rl_code It's a reinforce learning for gym, now is playing LunarLander-v2 game result the policy grdient result is like: the ppo result is like: 一些分析 对于a2c来说,learning rate不能过大,在小learning rate情况下表现稳定,如下图: 对于ac来说,参数仍需要调整,在后期会有崩溃情况出现,在使用mc算reward时,会有崩溃现象发生,如下图,但是神奇的是,崩溃后会恢复,应该是从local min跳出来了: