Examples

Example 1: Data Download and Processing

This example shows how to download and process financial data using the provided shell scripts.

Download Scripts

Use the provided download scripts in examples/download.sh:

#!/bin/bash

# Download HS300 data
python scripts/download/download.py --config configs/download/hs300/hs300_fmp_price_1day.py
python scripts/download/download.py --config configs/download/hs300/hs300_fmp_price_1min.py

# Download SP500 data
python scripts/download/download.py --config configs/download/sp500/sp500_fmp_price_1day.py
python scripts/download/download.py --config configs/download/sp500/sp500_fmp_price_1min.py

# Download DJ30 data
python scripts/download/download.py --config configs/download/dj30/dj30_fmp_price_1day.py
python scripts/download/download.py --config configs/download/dj30/dj30_fmp_price_1min.py

Processing Scripts

Use the provided processing scripts in examples/process.sh:

#!/bin/bash

# Process HS300 data
python scripts/process/process.py --config configs/process/hs300.py

# Process SP500 data
python scripts/process/process.py --config configs/process/sp500.py

# Process DJ30 data
python scripts/process/process.py --config configs/process/dj30.py

Key Features

• Multiple Data Sources: FMP, Alpaca, AkShare, TuShare
• Different Time Frames: Daily and minute-level data
• Batch Processing: Process multiple datasets simultaneously
• Data Quality: Built-in validation and cleaning

Example 2: Time Series Forecasting

This example demonstrates time series forecasting using different model architectures.

Training Scripts

Use the provided time series training scripts in examples/time.sh:

#!/bin/bash

# Train Autoformer model
CUDA_VISIBLE_DEVICES=4,5 torchrun --master_port=29510 --nproc_per_node=2 scripts/time/train.py --config configs/time/dj30_autoformer.py

# Train Crossformer model
CUDA_VISIBLE_DEVICES=4,5 torchrun --master_port=29511 --nproc_per_node=2 scripts/time/train.py --config configs/time/dj30_crossformer.py

# Train DLinear model
CUDA_VISIBLE_DEVICES=4,5 torchrun --master_port=29512 --nproc_per_node=2 scripts/time/train.py --config configs/time/dj30_dlinear.py

# Train ETSformer model
CUDA_VISIBLE_DEVICES=4,5 torchrun --master_port=29513 --nproc_per_node=2 scripts/time/train.py --config configs/time/dj30_etsformer.py

Key Features

• Multiple Architectures: Autoformer, Crossformer, DLinear, ETSformer
• Distributed Training: Multi-GPU support for faster training
• Different Datasets: DJ30, SP500, HS300, SSE50
• Performance Comparison: Easy to compare different models

Example 3: RL Trading with PPO

This example demonstrates reinforcement learning trading using PPO algorithm across multiple stocks.

Training Scripts

Use the provided PPO trading scripts in examples/ppo_trading.sh:

#!/usr/bin/env bash

# Train PPO trading agents for different stocks
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/ppo/AAPL_ppo_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/ppo/AMZN_ppo_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/ppo/GOOGL_ppo_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/ppo/META_ppo_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/ppo/MSFT_ppo_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/ppo/TSLA_ppo_trading.py &

Key Features

• Multi-Stock Training: Train on multiple stocks simultaneously
• PPO Algorithm: Stable and sample-efficient RL algorithm
• Risk Management: Built-in position limits and transaction costs
• Performance Metrics: Sharpe ratio, returns, drawdown analysis

Example 4: RL Trading with SAC

This example demonstrates reinforcement learning trading using SAC (Soft Actor-Critic) algorithm.

Training Scripts

Use the provided SAC trading scripts in examples/sac_trading.sh:

#!/usr/bin/env bash

# Kill any existing SAC trading processes
# ps -ef | grep sac | grep -v grep | awk '{print $2}' | xargs kill -9

CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/sac/AAPL_sac_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/sac/AMZN_sac_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/sac/GOOGL_sac_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/sac/META_sac_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/sac/MSFT_sac_trading.py &
CUDA_VISIBLE_DEVICES=4 python scripts/rl_trading/train.py --config=configs/rl_trading/sac/TSLA_sac_trading.py &

Key Features

• SAC Algorithm: Sample-efficient off-policy RL algorithm
• Experience Replay: Better sample efficiency with replay buffer
• Entropy Regularization: Automatic exploration-exploitation balance
• Double Q-Learning: Reduces overestimation bias

Example 5: RL Portfolio with PPO

This example demonstrates reinforcement learning portfolio optimization using PPO algorithm across different asset classes.

Training Scripts

Use the provided PPO portfolio scripts in examples/ppo_portfolio.sh:

#!/usr/bin/env bash

CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/ppo/dj30_ppo_portfolio.py &
CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/ppo/sse50_ppo_portfolio.py &
CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/ppo/hs300_ppo_portfolio.py &
CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/ppo/sp500_ppo_portfolio.py &

Key Features

• Multi-Asset Portfolio: Optimize across multiple asset classes
• Risk-Adjusted Returns: Focus on Sharpe ratio and risk management
• Diversification: Built-in diversification constraints
• Rebalancing: Dynamic portfolio rebalancing strategies

Example 6: RL Portfolio with SAC

This example demonstrates reinforcement learning portfolio optimization using SAC algorithm.

Training Scripts

Use the provided SAC portfolio scripts in examples/sac_portfolio.sh:

#!/usr/bin/env bash

CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/sac/dj30_sac_portfolio.py &
CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/sac/sse50_sac_portfolio.py &
CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/sac/hs300_sac_portfolio.py &
CUDA_VISIBLE_DEVICES=5 python scripts/rl_portfolio/train.py --config=configs/rl_portfolio/sac/sp500_sac_portfolio.py &

Key Features

• Continuous Action Space: Fine-grained portfolio allocation
• Sample Efficiency: SAC's off-policy learning advantage
• Stable Training: Reduced variance in portfolio optimization
• Adaptive Strategies: Dynamic adjustment to market conditions

Getting Started with Examples

To run these examples, follow these steps:

1. Prepare Data: Download the required datasets using the download tutorials
2. Install Dependencies: Ensure all required packages are installed
3. Configure Environment: Set up your configuration files
4. Run Examples: Execute the example scripts
5. Analyze Results: Review performance metrics and visualizations

For more detailed information, check out our Tutorials and API Reference.