How ntorm1/atlas works
Project Type and Purpose
• Atlas is a high-performance, modular algorithmic trading backtesting library written in C++20.
• It exposes a Python API (AtlasPy, built with pybind11) and a Qt-based GUI (AtlasX).
• It is intended for large-scale backtests: ingesting CSV/HDF5 market data, composing strategy ASTs, running simulations, and collecting performance metrics.
High-Level Architectural Components
A. Core C++ Runtime (“Atlas”)
• Built as a Visual Studio DLL (Atlas.sln, Atlas.vcxproj).
• Contains the bulk of the logic, organized into “modules/”:
• ast (AST node definitions)
• exchange (data ingestion & exchange view)
• hydra (runtime orchestration)
• model (ML models: LinearRegression, XGBoost, Torch)
• standard (utilities, serialization, stats, time)
• strategy (allocation, meta-strategy, tracing, measurement)
• Public headers in “include/” expose core types and macros.
• Links to external libraries: Eigen, H5CPP, xgboost, unordered_dense, expected.hpp.
B. Python Binding Layer (“AtlasPy”)
• Built as a pybind11 wrapper (AtlasPy/AtlasPy.vcxproj).
• Exposes C++ functionality through Python modules: core (Hydra, Portfolio, Strategy), ast, parser, strategy, logging.
• Test suite under AtlasPy/tests calls into the Python API to drive the backtester.
C. GUI Application (“AtlasX”)
• Qt5/6-based desktop application (AtlasX/).
• Uses QScintilla, Advanced Docking System.
• Implements a graphical strategy editor, tree view of AST, and real-time tracing via the C++ core.
D. Test Harnesses
• AtlasTest: a C++ unit-test project verifying core internals (AST, exchange, modules).
• AtlasPy/tests: Python-level integration tests driving the backtester end-to-end with sample data.
E. Data Sources & Formats
• .csv and .h5 files ingested by ExchangeIO under modules/exchange.
• Contains sample datasets under AtlasPy/tests/files.