aditijc/snooker-testbed-canary-7020518-warmstart-v1

该数据集记录了在斯诺克测试平台上进行的强化学习实验的结果，具体是在PPO算法下对已有模型进行500k额外步骤的warmstart训练后的性能指标。数据集包含11行数据，每行记录了训练过程中的关键指标，如全局时间步、训练课程阶段、平均得分、最高得分、平均击球次数、平均效率、平均犯规率等。实验结果显示，此次训练未能成功，犯规率上升且策略标准差增加，表明新的奖励信号不足。

This dataset records the results of a reinforcement learning experiment conducted on a snooker testbed, specifically the performance metrics after 500k additional PPO steps of warmstart training on an existing model. The dataset contains 11 rows, each recording key metrics during the training process, such as global timestep, curriculum stage, mean score, max score, mean shots, mean efficiency, mean foul rate, etc. The experimental results show that this training was unsuccessful, with an increased foul rate and an upward drift in policy std, indicating insufficient gradient signal from the new reward.