llm agent added

utils/helper.py

@@ -1,9 +1,14 @@
 # helper.py
-import yfinance as yf
+import os
+from typing import Dict, List, Tuple
+
 import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-from typing import List, Tuple, Dict
+import yfinance as yf
+from langchain import LLMChain, PromptTemplate
+from langchain.agents import initialize_agent, load_tools
+from langchain.llms import OpenAI
 
 
 def download_stock_data(
@@ -283,3 +288,58 @@ def display_simulated_ef_with_random(
             min_vol_allocation.sum(axis=1), axis=0
         ),
     }
+
+
+OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]
+
+
+def run_langchain_agent_(
+    question: str = "What is your question?", interested_tickers: str = "AAPL, META"
+) -> str:
+    """
+    Executes a language chain agent to answer questions by using a series of tools.
+
+    This function creates an instance of an OpenAI model, sets up a prompt template, loads necessary tools,
+    initializes the agent, and runs the agent with the provided question. It returns the agent's output.
+
+    Parameters:
+        question (str): The question to be answered by the agent. Defaults to "What is your question?".
+
+    Returns:
+        str: The output response from the agent after processing the question.
+
+    """
+
+    # Instantiating an OpenAI language model with specific temperature setting
+    llm = OpenAI(
+        temperature=0.1
+    )  # model_name="text-davinci-003" implied but commented out
+
+    # Creating a prompt template that structures the input question and a step-by-step thinking format
+    template = """Question: {question};
+    Information given about interested stock tickers in the financial market: {interested_tickers}
+
+    You are a financial advisor and user has a question above regarding related tickers provided.
+    
+    Let's think step by step.
+    Answer: """
+    prompt = PromptTemplate(
+        template=template, input_variables=["question", "interested_tickers"]
+    )
+
+    # Building a chain of language model actions based on the prompt template
+    llm_chain = LLMChain(prompt=prompt, llm=llm)
+
+    # Loading additional language model tools like Wikipedia and math modules
+    tools = load_tools(["wikipedia", "llm-math"], llm=llm)
+
+    # Initializing the agent with the loaded tools, the language model, default name, and verbosity
+    agent = initialize_agent(
+        tools, llm, agent="zero-shot-react-description", verbose=True
+    )
+
+    # Running the agent to process the input question and generate an output
+    output_ = agent.run(question)
+
+    # Return
+    return output_

utils/ui_helper.py

@@ -171,7 +171,7 @@ def main_algo_trader():
                 # Checkpoint: ask user whether they want portfolio weights
                 if csv:
                     recent_selected_stocks = df["portfolio_history"][-1]
-                    recent_selected_stocks = ', '.join(recent_selected_stocks)
+                    recent_selected_stocks = ", ".join(recent_selected_stocks)
                     st.success(
                         f"The algorithm suggests to hold the following stocks for the current month (equally weighted): {recent_selected_stocks}"
                     )
@@ -185,6 +185,20 @@ def main_algo_trader():
                     mime="text/csv",
                 )
 
+            # Question-answering
+            if csv:
+                user_question = st.text_input(
+                    "Enter your question:",
+                    "Tell me about the company with stock ticker AAPL.",
+                )
+                st.write(f"Question entered: {user_question}")
+                with st.spinner("Running langchain agent with GPT3..."):
+                    final_ans = run_langchain_agent_(
+                        question=str(user_question),
+                        interested_tickers=str(recent_selected_stocks),
+                    )
+                    st.markdown(final_ans)
+
 
 # chinese
 def main_algo_trader_chinese():
@@ -348,7 +362,7 @@ def main_algo_trader_chinese():
                 # Checkpoint: ask user whether they want portfolio weights
                 if csv:
                     recent_selected_stocks = df["portfolio_history"][-1]
-                    recent_selected_stocks = ', '.join(recent_selected_stocks)
+                    recent_selected_stocks = ", ".join(recent_selected_stocks)
                     st.success(f"算法建议在本月持有以下股票（均仓位）: {recent_selected_stocks}")
 
             with col2: