File size: 26,103 Bytes
d68e12a c26cd3f d68e12a c26cd3f d68e12a b62e4ec d68e12a 9b08ac4 d68e12a 9b08ac4 d68e12a 9b08ac4 d68e12a b78f40d 9b08ac4 d68e12a 50d0005 d68e12a 499b4d5 c26cd3f 1974ab3 c26cd3f 1974ab3 c26cd3f 499b4d5 d21581b 499b4d5 8304537 499b4d5 8304537 499b4d5 8304537 499b4d5 8304537 499b4d5 8304537 499b4d5 d5a1ba3 4e8dd01 d5a1ba3 4e8dd01 d5a1ba3 4e8dd01 d5a1ba3 4e8dd01 d5a1ba3 4e8dd01 499b4d5 d5a1ba3 499b4d5 d5a1ba3 499b4d5 d5a1ba3 d21581b c26cd3f dcfac75 57d9a46 d68e12a 5034cbf 50d0005 d21581b 5034cbf d21581b 50d0005 5034cbf 50d0005 5034cbf 50d0005 5034cbf 50d0005 5034cbf d68e12a 9b08ac4 d68e12a 50d0005 9b08ac4 50d0005 f44bd60 b78f40d f44bd60 9b08ac4 50d0005 f44bd60 d68e12a b78f40d 9b08ac4 d68e12a bb8f6fb d68e12a 6404599 d68e12a 6404599 6f1251a d68e12a 5034cbf 6f1251a 5034cbf 6f1251a d68e12a 6f1251a 5034cbf 6f1251a d68e12a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 |
import os
import io
import sys
import re
import traceback
import subprocess
import warnings
import gradio as gr
import pandas as pd
from dotenv import load_dotenv
from crewai import Crew, Agent, Task, Process, LLM
from crewai_tools import FileReadTool
from pydantic import BaseModel, Field
# Filter out specific warnings
warnings.filterwarnings('ignore', category=FutureWarning, module='yfinance.*')
# Load environment variables
load_dotenv()
# Get API key from environment variables
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
if not OPENAI_API_KEY:
raise ValueError("OPENAI_API_KEY environment variable not set")
llm = LLM(
model="openai/gpt-4o",
api_key=OPENAI_API_KEY,
temperature=0.7
)
# 1) Query parser agent
query_parser_agent = Agent(
role="Stock Data Analyst",
goal="Extract stock details and fetch required data from this user query: {query}.",
backstory="You are a financial analyst specializing in stock market data retrieval.",
llm=llm,
verbose=True,
memory=True,
)
# Need to define QueryAnalysisOutput class here as it's used by the task
class QueryAnalysisOutput(BaseModel):
"""Structured output for the query analysis task."""
symbols: list[str] = Field(
...,
json_schema_extra={"description": "List of stock ticker symbols (e.g., ['TSLA', 'AAPL'])."}
)
timeframe: str = Field(
...,
json_schema_extra={"description": "Time period (e.g., '1d', '1mo', '1y')."}
)
action: str = Field(
...,
json_schema_extra={"description": "Action to be performed (e.g., 'fetch', 'plot')."}
)
query_parsing_task = Task(
description="Analyze the user query and extract stock details.",
expected_output="A dictionary with keys: 'symbol', 'timeframe', 'action'.",
output_pydantic=QueryAnalysisOutput,
agent=query_parser_agent,
)
# 2) Code writer agent
code_writer_agent = Agent(
role="Senior Python Developer",
goal="Write Python code to visualize stock data.",
backstory="""You are a Senior Python developer specializing in stock market data visualization.
You are also a Pandas, Matplotlib and yfinance library expert.
You are skilled at writing production-ready Python code.
Ensure the code handles potential variations in the DataFrame structure returned by yfinance,
especially for different timeframes or delisted stocks.
