 |
return model
|
42 |
|
43 |
|
44 |
+
encoder = SentenceTransformer('all-MiniLM-L6-v2')
|
45 |
+
# model = retrieve_intent_classification_model()
|
46 |
+
DATA_DIR = Path(__file__).parent.parent / "mathtext_fastapi" / "data" / "intent_classification_model.joblib"
|
47 |
+
model = load(DATA_DIR)
|
48 |
+
|
49 |
+
|
50 |
def predict_message_intent(message):
|
|
|
|
|
51 |
tokenized_utterance = np.array([list(encoder.encode(message))])
|
52 |
predicted_label = model.predict(tokenized_utterance)
|
53 |
predicted_probabilities = model.predict_proba(tokenized_utterance)
|
mathtext_fastapi/nlu.py
CHANGED
@@ -110,7 +110,10 @@ def run_intent_classification(message_text):
|
|
110 |
]
|
111 |
|
112 |
for command in commands:
|
113 |
-
|
|
|
|
|
|
|
114 |
if ratio > 80:
|
115 |
nlu_response['data'] = command
|
116 |
nlu_response['confidence'] = ratio / 100
|
@@ -129,7 +132,7 @@ def evaluate_message_with_nlu(message_data):
|
|
129 |
"""
|
130 |
# Keeps system working with two different inputs - full and filtered @event object
|
131 |
try:
|
132 |
-
message_text = message_data['message_body']
|
133 |
except KeyError:
|
134 |
message_data = {
|
135 |
'author_id': message_data['message']['_vnd']['v1']['chat']['owner'],
|
@@ -141,11 +144,12 @@ def evaluate_message_with_nlu(message_data):
|
|
141 |
'message_inserted_at': message_data['message']['_vnd']['v1']['chat']['inserted_at'],
|
142 |
'message_updated_at': message_data['message']['_vnd']['v1']['chat']['updated_at'],
|
143 |
}
|
144 |
-
message_text = message_data['message_body']
|
145 |
|
146 |
# Run intent classification only for keywords
|
147 |
intent_api_response = run_intent_classification(message_text)
|
148 |
if intent_api_response['data']:
|
|
|
149 |
return intent_api_response
|
150 |
|
151 |
number_api_resp = text2int(message_text.lower())
|
|
|
110 |
]
|
111 |
|
112 |
for command in commands:
|
113 |
+
try:
|
114 |
+
ratio = fuzz.ratio(command, message_text.lower())
|
115 |
+
except:
|
116 |
+
ratio = 0
|
117 |
if ratio > 80:
|
118 |
nlu_response['data'] = command
|
119 |
nlu_response['confidence'] = ratio / 100
|
|
|
132 |
"""
|
133 |
# Keeps system working with two different inputs - full and filtered @event object
|
134 |
try:
|
135 |
+
message_text = str(message_data['message_body'])
|
136 |
except KeyError:
|
137 |
message_data = {
|
138 |
'author_id': message_data['message']['_vnd']['v1']['chat']['owner'],
|
|
|
144 |
'message_inserted_at': message_data['message']['_vnd']['v1']['chat']['inserted_at'],
|
145 |
'message_updated_at': message_data['message']['_vnd']['v1']['chat']['updated_at'],
|
146 |
}
|
147 |
+
message_text = str(message_data['message_body'])
|
148 |
|
149 |
# Run intent classification only for keywords
|
150 |
intent_api_response = run_intent_classification(message_text)
|
151 |
if intent_api_response['data']:
|
152 |
+
prepare_message_data_for_logging(message_data, intent_api_response)
|
153 |
return intent_api_response
|
154 |
|
155 |
number_api_resp = text2int(message_text.lower())
|
pyproject.toml
CHANGED
@@ -31,6 +31,7 @@ supabase = "*"
|
|
31 |
uvicorn = "0.17.*"
|
32 |
pandas = "^1.5.3"
|
33 |
scipy = "^1.10.1"
|
|
|
34 |
|
35 |
[tool.poetry.group.dev.dependencies]
|
36 |
pytest = "^7.2"
|
|
|
31 |
uvicorn = "0.17.*"
|
32 |
pandas = "^1.5.3"
|
33 |
scipy = "^1.10.1"
|
34 |
+
sentry_sdk = "^1.19.1"
|
35 |
|
36 |
