Spaces:
Sleeping
Sleeping
Added new endpoint for Income/Expenses prediction (to be updated along the way)
Browse files
main.py
CHANGED
|
@@ -4,152 +4,137 @@ import os
|
|
| 4 |
import io
|
| 5 |
from flask import Flask, request, jsonify
|
| 6 |
from flask_cors import CORS, cross_origin
|
| 7 |
-
import pandas as pd
|
| 8 |
import firebase_admin
|
| 9 |
-
from firebase_admin import credentials, firestore
|
| 10 |
-
import
|
| 11 |
-
|
| 12 |
-
from
|
| 13 |
-
import os
|
| 14 |
-
from pandasai.llm import GoogleGemini
|
| 15 |
-
from pandasai import SmartDataframe, SmartDatalake
|
| 16 |
-
|
| 17 |
-
from pandasai.responses.response_parser import ResponseParser
|
| 18 |
-
import matplotlib.pyplot as plt
|
| 19 |
-
from wordcloud import WordCloud
|
| 20 |
-
import random
|
| 21 |
from langchain.prompts import PromptTemplate
|
| 22 |
from langchain.chains import LLMChain
|
| 23 |
-
from
|
| 24 |
-
import
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
from dotenv import load_dotenv
|
| 28 |
|
| 29 |
load_dotenv()
|
| 30 |
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
app = Flask(__name__)
|
| 35 |
cors = CORS(app)
|
| 36 |
|
| 37 |
-
class FlaskResponse(ResponseParser):
|
| 38 |
-
def __init__(self, context) -> None:
|
| 39 |
-
super().__init__(context)
|
| 40 |
-
|
| 41 |
-
def format_dataframe(self, result):
|
| 42 |
-
return result['value'].to_html()
|
| 43 |
-
|
| 44 |
-
def format_plot(self, result):
|
| 45 |
-
# Save the plot using savefig
|
| 46 |
-
try:
|
| 47 |
-
|
| 48 |
-
img_path = result['value']
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
except ValueError:
|
| 52 |
-
img_path = str(result['value'])
|
| 53 |
-
print("value error!", img_path)
|
| 54 |
-
|
| 55 |
-
print("response_class_path:", img_path)
|
| 56 |
-
return img_path
|
| 57 |
-
|
| 58 |
-
def format_other(self, result):
|
| 59 |
-
return str(result['value'])
|
| 60 |
-
|
| 61 |
-
gemini_api_key = os.environ['Gemini']
|
| 62 |
-
|
| 63 |
-
@app.route("/", methods=["GET"])
|
| 64 |
-
def home():
|
| 65 |
-
|
| 66 |
-
return "Hello Qx!"
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
llm = GoogleGemini(api_key=gemini_api_key)
|
| 71 |
-
|
| 72 |
-
llm2 = ChatGoogleGenerativeAI(model='gemini-1.5-flash-001', temperature=0.1)
|
| 73 |
-
|
| 74 |
# Initialize Firebase app
|
| 75 |
if not firebase_admin._apps:
|
| 76 |
-
|
| 77 |
cred = credentials.Certificate("quant-app-99d09-firebase-adminsdk-6prb1-37f34e1c91.json")
|
| 78 |
firebase_admin.initialize_app(cred)
|
| 79 |
|
| 80 |
db = firestore.client()
|
| 81 |
|
|
|
|
|
|
|
|
|
|
| 82 |
|
|
|
|
|
|
|
| 83 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
@app.route("/predict", methods=["POST"])
|
| 86 |
@cross_origin()
|
| 87 |
def bot():
|
| 88 |
-
|
| 89 |
-
|
| 90 |
user_id = request.json.get("user_id")
|
| 91 |
user_question = request.json.get("user_question")
|
| 92 |
-
load_dotenv()
|
| 93 |
-
|
| 94 |
-
inventory_ref = db.collection("system_users").document(user_id).collection('inventory')
|
| 95 |
|
|
|
|
| 96 |
tasks_ref = db.collection("system_users").document(user_id).collection('tasks')
|
| 97 |
-
|
| 98 |
transactions_ref = db.collection("system_users").document(user_id).collection('transactions')
|
| 99 |
|
| 100 |
-
inventory_list = []
|
| 101 |
-
for doc in
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
tasks_list = []
|
| 106 |
-
for doc in tasks_ref.stream():
|
| 107 |
-
a = doc.to_dict()
|
| 108 |
-
tasks_list.append(a)
|
| 109 |
-
|
| 110 |
-
transactions_list = []
|
| 111 |
-
for doc in transactions_ref.stream():
|
| 112 |
-
a = doc.to_dict()
|
| 113 |
-
transactions_list.append(a)
|
| 114 |
-
|
| 115 |
inventory_df = pd.DataFrame(inventory_list)
|
| 116 |
transactions_df = pd.DataFrame(transactions_list)
|
| 117 |
tasks_df = pd.DataFrame(tasks_list)
|
| 118 |
|
| 119 |
-
lake = SmartDatalake([inventory_df, transactions_df, tasks_df], config={"llm":llm, "response_parser":FlaskResponse, "enable_cache": False, "save_logs":False})
|
| 120 |
response = lake.chat(user_question)
|
| 121 |
print(user_question)
|
| 122 |
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
return jsonify(resp)
|
| 126 |
|
|
|
|
| 127 |
@app.route("/mrec", methods=["POST"])
|
| 128 |
@cross_origin()
|
| 129 |
def marketing_rec():
|
| 130 |
user_id = request.json.get("user_id")
|
| 131 |
|
| 132 |
transactions_ref = db.collection("system_users").document(user_id).collection('transactions')
|
| 133 |
-
transactions_list = []
|
| 134 |
-
for doc in transactions_ref.stream():
|
| 135 |
-
a = doc.to_dict()
|
| 136 |
-
transactions_list.append(a)
|
| 137 |
|
| 138 |
transactions_df = pd.DataFrame(transactions_list)
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
chain = LLMChain(llm=llm2, prompt=prompt, verbose=True)
|
| 144 |
-
data_frame = transactions_df
|
| 145 |
-
response = chain.invoke(input=data_frame)
|
| 146 |
print(response)
|
| 147 |
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
return jsonify(resp)
|
| 151 |
-
|
| 152 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
|
| 154 |
if __name__ == "__main__":
|
| 155 |
-
app.run(debug=True,host="0.0.0.0", port=7860)
|
|
|
|
| 4 |
import io
|
| 5 |
from flask import Flask, request, jsonify
|
| 6 |
from flask_cors import CORS, cross_origin
|
|
|
|
| 7 |
import firebase_admin
|
| 8 |
+
from firebase_admin import credentials, firestore
|
| 9 |
+
from dotenv import load_dotenv
|
| 10 |
+
from pandasai import SmartDatalake
|
| 11 |
+
from pandasai.responses.response_parser import ResponseParser
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
from langchain.prompts import PromptTemplate
|
| 13 |
from langchain.chains import LLMChain
|
| 14 |
+
from datetime import datetime
|
| 15 |
+
import matplotlib.pyplot as plt
|
| 16 |
+
from statsmodels.tsa.holtwinters import ExponentialSmoothing
|
| 17 |
+
from fbprophet import Prophet
|
|
|
|
| 18 |
|
| 19 |
load_dotenv()
|
| 20 |
|
|
|
|
|
|
|
|
|
|
| 21 |
app = Flask(__name__)
|
| 22 |
cors = CORS(app)
|
| 23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
# Initialize Firebase app
|
| 25 |
if not firebase_admin._apps:
|
|
|
|
| 26 |
cred = credentials.Certificate("quant-app-99d09-firebase-adminsdk-6prb1-37f34e1c91.json")
|
| 27 |
firebase_admin.initialize_app(cred)
|
| 28 |
|
| 29 |
db = firestore.client()
|
| 30 |
|
| 31 |
+
class FlaskResponse(ResponseParser):
|
| 32 |
+
def __init__(self, context) -> None:
|
| 33 |
+
super().__init__(context)
|
| 34 |
|
| 35 |
+
def format_dataframe(self, result):
|
| 36 |
+
return result['value'].to_html()
|
| 37 |
|
| 38 |
+
def format_plot(self, result):
|
| 39 |
+
try:
|
| 40 |
+
img_path = result['value']
|
| 41 |
+
except ValueError:
|
| 42 |
+
img_path = str(result['value'])
|
| 43 |
+
print("ValueError:", img_path)
|
| 44 |
+
print("response_class_path:", img_path)
|
| 45 |
+
return img_path
|
| 46 |
|
| 47 |
+
def format_other(self, result):
|
| 48 |
+
return str(result['value'])
|
| 49 |
+
|
| 50 |
+
gemini_api_key = os.getenv('Gemini')
|
| 51 |
+
llm = ChatGoogleGenerativeAI(api_key=gemini_api_key, model='gemini-1.5-flash-001', temperature=0.1)
|
| 52 |
+
|
| 53 |
+
# Endpoint for handling questions to the bot using transaction data
|
| 54 |
@app.route("/predict", methods=["POST"])
|
| 55 |
@cross_origin()
|
| 56 |
def bot():
|
|
|
|
|
|
|
| 57 |
user_id = request.json.get("user_id")
|
| 58 |
user_question = request.json.get("user_question")
|
|
|
|
|
|
|
|
|
|
| 59 |
|
| 60 |
+
inventory_ref = db.collection("system_users").document(user_id).collection('inventory')
|
| 61 |
tasks_ref = db.collection("system_users").document(user_id).collection('tasks')
|
|
|
|
| 62 |
transactions_ref = db.collection("system_users").document(user_id).collection('transactions')
|
| 63 |
|
| 64 |
+