Crucially, ensure the generated script saves any generated plot as 'plot.png' using `plt.savefig('plot.png')` before the script ends.""",
llm=llm,
verbose=True,
)
code_writer_task = Task(
description="""Write Python code to visualize stock data based on the inputs from the stock analyst
where you would find stock symbol, timeframe and action.""",
expected_output="A clean and executable Python script file (.py) for stock visualization.",
agent=code_writer_agent,
)
# 3) Code output agent (instead of execution agent)
code_output_agent = Agent(
role="Python Code Presenter",
goal="Present the generated Python code for stock visualization.",
backstory="You are an expert in presenting Python code in a clear and readable format.",
allow_delegation=False, # This agent just presents the code
llm=llm,
verbose=True,
)
code_output_task = Task(
description="""Receive the Python code for stock visualization from the code writer agent and present it.""",
expected_output="The complete Python script for stock visualization.",
agent=code_output_agent,
)
crew = Crew(
agents=[query_parser_agent, code_writer_agent, code_output_agent], # Use code_output_agent
tasks=[query_parsing_task, code_writer_task, code_output_task], # Use code_output_task
process=Process.sequential
)
def run_crewai_process(user_query, model, temperature):
"""
Runs the CrewAI process, captures agent thoughts, gets generated code,
executes the code, and returns results, including plot.
Args:
user_query (str): The user's query for the CrewAI process.
model (str): The model to use for the LLM.
temperature (float): The temperature to use for the LLM.
Yields:
tuple: A tuple containing the agent thoughts (str), the final answer (list of dicts),
the generated code (str), the execution output (str), and plot file path (str or None).
"""
# Create a string buffer to capture stdout
output_buffer = io.StringIO()
original_stdout = sys.stdout
sys.stdout = output_buffer
agent_thoughts = ""
generated_code = ""
execution_output = ""
generated_plot_path = None
final_answer_chat = [{"role": "user", "content": user_query}]
try:
# Initial status update with proper message format
initial_message = {"role": "assistant", "content": "Starting CrewAI process..."}
final_answer_chat = [{"role": "user", "content": str(user_query)}, initial_message]
yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None
# Run the crew process
final_result = crew.kickoff(inputs={"query": user_query})
# Get the captured CrewAI output (agent thoughts)
agent_thoughts = output_buffer.getvalue()
# Update with processing message
processing_message = {"role": "assistant", "content": "Processing complete. Generating code..."}
final_answer_chat = [{"role": "user", "content": str(user_query)}, processing_message]
yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None
# The final result is the generated code from the code_output_agent
generated_code_raw = str(final_result).strip()
# Use regex to extract the code block
code_match = re.search(r"```python\n(.*?)\n```", generated_code_raw, re.DOTALL)
if code_match:
generated_code = code_match.group(1).strip()
else:
# If no code block is found, assume the entire output is code (or handle as error)
generated_code = generated_code_raw
if not generated_code.strip(): # Handle cases where output is empty or just whitespace
execution_output = "CrewAI process completed, but no code was generated."
final_answer_chat.append({"role": "assistant", "content": execution_output})
yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path
return # Exit the generator
# Format for Gradio Chatbot (list of dictionaries with 'role' and 'content' keys only)
code_gen_message = {"role": "assistant", "content": "Code generation complete. See the 'Generated Code' box. Attempting to execute code..."}
final_answer_chat = [{"role": "user", "content": str(user_query)}, code_gen_message]
yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None
# --- Execute the generated code ---
# Check for common plot filename patterns
# First check for symbol-specific plots (e.g., META_plot.png)
symbol_plot_patterns = ['META_plot.png', 'AAPL_plot.png', 'MSFT_plot.png', 'GOOG_plot.png', 'TSLA_plot.png']
# Also check for generic plot filenames
generic_plot_patterns = ['plot.png', 'output.png', 'figure.png']
# Combine all patterns to check
plot_file_paths = symbol_plot_patterns + generic_plot_patterns
if generated_code:
try:
# Write the generated code to a temporary file
temp_script_path = "generated_script.py"
with open(temp_script_path, "w") as f:
f.write(generated_code)
# Read the generated script
with open(temp_script_path, 'r') as f:
