YAWNING-TITAN (YT) modified environment

该数据集是一个多代理环境，其中攻击代理和防御代理同时进行训练，融入了噪声，并且具有不同的观察/动作空间。环境支持乐观（零知识）和悲观（全知识）的训练设置。每个代理均接受3000个回合的训练和500个回合的评估。该数据集的规模为50节点网络，具有60%的连通性。任务是在网络安全背景下对强化学习代理进行训练和评估。

This dataset is a multi-agent environment where attack and defense agents are trained simultaneously, with injected noise and distinct observation/action spaces. The environment supports two training modes: optimistic (zero-knowledge) and pessimistic (full-knowledge). Each agent undergoes 3000 episodes of training and 500 episodes of evaluation. This dataset is built on a 50-node network with 60% connectivity. The core task is to train and evaluate reinforcement learning agents within the cybersecurity context.