search

laermannjan / nip-deeprl-project

Student project in deep reinforcement learning with the OpenAI Gym. We evaluated and analyzed how different model architectures performed as agents in various games.
0 stars 0 forks source link

misc_util.RunningAvg #2

Closed laermannjan closed 7 years ago

laermannjan commented 7 years ago

Use RunningAvg util function instead of or in addition to our sliding window mean calculations(convolution style).

laermannjan commented 7 years ago

Seems less scientific as it depends on parameter gamma which is alters smoothening with growing timesteps. On the other side it might give a more accurate smoothing when functions start oscillating.

laermannjan commented 7 years ago

We could use this with a filter ignoring present changes but which is highly sensible to those from a few steps ago and slowly fades off in sensibility.

For now (7b7ce636f6a7d360954c73a85bba0f5431e5f61d) we just use it for an ETA of our experiment.