script_content = f.read()
# Update yf.download() calls to include auto_adjust parameter
def add_auto_adjust(match):
# Check if auto_adjust is already in the arguments
args = match.group(1).strip()
if 'auto_adjust' not in args:
# Add auto_adjust=True to the arguments
if args.endswith(','):
return f'yf.download({args} auto_adjust=True)'
elif args: # If there are existing arguments
return f'yf.download({args}, auto_adjust=True)'
else: # If no arguments
return 'yf.download(auto_adjust=True)'
return match.group(0) # Return unchanged if auto_adjust is already present
# This pattern matches yf.download() with any arguments
script_content = re.sub(
r'yf\.download\(([^)]*)\)',
add_auto_adjust,
script_content
)
# Add helper functions at the beginning of the script
helpers = """
# Standard plot filename to use
PLOT_FILENAME = 'generated_plot.png'
# Helper functions for data processing
def safe_get_column(df, column):
# Handle case where column is a tuple (e.g., from multi-index)
if isinstance(column, tuple):
column = column[0] # Take the first element of the tuple
# Convert column to string in case it's not
column = str(column)
# Try exact match first
if column in df.columns:
return df[column]
# Try case-insensitive match
try:
col_lower = column.lower()
for col in df.columns:
if str(col).lower() == col_lower:
return df[col]
except (AttributeError, TypeError):
pass # Skip case-insensitive matching if not applicable
# If not found, try common variations
variations = {
'close': ['Close', 'Adj Close', 'close', 'adj close', 'CLOSE', 'Adj. Close'],
'adj close': ['Adj Close', 'adj close', 'ADJ CLOSE', 'Close', 'close', 'CLOSE', 'Adj. Close']
}
for var_list in variations.values():
for var in var_list:
if var in df.columns:
return df[var]
# If still not found, try to find any column containing 'close'
for col in df.columns:
if 'close' in str(col).lower():
return df[col]
# If still not found, raise a helpful error
raise KeyError(f"Column '{column}' not found in DataFrame. Available columns: {list(df.columns)}")
def show_plot(plt):
try:
# Use a non-interactive backend to avoid GUI issues
import matplotlib
matplotlib.use('Agg')
# Create a temporary file to store the plot
import io
import base64
# Save plot to a bytes buffer
buf = io.BytesIO()
plt.savefig(buf, format='png', bbox_inches='tight', dpi=100)
plt.close()
# Convert to base64 for display
buf.seek(0)
img_str = base64.b64encode(buf.read()).decode('utf-8')
buf.close()
# Return HTML to display the image
return f'<img src="data:image/png;base64,{img_str}" />'
except Exception as e:
print(f"[ERROR] Failed to display plot: {str(e)}")
return None
# Monkey patch DataFrame to add safe column access
import pandas as pd
pd.DataFrame.safe_get = safe_get_column
"""
# Insert the helper functions after imports
if 'import ' in script_content:
# Insert after the last import
last_import = script_content.rfind('import ')
insert_pos = script_content.find('\n', last_import) + 1
script_content = script_content[:insert_pos] + '\n' + helpers + script_content[insert_pos:]
else:
# Insert at the beginning if no imports found
script_content = helpers + '\n' + script_content
# Replace common column access patterns with our safe version
script_content = script_content.replace("['Adj Close']", ".safe_get('close')")
script_content = script_content.replace("['Close']", ".safe_get('close')")
script_content = script_content.replace("['close']", ".safe_get('close')")
# Replace plt.show() calls with our helper
script_content = re.sub(
r'plt\\.show\\(\s*\\)',
r'print(show_plot(plt))',
script_content
)
# If no show() call is found, add one at the end of the script
if 'plt.show()' not in script_content:
script_content += "\n# Display the plot if any figures exist\nif 'plt' in locals() and len(plt.get_fignums()) > 0:\n print(show_plot(plt))\n"
# Write the updated script back
with open(temp_script_path, 'w') as f:
f.write(script_content)
# Execute the temporary script using subprocess
# Use python3 to ensure correct interpreter in Colab
process = subprocess.run(
["python3", temp_script_path],
capture_output=True,
text=True, # Capture stdout and stderr as text
check=False # Don't raise exception for non-zero exit codes
)
execution_output = process.stdout + process.stderr
# Check for specific errors in execution output
if "KeyError" in execution_output:
execution_output += "\n\nPotential Issue: The generated script encountered a KeyError. This might mean the script tried to access a column or data point that wasn't available for the specified stock or timeframe. Please try a different query or timeframe."
