d3f12d1622f736e649fcec853044b05fe68e05ba,catalyst/rl/onpolicy/algorithms/ppo.py,PPO,postprocess_buffer,#PPO#Any#Any#,230

Before Change

    def postprocess_buffer(self, buffers, len):
        adv_centered = \
            buffers["advantage"][:len] \
            - buffers["advantage"][:len].mean(axis=0)
        adv_std = buffers["advantage"][:len].std(axis=0)
        buffers["advantage"][:len] = adv_centered / (adv_std + 1e-6)

    def train(self, batch, **kwargs):
        (states_t, actions_t, returns_t, states_tp1, done_t,
         values_t, advantages_t, action_logprobs_t) = (

After Change

    def postprocess_buffer(self, buffers, len):
        adv = buffers["advantage"][:len]
        adv = (adv - adv.mean(axis=0)) / (adv.std(axis=0) + 1e-8)
        buffers["advantage"][:len] = adv

    def train(self, batch, **kwargs):
        (states_t, actions_t, returns_t, states_tp1, done_t,

Instances