This project is a collection of scripts that generate and submit alphas to the WorldQuant platform.

For the best performance and user experience, we recommend using the Naive-Ollama Alpha Generator which features:

• 🚀 3-5x faster alpha generation with local Ollama LLM
• 🖥️ GPU acceleration for optimal performance
• 📊 Real-time web dashboard for monitoring and control
• 🤖 Fully automated 24/7 operation
• 🔒 Local processing - no external API costs or privacy concerns
• 🐳 Docker support for easy deployment
• 📈 Advanced orchestration with intelligent scheduling

Quick Start:

cd naive-ollama
# Setup credentials in credential.txt
docker-compose -f docker-compose.gpu.yml up -d
# Access dashboard at http://localhost:5000

---

 __      __            .__       .___                          __            .__                     
/  \    /  \___________|  |    __| _/________ _______    _____/  |_    _____ |__| ____   ___________ 
\   \/\/   /  _ \_  __ \  |   / __ |/ ____/  |  \__  \  /    \   __\  /     \|  |/    \_/ __ \_  __ \
 \        (  <_> )  | \/  |__/ /_/ < <_|  |  |  // __ \|   |  \  |   |  Y Y  \  |   |  \  ___/|  | \/
  \__/\  / \____/|__|  |____/\____ |\__   |____/(____  /___|  /__|   |__|_|  /__|___|  /\___  >__|   
       \/                         \/   |__|          \/     \/             \/        \/     \/       

Discord: https://discord.gg/3B2TmHQw

Usage Tutorial(WIP) on the web version: https://www.youtube.com/watch?v=xwr9atsulSA Further usage tutorial on the local ollama version: https://www.youtube.com/watch?v=EAeujBRrKiI

This is a Rust implementation of the alpha generator.

cargo build --release

cargo run --release

This is a Python implementation of the alpha generator.

• Pre-Consultant has maximum 5 concurrent simulations
• Pre-Consultant has fewer operators and data fields options

Incoming... lol

A sophisticated alpha factor generation system that uses Ollama with financial language models to generate, test, and submit alpha factors to WorldQuant Brain. This system replaces the previous Kimi interface with a local Ollama-based solution for better performance and control.

• Local LLM Integration: Uses Ollama with llama3.2:3b or llama2:7b models
• GPU Acceleration: Full NVIDIA GPU support for faster inference
• Web Dashboard: Real-time monitoring and control interface
• Automated Orchestration: Continuous alpha generation, mining, and submission
• WorldQuant Brain Integration: Direct API integration for testing and submission
• Docker Support: Easy deployment with Docker and Docker Compose
• Daily Rate Limiting: Ensures compliance with WorldQuant submission limits

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Web Dashboard │    │ Alpha Generator │    │  WorldQuant API │
│   (Flask)       │◄──►│   (Ollama)      │◄──►│   (External)    │
│   Port 5000     │    │   Port 11434    │    │                 │
└─────────────────┘    └─────────────────┘    └─────────────────┘
         │                       │                       │
         │              ┌─────────────────┐              │
         └──────────────►│ Alpha Orchestrator │◄─────────────┘
                        │   (Python)      │
                        └─────────────────┘
                                │
                                ▼
                        ┌─────────────────┐
                        │   Results &     │
                        │   Logs Storage  │
                        └─────────────────┘

Create naive-ollama/credential.txt with your WorldQuant Brain credentials:

["your.email@worldquant.com", "your_password"]

cd naive-ollama
# Start the complete system with GPU acceleration
docker-compose -f docker-compose.gpu.yml up -d

# Or use the convenience script
start_gpu.bat

Open your browser and navigate to:

• Main Dashboard: http://localhost:5000
• Ollama WebUI: http://localhost:3000
• Ollama API: http://localhost:11434

The web dashboard provides real-time monitoring and control:

• GPU Status: Memory usage, utilization, temperature
• Ollama Status: Model loading, API connectivity
• Orchestrator Status: Generation activity, mining schedule
• WorldQuant Status: API connectivity, authentication
• Statistics: Generated alphas, success rates, 24h metrics

• Generate Alpha: Trigger single alpha generation
• Trigger Mining: Run alpha expression mining
• Trigger Submission: Submit successful alphas
• Refresh Status: Update all metrics