elif "FileNotFoundError: [Errno 2] No such file or directory: 'plot.png'" in execution_output and "Figure(" in execution_output:
execution_output += "\n\nPlot Generation Issue: The script seems to have created a plot but did not save it to 'plot.png'. Please ensure the generated code includes `plt.savefig('plot.png')`."
elif "FileNotFoundError: [Errno 2] No such file or directory: 'plot.png'" in execution_output:
execution_output += "\n\nPlot Generation Issue: The script ran, but the plot file was not created. Ensure the generated code includes commands to save the plot to 'plot.png'."
# Check for the generated plot file in all possible locations
generated_plot_path = None
plot_found = False
# First check the standard plot filename
plot_file_paths = ['generated_plot.png', 'plot.png', 'META_plot.png', 'AAPL_plot.png', 'MSFT_plot.png', 'output.png']
# Also check for any .png files in the current directory
current_dir = os.path.abspath('.')
png_files = [f for f in os.listdir(current_dir)
if f.endswith('.png') and not f.startswith('gradio_')]
# Add any found .png files to our search paths
plot_file_paths.extend(png_files)
# Make paths absolute and remove duplicates
plot_file_paths = list(dict.fromkeys([os.path.abspath(f) for f in plot_file_paths]))
print(f"[DEBUG] Looking for plot files in: {plot_file_paths}")
for plot_file in plot_file_paths:
try:
if os.path.exists(plot_file) and os.path.getsize(plot_file) > 0:
print(f"[DEBUG] Found plot file: {plot_file}")
generated_plot_path = plot_file
plot_found = True
break
except Exception as e:
print(f"[DEBUG] Error checking plot file {plot_file}: {e}")
# Add the plot to the execution output
try:
import base64
with open(plot_abs_path, 'rb') as img_file:
img_str = base64.b64encode(img_file.read()).decode('utf-8')
execution_output += f"\n\n"
except Exception as e:
execution_output += f"\n\nNote: Could not embed plot in output: {str(e)}"
break
if not plot_found:
# If no plot file was found, check the current directory for any .png files
current_dir = os.path.abspath('.')
png_files = [f for f in os.listdir(current_dir) if f.endswith('.png') and not f.startswith('gradio_')]
if png_files:
# Use the first .png file found
plot_abs_path = os.path.abspath(png_files[0])
generated_plot_path = plot_abs_path
print(f"Using plot file found at: {plot_abs_path}")
# Add the plot to the execution output
try:
import base64
with open(plot_abs_path, 'rb') as img_file:
img_str = base64.b64encode(img_file.read()).decode('utf-8')
execution_output += f"\n\n"
except Exception as e:
execution_output += f"\n\nNote: Could not embed plot in output: {str(e)}"
else:
print(f"No plot file found in {current_dir}")
execution_output += "\nNo plot file was found after execution.\n\nMake sure the generated code includes:\n1. `plt.savefig('plot.png')` to save the plot\n2. `plt.close()` to close the figure after saving"
except Exception as e:
traceback_str = traceback.format_exc()
execution_output = f"An error occurred during code execution: {e}\n{traceback_str}"
finally:
# Clean up the temporary script file
if os.path.exists(temp_script_path):
os.remove(temp_script_path)
else:
execution_output = "No code was generated to execute."
