on_boarding.md

graph LR
    Application_Orchestrator["Application Orchestrator"]
    Configuration_Utilities["Configuration & Utilities"]
    Market_Data_Management["Market Data Management"]
    Reinforcement_Learning_Core["Reinforcement Learning Core"]
    Traditional_Trading_Strategies["Traditional Trading Strategies"]
    Training_Optimization_Engine["Training & Optimization Engine"]
    Backtesting_Simulation_Engine["Backtesting & Simulation Engine"]
    Results_Analysis_Visualization["Results Analysis & Visualization"]
    Application_Orchestrator -- "Initializes & Loads Config" --> Configuration_Utilities
    Application_Orchestrator -- "Triggers Training" --> Training_Optimization_Engine
    Application_Orchestrator -- "Initiates Backtest" --> Backtesting_Simulation_Engine
    Configuration_Utilities -- "Provides Data Settings" --> Market_Data_Management
    Configuration_Utilities -- "Configures Agent" --> Reinforcement_Learning_Core
    Configuration_Utilities -- "Configures Algorithms" --> Traditional_Trading_Strategies
    Market_Data_Management -- "Provides Market Data & Experiences" --> Reinforcement_Learning_Core
    Market_Data_Management -- "Feeds Processed Data" --> Traditional_Trading_Strategies
    Market_Data_Management -- "Provides Training Batches" --> Training_Optimization_Engine
    Reinforcement_Learning_Core -- "Stores Experiences" --> Market_Data_Management
    Reinforcement_Learning_Core -- "Provides Model for Training" --> Training_Optimization_Engine
    Reinforcement_Learning_Core -- "Provides Trading Decisions" --> Backtesting_Simulation_Engine
    Traditional_Trading_Strategies -- "Provides Strategy for Evaluation" --> Training_Optimization_Engine
    Traditional_Trading_Strategies -- "Provides Trading Decisions" --> Backtesting_Simulation_Engine
    Training_Optimization_Engine -- "Triggers Training Steps" --> Reinforcement_Learning_Core
    Training_Optimization_Engine -- "Triggers Strategy Updates" --> Traditional_Trading_Strategies
    Training_Optimization_Engine -- "Provides Trained Model/Strategy" --> Backtesting_Simulation_Engine
    Backtesting_Simulation_Engine -- "Requests Decisions" --> Reinforcement_Learning_Core
    Backtesting_Simulation_Engine -- "Requests Decisions" --> Traditional_Trading_Strategies
    Backtesting_Simulation_Engine -- "Outputs Backtest Results" --> Results_Analysis_Visualization
    click Application_Orchestrator href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGPortfolio/Application_Orchestrator.md" "Details"
    click Market_Data_Management href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGPortfolio/Market_Data_Management.md" "Details"
    click Reinforcement_Learning_Core href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGPortfolio/Reinforcement_Learning_Core.md" "Details"
    click Traditional_Trading_Strategies href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGPortfolio/Traditional_Trading_Strategies.md" "Details"
    click Training_Optimization_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGPortfolio/Training_Optimization_Engine.md" "Details"
    click Backtesting_Simulation_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGPortfolio/Backtesting_Simulation_Engine.md" "Details"

Loading

Details

The PGPortfolio system is designed as a comprehensive deep reinforcement learning platform for cryptocurrency portfolio management. The Application Orchestrator acts as the central control, initiating the system by leveraging the Configuration & Utilities component to load and preprocess all necessary settings. Market Data Management is responsible for acquiring, storing, and providing processed historical market data and experience samples to both the Reinforcement Learning Core and Traditional Trading Strategies. The Reinforcement Learning Core also contributes by storing new experiences back into Market Data Management.\n\nThe Training & Optimization Engine orchestrates the learning process for both types of agents: it receives models from the Reinforcement Learning Core and strategies from Traditional Trading Strategies, and in turn, triggers training steps and strategy updates within these components. Once trained, the optimized models and strategies are passed to the Backtesting & Simulation Engine. This engine simulates trading on historical data, requesting real-time decisions from both the Reinforcement Learning Core and Traditional Trading Strategies. Finally, the Backtesting & Simulation Engine delivers the performance outcomes to the Results Analysis & Visualization component for comprehensive analysis and graphical representation. This structured flow ensures a clear separation of concerns and efficient data progression through the system.

Application Orchestrator [Expand]

The central control unit that initiates and manages the overall execution flow, from setup to the main training/backtesting process.

Related Classes/Methods:

• main.main:45-105
• pgportfolio.autotrain.training.train_all:35-81

Configuration & Utilities

Handles the loading, validation, and preprocessing of all system configurations, ensuring consistent settings across components, and provides general utility functions.

Related Classes/Methods:

• pgportfolio.tools.configprocess.load_config:97-108
• pgportfolio.tools.configprocess.preprocess_config:16-21

Market Data Management [Expand]

Responsible for acquiring, storing, preprocessing, and providing raw and processed historical market data, including managing the experience replay buffer.

Related Classes/Methods:

• pgportfolio.marketdata.coinlist.py
• pgportfolio.marketdata.globaldatamatrix.get_global_data_matrix:43-47
• pgportfolio.marketdata.datamatrices.create_from_config:86-111
• pgportfolio.marketdata.datamatrices.next_batch:149-157
• pgportfolio.marketdata.replaybuffer.__init__:52-53
• pgportfolio.marketdata.replaybuffer.append_experience:20-22
• pgportfolio.marketdata.replaybuffer.next_experience_batch:36-48

Reinforcement Learning Core [Expand]

Implements the core deep reinforcement learning logic, including the policy network, loss functions, and decision-making process based on observed states.

Related Classes/Methods:

• pgportfolio.learn.nnagent.__init__:9-46
• pgportfolio.learn.nnagent.train:148-150
• pgportfolio.learn.nnagent.decide_by_history:204-213
• pgportfolio.learn.network.__init__:35-36
• pgportfolio.learn.network._build_network:44-151

Traditional Trading Strategies [Expand]

A collection of classical portfolio management and trading algorithms that can be used as baselines or alternative strategies.

Related Classes/Methods:

• pgportfolio.tdagent.algorithms.anticor1.decide_by_history:15-37
• pgportfolio.tdagent.algorithms.bcrp.decide_by_history:23-34
• pgportfolio.tdagent.algorithms.olmar.decide_by_history:35-66

Training & Optimization Engine [Expand]

Orchestrates the training process for both RL and traditional agents, managing training loops, model updates, and performance evaluation during training.

Related Classes/Methods:

• pgportfolio.learn.tradertrainer.train_net:167-207
• pgportfolio.learn.rollingtrainer.rolling_train:52-59

Backtesting & Simulation Engine [Expand]

Simulates the execution of trading strategies on historical market data to evaluate their performance under realistic conditions.

Related Classes/Methods:

• pgportfolio.trade.trader.start_trading:106-123
• pgportfolio.trade.trader.__trade_body:88-104

Results Analysis & Visualization

Processes the output of backtesting and training runs to generate insightful plots and summary tables for performance analysis.

Related Classes/Methods:

• pgportfolio.resultprocess.plot.plot_backtest:44-100
• pgportfolio.resultprocess.table.table_backtest

FAQ