[tool.poetry.group.dev.dependencies]
|
37 |
pytest = "^7.2"
|
requirements.txt
CHANGED
@@ -3,7 +3,7 @@ en-core-web-sm @ https://github.com/explosion/spacy-models/releases/download/en_
|
|
3 |
fuzzywuzzy
|
4 |
jsonpickle
|
5 |
mathtext @ git+https://gitlab.com/tangibleai/community/mathtext@main
|
6 |
-
mathactive @ git+https://gitlab.com/tangibleai/community/mathactive@
|
7 |
fastapi
|
8 |
pydantic
|
9 |
requests
|
@@ -11,6 +11,7 @@ sentencepiece
|
|
11 |
openpyxl
|
12 |
python-Levenshtein
|
13 |
sentence-transformers
|
|
|
14 |
supabase
|
15 |
transitions
|
16 |
uvicorn
|
|
|
3 |
fuzzywuzzy
|
4 |
jsonpickle
|
5 |
mathtext @ git+https://gitlab.com/tangibleai/community/mathtext@main
|
6 |
+
mathactive @ git+https://gitlab.com/tangibleai/community/mathactive@vlad
|
7 |
fastapi
|
8 |
pydantic
|
9 |
requests
|
|
|
11 |
openpyxl
|
12 |
python-Levenshtein
|
13 |
sentence-transformers
|
14 |
+
sentry-sdk[fastapi]
|
15 |
supabase
|
16 |
transitions
|
17 |
uvicorn
|
scripts/env_loader.py
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import dotenv
|
2 |
+
import os
|
3 |
+
|
4 |
+
dotenv.load_dotenv(".env")
|
5 |
+
envs = {}
|
6 |
+
|
7 |
+
|
8 |
+
for var in ["SENTRY_URL"]:
|
9 |
+
envs[var] = os.getenv(var)
|
scripts/logger.py
ADDED
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import sentry_sdk
|
2 |
+
from . import env_loader
|
3 |
+
|
4 |
+
sentry_sdk.init(
|
5 |
+
dsn=env_loader.envs.get("SENTRY_URL"),
|
6 |
+
|
7 |
+
# Set traces_sample_rate to 1.0 to capture 100%
|
8 |
+
# of transactions for performance monitoring.
|
9 |
+
# We recommend adjusting this value in production.
|
10 |
+
traces_sample_rate=1.0
|
11 |
+
)
|
scripts/make_request.py
CHANGED
@@ -66,21 +66,19 @@ def run_simulated_request(endpoint, sample_answer, context=None):
|
|
66 |
# run_simulated_request('intent-classification', "I'm not sure")
|
67 |
# run_simulated_request('sentiment-analysis', 'I reject it')
|
68 |
# run_simulated_request('text2int', 'seven thousand nine hundred fifty seven')
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
# run_simulated_request('v2/manager', '')
|
83 |
-
# run_simulated_request('v2/manager', '5')
|
84 |
# run_simulated_request('manager', '')
|
85 |
# run_simulated_request('manager', 'add')
|
86 |
# run_simulated_request('manager', 'subtract')
|
|
|
66 |
# run_simulated_request('intent-classification', "I'm not sure")
|
67 |
# run_simulated_request('sentiment-analysis', 'I reject it')
|
68 |
# run_simulated_request('text2int', 'seven thousand nine hundred fifty seven')
|
69 |
+
run_simulated_request('nlu', 'test message')
|
70 |
+
run_simulated_request('nlu', 'eight')
|
71 |
+
run_simulated_request('nlu', 'is it 8')
|
72 |
+
run_simulated_request('nlu', 'can I know how its 0.5')
|
73 |
+
run_simulated_request('nlu', 'eight, nine, ten')
|
74 |
+
run_simulated_request('nlu', '8, 9, 10')
|
75 |
+
run_simulated_request('nlu', '8')
|
76 |
+
run_simulated_request('nlu', "I don't know")
|
77 |
+
run_simulated_request('nlu', "I don't know eight")
|
78 |
+
run_simulated_request('nlu', "I don't 9")
|
79 |
+
run_simulated_request('nlu', "0.2")
|
80 |
+
run_simulated_request('nlu', 'Today is a wonderful day')
|
81 |
+
run_simulated_request('nlu', 'IDK 5?')
|
|
|
|
|
82 |
# run_simulated_request('manager', '')
|
83 |
# run_simulated_request('manager', 'add')
|
84 |
# run_simulated_request('manager', 'subtract')
|