# Update final answer chat to reflect execution attempt
execution_complete_msg = "Code execution finished. See 'Execution Output'."
if generated_plot_path:
plot_msg = f"Plot generated successfully at: {generated_plot_path}"
final_answer_chat = [
{"role": "user", "content": str(user_query)},
{"role": "assistant", "content": execution_complete_msg},
{"role": "assistant", "content": plot_msg}
]
# Return the plot path to be displayed
yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, generated_plot_path
else:
no_plot_msg = "No plot was generated. Make sure your query includes a request for a visualization. Check the 'Execution Output' tab for any errors."
final_answer_chat = [
{"role": "user", "content": str(user_query)},
{"role": "assistant", "content": execution_complete_msg},
{"role": "assistant", "content": no_plot_msg}
]
# Return None for plot path
yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None
yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path
except Exception as e:
# If an error occurs during CrewAI process, return the error message
traceback_str = traceback.format_exc()
agent_thoughts += f"\nAn error occurred during CrewAI process: {e}\n{traceback_str}"
error_message = f"An error occurred during CrewAI process: {e}"
final_answer_chat = [
{"role": "user", "content": str(user_query)},
{"role": "assistant", "content": error_message}
]
yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None
finally:
# Restore original stdout
sys.stdout = original_stdout
def create_interface():
"""Create and return the Gradio interface."""
with gr.Blocks(title="Financial Analytics Agent", theme=gr.themes.Soft()) as interface:
gr.Markdown("# 📊 Financial Analytics Agent")
gr.Markdown("Enter your financial query to analyze stock data and generate visualizations.")
with gr.Row():
with gr.Column(scale=2):
user_query_input = gr.Textbox(
label="Enter your financial query",
placeholder="e.g., Show me the stock performance of AAPL and MSFT for the last year",
lines=3
)
submit_btn = gr.Button("Analyze", variant="primary")
with gr.Accordion("Advanced Options", open=False):
gr.Markdown("### Model Settings")
model_dropdown = gr.Dropdown(
["gpt-4o", "gpt-4-turbo", "gpt-3.5-turbo"],
value="gpt-4o",
label="Model"
)
temperature = gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.7,
step=0.1,
label="Creativity (Temperature)"
)
with gr.Column(scale=3):
with gr.Tabs():
with gr.TabItem("Analysis"):
final_answer_chat = gr.Chatbot(
label="Analysis Results",
height=300,
show_copy_button=True,
type="messages" # Explicitly set to use OpenAI-style message format
)
with gr.TabItem("Agent Thoughts"):
agent_thoughts = gr.Textbox(
label="Agent Thinking Process",
interactive=False,
lines=15,
max_lines=30,
show_copy_button=True
)
with gr.TabItem("Generated Code"):
generated_code = gr.Code(
label="Generated Python Code",
language="python",
interactive=False,
lines=15
)
with gr.TabItem("Execution Output"):
execution_output = gr.Textbox(
label="Code Execution Output",
interactive=False,
lines=10,
show_copy_button=True
)
with gr.Row():
with gr.Column():
plot_output = gr.Plot(
label="Generated Visualization",
visible=False
)
image_output = gr.Image(
label="Generated Plot",
type="filepath",
visible=False
)
# Handle form submission
inputs = [user_query_input, model_dropdown, temperature]
outputs = [
final_answer_chat,
agent_thoughts,
generated_code,
execution_output,
plot_output, # Pass the actual component, not None
image_output # Pass the actual component, not None
]
def process_results(chat, thoughts, code, output, plot_path):
# This function will be called after run_crewai_process
# Show the image in the image_output component
return [
chat,
thoughts,
code,
output,
gr.update(visible=plot_path is not None and os.path.exists(plot_path)),
gr.update(value=plot_path if (plot_path and os.path.exists(plot_path)) else None,
visible=plot_path is not None and os.path.exists(plot_path))
]
# First, run the crewAI process
click_event = submit_btn.click(
fn=run_crewai_process,
inputs=inputs,
outputs=outputs,
api_name="analyze"
)
# Then update the UI with the results
click_event.then(
fn=process_results,
inputs=[final_answer_chat, agent_thoughts, generated_code, execution_output, image_output],
outputs=outputs
)
return interface
def main():
"""Run the Gradio interface."""
interface = create_interface()
interface.launch(share=False, server_name="0.0.0.0", server_port=7860)
if __name__ == "__main__":
main() |