UG-CPPO 论文 v3（FinAI 竞赛 2025，NeurIPS 2026 投稿）的训练模型。
作者 — Grace-Esther Dong · Aivancity Paris-Cachan 论文 —
UG_CPPO_preprint_PAT_corrected.pdf代码 — https://github.com/graceesthi/ug_cppo 预印本 — arXiv [待公布]
30 个训练好的 Stable-Baselines3 智能体（10 个随机种子 × 3 种算法）：
{mode}_seed{seed}.zip
ppo_seed42.zip → Vanilla PPO, seed 42
cppo_seed42.zip → CVaR-PPO, seed 42
ug_cppo_seed42.zip → UG-CPPO (ours), seed 42
... (3 modes × 10 seeds = 30 files total)累积收益（均值±标准差）
| 模型 | 均值 | 标准差 | Rachev | MDD | Wilcoxon p（与 PPO 对比） |
|---|---|---|---|---|---|
| PPO | 43.94% | ±32.18% | 0.9445 | −27.95% | — |
| CPPO | 39.71% | ±46.01% | 0.9408 | −31.08% | 0.1720 |
| UG-CPPO | 35.99% | ±38.70% | 0.9420 | −29.72% | 0.8127 |
解读：
表现最佳者（按 Rachev 指标）：
from stable_baselines3 import PPO
from huggingface_hub import hf_hub_download
# Download UG-CPPO seed 47 (top performer)
path = hf_hub_download(
repo_id="graceesthi/ug-cppo-finai-2025",
filename="ug_cppo_seed47.zip"
)
agent = PPO.load(path)ppo_seed*.zip、cppo_seed*.zip、ug_cppo_seed*.zip——训练好的智能体multiseed_report_v13.json——包含Wilcoxon检验的完整结果UG_CPPO_paper.pdf——包含方法论的完整论文multiseed_performance.png——性能对比图@inproceedings{dong2026ugcppo,
title={UG-CPPO: Uncertainty-Gated LLM Infusion for Risk-Sensitive
Reinforcement Learning Trading Agents},
author={Dong, Grace-Esther},
booktitle={NeurIPS 2026 — FinAI Contest 2025, Task 1},
year={2026},
note={v3: multi-seed honest evaluation with PAT corrections}
}