inventory_list = [doc.to_dict() for doc in inventory_ref.stream()]
|
| 65 |
+
tasks_list = [doc.to_dict() for doc in tasks_ref.stream()]
|
| 66 |
+
transactions_list = [doc.to_dict() for doc in transactions_ref.stream()]
|
| 67 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 68 |
inventory_df = pd.DataFrame(inventory_list)
|
| 69 |
transactions_df = pd.DataFrame(transactions_list)
|
| 70 |
tasks_df = pd.DataFrame(tasks_list)
|
| 71 |
|
| 72 |
+
lake = SmartDatalake([inventory_df, transactions_df, tasks_df], config={"llm": llm, "response_parser": FlaskResponse, "enable_cache": False, "save_logs": False})
|
| 73 |
response = lake.chat(user_question)
|
| 74 |
print(user_question)
|
| 75 |
|
| 76 |
+
return jsonify(str(response))
|
|
|
|
|
|
|
| 77 |
|
| 78 |
+
# Marketing recommendations endpoint
|
| 79 |
@app.route("/mrec", methods=["POST"])
|
| 80 |
@cross_origin()
|
| 81 |
def marketing_rec():
|
| 82 |
user_id = request.json.get("user_id")
|
| 83 |
|
| 84 |
transactions_ref = db.collection("system_users").document(user_id).collection('transactions')
|
| 85 |
+
transactions_list = [doc.to_dict() for doc in transactions_ref.stream()]
|
|
|
|
|
|
|
|
|
|
| 86 |
|
| 87 |
transactions_df = pd.DataFrame(transactions_list)
|
| 88 |
+
prompt = PromptTemplate.from_template('You are a business analyst. Write a brief analysis and marketing tips for a small business using this transactions data {data_frame}')
|
| 89 |
+
chain = LLMChain(llm=llm, prompt=prompt, verbose=True)
|
| 90 |
+
|
| 91 |
+
response = chain.invoke(input=transactions_df)
|
|
|
|
|
|
|
|
|
|
| 92 |
print(response)
|
| 93 |
|
| 94 |
+
return jsonify(str(response['text']))
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
+
# Income/Expenses Prediction endpoint
|
| 97 |
+
@app.route("/predict_revenue", methods=["POST"])
|
| 98 |
+
@cross_origin()
|
| 99 |
+
def predict_revenue():
|
| 100 |
+
request_data = request.json
|
| 101 |
+
user_id = request_data.get("user_id")
|
| 102 |
+
interval = request_data.get("interval", 30)
|
| 103 |
+
transaction_type = request_data.get("transaction_type", "Income")
|
| 104 |
+
|
| 105 |
+
# Fetch transaction data based on user and transaction type
|
| 106 |
+
transactions_ref = db.collection("system_users").document(user_id).collection("transactions")
|
| 107 |
+
query = transactions_ref.where("transactionType", "==", transaction_type).stream()
|
| 108 |
+
|
| 109 |
+
data = []
|
| 110 |
+
for doc in query:
|
| 111 |
+
transaction = doc.to_dict()
|
| 112 |
+
data.append({"date": transaction["date"].to_date(), "amountDue": transaction["amountDue"]})
|
| 113 |
+
|
| 114 |
+
# Create DataFrame from transaction data
|
| 115 |
+
df = pd.DataFrame(data)
|
| 116 |
+
df = df.sort_values("date").set_index("date")
|
| 117 |
+
df = df.resample("D").sum().reset_index() # Resample daily to ensure regular intervals
|
| 118 |
+
df.columns = ["ds", "y"] # Rename columns for Prophet (ds: date, y: target)
|
| 119 |
+
|
| 120 |
+
# Check if there's enough data to train the model
|
| 121 |
+
if df.shape[0] < 10:
|
| 122 |
+
return jsonify({"error": "Not enough data for prediction"})
|
| 123 |
+
|
| 124 |
+
# Initialize and fit the Prophet model
|
| 125 |
+
model = Prophet(daily_seasonality=True, yearly_seasonality=True)
|
| 126 |
+
model.fit(df)
|
| 127 |
+
|
| 128 |
+
# dataframe for future predictions
|
| 129 |
+
future_dates = model.make_future_dataframe(periods=interval)
|
| 130 |
+
forecast = model.predict(future_dates)
|
| 131 |
+
|
| 132 |
+
# Extract the forecast for the requested interval
|
| 133 |
+
forecast_data = forecast[['ds', 'yhat']].tail(interval)
|
| 134 |
+
predictions = forecast_data['yhat'].tolist()
|
| 135 |
+
|
| 136 |
+
# Return predictions in JSON format
|
| 137 |
+
return jsonify({"predictedData": predictions})
|
| 138 |
|
| 139 |
if __name__ == "__main__":
|
| 140 |
+
app.run(debug=True, host="0.0.0.0", port=7860)
|