stocker-env / openenv.yaml
Hydr473's picture
Add scripts for rendering charts and serving vLLM; enhance task validation and introduce council tests
70a0c37
Raw
History Blame
1.05 kB
name: stocker
version: 0.1.0
description: >
A stock-trading RL environment built on OpenEnv. Long-term backtest over
3 historical episodes (AAPL, INTC, META). Per-step decisions are produced
by a council of 7 LLM specialist agents (chart, seasonal/long-term, technical
indicators, news, forum sentiment, peers + commodity, geopolitics/macro)
whose votes are merged by a moderator LLM. The moderator is GRPO-trained
via TRL on top of google/gemma-4-E4B-it.
tasks:
- task_easy
- task_medium
- task_hard
observation_space:
ticker: str
date: str
price: float
price_history: "list[float]"
fundamentals: dict
cash: float
position: int
portfolio_value: float
task_id: str
step_number: int
total_steps: int
chart_path: str
headlines: "list[dict]"
forum_excerpts: "list[dict]"
indicators: dict
peers: dict
macro: "list[dict]"
action_space:
side: "literal[buy, sell, hold]"
quantity: int
entrypoint: app.core.environment:StockerEnv
tags:
- openenv
- finance
- trading
- multi-agent
- rl