• Alpha Generator Logs: Filtered logs showing alpha generation activity
• System Logs: Complete system activity
• Recent Activity: Timeline of recent events

• Continuous Mode: Generates alphas every 6 hours
• Batch Processing: Generates 3 alphas per batch
• Ollama Integration: Uses local LLM for alpha idea generation
• WorldQuant Testing: Tests each alpha immediately

• Expression Mining: Analyzes promising alphas for variations
• Pattern Recognition: Identifies successful alpha patterns
• Optimization: Suggests improvements to existing alphas

• Daily Limit: Submits only once per day
• Success Filtering: Only submits alphas with good performance
• Rate Limiting: Respects WorldQuant API limits

• Before: ~10-15 seconds per alpha (Kimi API)
• After: ~3-5 seconds per alpha (Local Ollama + GPU)

• Before: Manual intervention required
• After: Fully automated 24/7 operation

naive-ollama/
├── alpha_generator_ollama.py      # Main alpha generation script
├── alpha_orchestrator.py          # Orchestration and scheduling
├── alpha_expression_miner.py      # Alpha expression mining
├── successful_alpha_submitter.py  # Alpha submission to WorldQuant
├── web_dashboard.py               # Flask web dashboard
├── templates/
│   └── dashboard.html             # Dashboard HTML template
├── results/                       # Generated alpha results
├── logs/                          # System logs
├── Dockerfile                     # Docker image definition
├── docker-compose.gpu.yml         # GPU-enabled deployment
├── docker-compose.yml             # CPU-only deployment
├── requirements.txt               # Python dependencies
├── credential.txt                 # WorldQuant credentials
├── start_gpu.bat                  # Windows GPU startup script
├── start_dashboard.bat            # Windows dashboard startup script
├── README.md                      # Detailed documentation
├── README_Docker.md               # Docker-specific documentation
└── CHANGELOG.md                   # Version history

• Python 3.8: Main application language
• Flask: Web dashboard framework
• Requests: HTTP client for APIs
• Schedule: Task scheduling
• PyTorch: GPU acceleration support

• Docker: Containerization
• Docker Compose: Multi-service orchestration
• NVIDIA CUDA: GPU acceleration
• Ollama: Local LLM serving

• HTML5/CSS3: Dashboard interface
• JavaScript: Real-time updates
• Responsive Design: Mobile-friendly layout

• Local Processing: All LLM inference happens locally
• Credential Protection: Credentials stored in mounted volume
• Network Isolation: Docker network isolation
• API Rate Limiting: Respects external API limits

For detailed documentation, see:

• naive-ollama/README.md - Main project documentation
• naive-ollama/README_Docker.md - Docker-specific documentation
• naive-ollama/CHANGELOG.md - Version history

1. Performance: 3-5x faster than Kimi API
2. Cost: No external API costs
3. Privacy: All processing happens locally
4. Control: Full control over the LLM and workflow
5. Automation: 24/7 continuous operation
6. Monitoring: Real-time web dashboard
7. Scalability: GPU acceleration support
8. Reliability: Docker containerization

This is basically a closed loop system where it starts with alpha_generator.py, which generates alpha ideas with Kimi AI. Then it dumps promising alpha to a local log file, which promising_alpha_miner.py will mine for better results, and dumps refined alpha to a local log file, which you can run successful_alpha_submitter.py to submit to the WorldQuant platform. These script can run concurrently.

While alpha_expression_miner.py is more of a utility script to mine alpha expressions from a given expression manually but not within the aforementioned closed loop system.

alpha_101_testing is currently under development.

alpha_polisher.py is currently under development.

pip install -r requirements.txt

The pre-consultant Python alpha generator uses Kimi AI to generate alpha expressions. Notice this may get you some alpha ideas as long as you are fine with the cost.

