██████╗██╗ ██╗██████╗ ███████╗██████╗ ████████╗██████╗ █████╗ ██████╗ ███████╗██████╗
██╔════╝╚██╗ ██╔╝██╔══██╗██╔════╝██╔══██╗╚══██╔══╝██╔══██╗██╔══██╗██╔══██╗██╔════╝██╔══██╗
██║ ╚████╔╝ ██████╔╝█████╗ ██████╔╝ ██║ ██████╔╝███████║██║ ██║█████╗ ██████╔╝
██║ ╚██╔╝ ██╔══██╗██╔══╝ ██╔══██╗ ██║ ██╔══██╗██╔══██║██║ ██║██╔══╝ ██╔══██╗
╚██████╗ ██║ ██████╔╝███████╗██║ ██║ ██║ ██║ ██║██║ ██║██████╔╝███████╗██║ ██║
╚═════╝ ╚═╝ ╚═════╝ ╚══════╝╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═╝╚═════╝ ╚══════╝╚═╝ ╚═╝
[ NIGHT CITY MARKET EXPLOITATION SYSTEM ]
"In 2077, what makes someone a successful trader? Getting rich." — V, probably
Choom, welcome to the most preem stock trading ICE-breaker this side of Night City. This neural network runs hotter than a Militech shard, trained using Deep Reinforcement Learning (Deep Q-Learning) to hack the corpo markets and extract maximum eddies.
Implementation is delta-grade — clean, minimal, and optimized for those chooms who want to understand the tech behind the chrome.
In the dark future of automated trading, Reinforcement Learning is the closest thing to true machine consciousness. These algorithms learn like street samurai — through trial, error, and a whole lot of flatlined trades.
The beauty? This technique adapts to any market situation that can be described as a Markovian process — which in corpo-speak means: "The future depends only on the present, not the past."
💡 NetRunner Tip: Traditional supervised learning is like following a corpo playbook. RL is like being a solo — you learn what works through experience on the streets.
This daemon utilizes Model-free Reinforcement Learning via Deep Q-Learning — think of it as installing a Sandevistan for your trading decisions.
The Loop:
[JACK IN] → Observe market state → Execute action (BUY/SELL/HOLD) →
Receive reward signal → Update neural weights → [REPEAT]
- 🧠 Vanilla DQN — Base neural implant
- 🎯 DQN with Fixed Target Distribution — Stabilized targeting system
- 🔄 Double DQN — Dual-core processing for better value estimation
- ⚡ Batch Prediction — Overclocked training speed
- 📊 Prioritized Experience Replay — Memory optimization (coming soon)
- 🏗️ Dueling Network Architectures — Advanced combat protocols (coming soon)
╔══════════════════════════════════════════════════╗
║ MINIMUM REQUIREMENTS FOR NEURAL LINK ║
╠══════════════════════════════════════════════════╣
║ ► Python 3.9+ (Neural Interface) ║
║ ► TensorFlow 2.16+ (Cortex Processor) ║
║ ► Keras 3.x (Synaptic Framework) ║
╚══════════════════════════════════════════════════╝
Target: GOOG corpo stock (2010-17 training data)
Mission Status: ✅ COMPLETE
Profit Extracted: $1,141.45 (2019 test) | $863.41 (2018 validation)
"That's a lot of eddies, choom."
Check out the DataKrash Visualization Notebook for detailed analytics of your runs.
| Issue | Description |
|---|---|
| 🎯 Single-Stock Mode | Agent trades one share at a time — keeps the neural load manageable, choom |
| 📈 Normalized Vectors | N-day window uses sigmoid normalization [0,1] — standard Arasaka protocols |
| 🖥️ CPU Training | Sequential nature means CPU outperforms GPU — no Kiroshi optics needed here |
Download market data from Yahoo! Finance or use the included datasets in data/ directory — pre-extracted from Arasaka servers.
# Initialize cyberware dependencies
pip3 install -r requirements.txt# Jack into the training matrix
python3 train.py data/GOOG.csv data/GOOG_2018.csv --strategy t-dqn# Unleash the trading ICE-breaker
python3 eval.py data/GOOG_2019.csv --model-name model_debug_10.keras --debug╔════════════════════════════════════════════════════════════╗
║ > SYSTEM INITIALIZED ║
║ > NEURAL LINK: ESTABLISHED ║
║ > MARKET CONNECTION: ONLINE ║
║ > STATUS: READY TO EXTRACT EDDIES ║
╚════════════════════════════════════════════════════════════╝
Models are saved in Keras 3 .keras format — the new corpo standard. Legacy TensorFlow 1.x shards are incompatible. If you've got old chrome, you'll need to retrain from scratch.
Props to these legendary NetRunners:
- @keon — Original deep-q-learning architect
- @edwardhdlu — q-trader pioneer
Required Reading for Aspiring NetRunners:
- 📖 Playing Atari with Deep Reinforcement Learning — The OG shard
- 📖 Human Level Control Through Deep Reinforcement Learning — DeepMind's magnum opus
- 📖 Deep RL with Double Q-Learning — Dual-core optimization
- 📖 Prioritized Experience Replay — Memory enhancement protocols
- 📖 Dueling Network Architectures — Advanced combat systems
╔═══════════════════════════════════════════════════════════╗
║ ║
║ "The street finds its own uses for things." ║
║ — William Gibson ║
║ ║
║ Wake up, Samurai. We have markets to burn. 🔥 ║
║ ║
╚═══════════════════════════════════════════════════════════╝
Made with 💜 in Night City | 2077