google / dopamine

Dopamine is a research framework for fast prototyping of reinforcement learning algorithms.
https://github.com/google/dopamine
Apache License 2.0
10.45k stars 1.37k forks source link

Is the baseline data smoothed? #91

Open snailrowen1337 opened 5 years ago

snailrowen1337 commented 5 years ago

When running the dqn algorithm, I consistently get higher variance than what the plots in https://google.github.io/dopamine/baselines/plots.html would suggest. Is the data there smoothed by a five-point-window or similar?

theovincent commented 1 year ago

@snailrowen1337 the data is not smoothed. Bootstrapped confidence intervals are used. This is why the variance that is shown is "so" small :)