Human in the loop control is the future development direction of this script.

python alpha_generator.py

This script is used to mine alpha expressions from a given expression.

python alpha_expression_miner.py --expression "expression"


PS ~> python .\alpha_expression_miner.py --expression "cashflow_stability = ts_mean(cashflow_op, 252) / (debt_lt + 0.01);
>> stability_z = zscore(cashflow_stability);
>> debt_ratio = debt_lt / (assets + 0.01);
>> combined_score = stability_z - zscore(debt_ratio);
>> -rank(combined_score)"
2025-05-04 01:37:50,111 - INFO - Starting alpha expression mining with parameters:
2025-05-04 01:37:50,111 - INFO - Expression: cashflow_stability = ts_mean(cashflow_op, 252) / (debt_lt + 0.01);
stability_z = zscore(cashflow_stability);
debt_ratio = debt_lt / (assets + 0.01);
combined_score = stability_z - zscore(debt_ratio);
-rank(combined_score)
2025-05-04 01:37:50,111 - INFO - Output file: mined_expressions.json
2025-05-04 01:37:50,112 - INFO - Initializing AlphaExpressionMiner
2025-05-04 01:37:50,112 - INFO - Loading credentials from ./credential.txt
2025-05-04 01:37:50,112 - INFO - Authenticating with WorldQuant Brain...
2025-05-04 01:37:51,303 - INFO - Authentication response status: 201
2025-05-04 01:37:51,303 - INFO - Authentication successful
2025-05-04 01:37:51,303 - INFO - Parsing expression: cashflow_stability = ts_mean(cashflow_op, 252) / (debt_lt + 0.01);
stability_z = zscore(cashflow_stability);
debt_ratio = debt_lt / (assets + 0.01);
combined_score = stability_z - zscore(debt_ratio);
-rank(combined_score)
2025-05-04 01:37:51,303 - INFO - Found 3 parameters to vary

Found the following parameters in the expression:
1. Value: 252.0 | Context: ...s_mean(cashflow_op, 252) / (debt_lt + 0.01)...
2. Value: 0.01 | Context: ..., 252) / (debt_lt + 0.01);
stability_z = zsc...
3. Value: 0.01 | Context: ...debt_lt / (assets + 0.01);
combined_score = ...

Enter the numbers of parameters to vary (comma-separated, or 'all'): all

Parameter: 252.0 | Context: ...s_mean(cashflow_op, 252) / (debt_lt + 0.01)...
Enter range (e.g., '10' for ±10, or '5,15' for 5 to 15): 25
Enter step size: 1

Parameter: 0.01 | Context: ..., 252) / (debt_lt + 0.01);
stability_z = zsc...
Enter range (e.g., '10' for ±10, or '5,15' for 5 to 15): -0.05,0.05
Enter step size: 0.01

Parameter: 0.01 | Context: ...debt_lt / (assets + 0.01);
combined_score = ...
Enter range (e.g., '10' for ±10, or '5,15' for 5 to 15): -0.05,0.05
Enter step size: 0.01
2025-05-04 01:38:18,371 - INFO - Generating variations based on selected parameters
2025-05-04 01:38:18,375 - INFO - Generated 5100 total variations
2025-05-04 01:38:18,376 - INFO - Testing variation 1/5100: cashflow_stability = ts_mean(cashflow_op, 227) / (debt_lt + -0.05);
stability_z = zscore(cashflow_stability);
debt_ratio = debt_lt / (assets + -0.05);
combined_score = stability_z - zscore(debt_ratio);
-rank(combined_score)
2025-05-04 01:38:18,376 - INFO - Testing alpha: cashflow_stability = ts_mean(cashflow_op, 227) / (debt_lt + -0.05);
stability_z = zscore(cashflow_stability);
debt_ratio = debt_lt / (assets + -0.05);
combined_score = stability_z - zscore(debt_ratio);
-rank(combined_score)
2025-05-04 01:38:18,754 - INFO - Simulation creation response: 201

This script is used to clean up the logs.

python clean_up_logs.py

This script is used to submit successful alphas to the WorldQuant platform. It's not recommended to use this script as of now because it submits alphas in one batch instead of once per day.

python successful_alpha_submitter.py

The machine_lib.py module provides core functionality for alpha generation and testing through the WorldQuant platform. Here are the key components:

The main class that handles interactions with WorldQuant's API and alpha generation logic:

• Authentication and session management with WorldQuant platform
• Fetching and processing data fields (matrix and vector types)
• Generating alpha expressions using operators and data fields
• Running simulations to test alpha performance
• Processing and analyzing simulation results

