DeepTraderX (DTX) — Documentation

Reference documentation for the DeepTraderX project: a deep-learning trading agent that competes inside the Threaded Bristol Stock Exchange (TBSE), a multi-threaded simulated limit-order-book (LOB) market.

This folder is the source for the project’s documentation site (published to GitHub Pages by .github/workflows/pages.yml). Anything committed here is public — do not put credentials, internal notes, or security findings in this directory.

What is this project?

DeepTraderX (DTX, internal trader code DTR) is an LSTM-based automated trader trained purely on Level-2 LOB market data. It is benchmarked against well-known public-domain trading strategies (ZIC, ZIP, GDX, AA, Giveaway, Shaver) inside TBSE — a Python multi-threaded fork of Dave Cliff’s Bristol Stock Exchange (BSE).

The project is the codebase behind the MEng dissertation and the ICAART 2024 paper “DeepTraderX: Challenging Conventional Trading Strategies with Deep Learning in Multi-Threaded Market Simulations.” Both PDFs are in the repo root.

Documentation index

Document	Contents
architecture.md	System components, threading model, end-to-end data flow
trader-agents.md	Every trading algorithm (ZIC/ZIP/GDX/AA/GVWY/SHVR/SNPR/DTR)
data-and-ml.md	The 13 LOB features, the LSTM model, the training pipeline
configuration-and-running.md	`config.py`, the three run modes, output formats
deployment.md	Docker image, Kubernetes job, AWS/EC2 provisioning

Repository layout (high level)

DeepTraderX/
├── deep_trader_tbse/            # Main application
│   ├── tbse.py                  # Entry point: orchestrates market sessions
│   ├── markets.csv              # Default batch of trader schedules
│   ├── Dockerfile               # Container image for cloud runs
│   └── src/
│       ├── config.py            # Simulation configuration + validation
│       ├── tbse/                # The exchange simulation
│       │   ├── tbse_exchange.py        # Order book + matching engine
│       │   ├── tbse_trader_agents.py   # All trading algorithms
│       │   ├── tbse_customer_orders.py # Supply/demand order generation
│       │   ├── tbse_msg_classes.py     # Order dataclass
│       │   ├── tbse_sys_consts.py      # System price bounds
│       │   └── RWD/                    # Real-world price series (IBM, GBP/USD, bonds)
│       └── deep_trader/         # The neural network
│           ├── neural_network.py       # Model load / inference / test
│           ├── lstm_architecture.py    # Model definition + training (offline)
│           ├── data_generator.py       # Keras Sequence for batched training
│           ├── utils.py                # Data pickling + normalization
│           └── Models/                 # Trained Keras models (.h5/.json/.csv)
├── configure_ec2.py             # AWS EC2 / kops cluster bootstrap (one-off)
├── generate_schedules.py        # Produces markets.csv permutations
├── market-simulations.yaml      # Kubernetes Job spec
├── requirements.txt
└── docs/                        # ← you are here

Quick start

$ python3 -m venv .venv && source .venv/bin/activate
$ pip install -r requirements.txt
$ cd deep_trader_tbse
$ python3 tbse.py            # run with the schedule in src/config.py

Important: TBSE must be run from inside the deep_trader_tbse/ directory. Module imports (import src.config) and the model-load path (./src/deep_trader/Models/...) are both resolved relative to the current working directory.

See configuration-and-running.md for the other two ways to specify a market (command-line and CSV) and the output formats.