• Automated alpha generation using combinations of:
  • Data fields (matrix and vector types)
  • Mathematical operators (+, -, *, /, etc.)
  • Ranking and scoring functions
  • Time series operations
• Simulation capabilities:
  • Single alpha testing
  • Batch simulation support
  • Performance metrics calculation
• Result processing:
  • Filtering based on performance thresholds
  • Storing successful alphas
  • Error handling and logging

• Data field categorization (matrix vs vector)
• Expression validation
• Performance metric calculations:
  • Information Ratio (IR)
  • Returns
  • Turnover
  • Correlation analysis

The library serves as the foundation for automated alpha mining and testing, providing the necessary tools to interact with WorldQuant's platform programmatically.

python machine_miner.py --username your_worldquant_username --password your_worldquant_password

Just like the pre-consultant non ai but without workaround like single simulation and skip for inaccessible data fields and operators.

pip install -r requirements.txt

python machine_miner.py --username your_worldquant_username --password your_worldquant_password

• Local LLM Integration: Ollama with llama3.2:3b/llama2:7b models
• GPU Acceleration: NVIDIA CUDA support for faster inference
• Web Dashboard: Real-time monitoring and control interface
• Automated Orchestration: Continuous alpha generation, mining, and submission
• Docker Support: Easy deployment with containerization
• Performance: 3-5x faster than previous approaches

• v1.0: Basic Kimi API integration with manual workflows
• v1.1: Alpha expression mining and optimization
• v1.2: Automated submission with rate limiting
• v2.0: Complete rewrite with Ollama, GPU support, and web dashboard

• Integrate more templates
• Integrate more datafields
• Integrate more operators
• Integrate more regions
• Integrate more universes
• Integrate more alphas

• An interim solution to manage WorldQuant Alpha Generator with python GUI

• An interim solution to manage agent networks with python GUI

• Key Points
  • A free(as of now because it is not done jajaja) user-friendly interface to create agent networks to work with the WorldQuant Alpha Generator
  • Open source and frontend only database interactions so you can see that the website does not save your WorldQuant credentials but only your email will be used to identify you
    • You would need to first verify with WorldQuant via API then verify with the site
    • No WorldQuant credentials are saved on the server side but your email will be used to identify you
  • Login required for managing agent networks
  • Free tier available
  • Leverage vector databases to store agent memories
• Features
  • Chat with agents
  • Create agent networks
  • Manage agent networks
  • Delete agent networks
  • View agent networks
  • View agent memories
  • Alpha Polisher - Polish existing alphas or generate new ideas using AI

• Key Points
  • Implementation of Agent-to-Agent (A2A) protocol for automated financial research
  • Prescriptive agent architecture mimicking real-world financial analyst workflows
  • Integration with existing WorldQuant Alpha Generator components
  • Automated alpha mining pipeline with specialized agents

• Data Collection Agents

  • Web Crawler Agent: Automated financial news and market data collection
  • PDF Extractor Agent: Research paper processing and information extraction
  • Data Processor Agent: Data transformation and knowledge base management

• Alpha Research Agents

  • Alpha Idea Generator Agent: Pattern recognition and idea generation
  • Alpha Idea Simulator Agent: Validation and testing of alpha ideas
  • Alpha Config Setter Agent: Parameter optimization and configuration

• Execution Agents

  • Multi-Arm Bandit Agent: Reinforcement learning for action selection
  • Executor Agent: Implementation and monitoring of selected actions

• Automated research workflow orchestration
• Structured data exchange between agents
• Performance feedback loops
• Knowledge base integration
• Real-time progress tracking
• Priority-based task scheduling

• Vector database integration for agent memories
• API integration with WorldQuant platform
• Web interface for monitoring and control
• Automated alpha submission pipeline

• Advanced natural language processing for research paper analysis
• Machine learning models for pattern recognition
• Automated hypothesis generation and testing
• Real-time market data integration
• Performance optimization and scaling
• Enhanced error handling and recovery mechanisms

We welcome contributions from the community! Here's how you can help:

1. Fork the repository
2. Create a new branch (git checkout -b feature/improvement)
3. Make your changes
4. Run tests to ensure nothing is broken
5. Commit your changes (git commit -am 'Add new feature')
6. Push to the branch (git push origin feature/improvement)
7. Create a Pull Request

• Use the GitHub issue tracker to report bugs
• Clearly describe the issue including steps to reproduce
• Make feature requests through GitHub issues
• Tag issues appropriately

• Help improve documentation
• Add code comments where needed
• Update the README with new features
• Write tutorials and examples

• Follow existing code style and conventions
• Write clear commit messages
• Add tests for new features
• Update documentation for changes
• Be respectful to other contributors

• Join our community chat
• Ask questions in GitHub issues
• Read existing documentation
• Check closed issues for solutions

We appreciate all contributions that help make this project better!

This project integrates components from Dify (agent-dify-api and agent-dify-web) for enhanced alpha mining capabilities. These components are used under the Apache License 2.0.

The Dify components (agent-dify-api and agent-dify-web) are licensed under the Apache License 2.0. This means:

1. You may use, reproduce, and distribute the Dify components
2. You may modify and create derivative works
3. You must include the original copyright notice
4. You must state significant changes made to the original software
5. You must include a copy of the Apache License 2.0

For the complete terms and conditions, please refer to the Apache License 2.0.

The Dify components are integrated into this project to enhance alpha mining capabilities:

1. agent-dify-api: Provides API endpoints for alpha generation and mining
2. agent-dify-web: Offers a web interface for alpha mining operations

The Dify components are used to:

• Generate and validate alpha expressions
• Process and analyze market data
• Provide a user-friendly interface for alpha mining
• Enable automated alpha generation and submission

This project uses components from Dify, which are licensed under the Apache License 2.0. The original copyright notices and license information are preserved in the respective component directories.

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

1. Docker and Docker Compose installed
2. Python 3.8 or higher
3. Node.js 16 or higher (for web interface)

1. Start the Dify Services

# Start the Dify API and Web services
docker-compose -f docker-compose.middleware.yaml up -d

2. Verify Services

# Check if services are running
docker ps

1. Access the Web Interface

   • Open your browser and navigate to http://localhost:3000
   • Log in with your credentials

2. Creating Alpha Mining Tasks

   • Click on "New Task" in the web interface
   • Select "Alpha Mining" as the task type
   • Configure your mining parameters:
     • Data fields to use
     • Time period
     • Universe selection
     • Mining strategy

3. Monitoring Mining Progress

   • View real-time mining progress in the dashboard
   • Check generated alphas in the "Results" section
   • Export successful alphas for submission

1. API Authentication

import requests

API_URL = "http://localhost:8000"
headers = {
    "Authorization": "Bearer your_api_key"
}

2. Creating Mining Tasks

# Create a new mining task
response = requests.post(
    f"{API_URL}/api/v1/mining/tasks",
    headers=headers,
    json={
        "name": "My Mining Task",
        "data_fields": ["close", "volume", "high", "low"],
        "time_period": "1Y",
        "universe": "US",
        "strategy": "correlation"
    }
)

3. Checking Task Status

# Get task status
task_id = response.json()["task_id"]
status = requests.get(
    f"{API_URL}/api/v1/mining/tasks/{task_id}",
    headers=headers
)

4. Retrieving Results

# Get mining results
results = requests.get(
    f"{API_URL}/api/v1/mining/tasks/{task_id}/results",
    headers=headers
)

1. Resource Management

   • Monitor system resources during mining
   • Adjust mining parameters based on available resources
   • Use appropriate timeouts for long-running tasks

2. Error Handling

   • Implement proper error handling in your API calls
   • Check task status regularly
   • Save intermediate results

3. Performance Optimization

   • Use appropriate batch sizes
   • Implement caching where possible
   • Monitor and adjust mining parameters

1. Service Issues

   • Check Docker container logs: docker logs <container_id>
   • Verify service health: docker-compose ps
   • Restart services if needed: docker-compose restart

2. API Issues

   • Verify API endpoint availability
   • Check authentication tokens
   • Monitor API rate limits

3. Mining Issues

   • Verify data field availability
   • Check universe configuration
   • Monitor memory usage

For more detailed information about specific features and configurations, refer to the Dify Documentation.