SamplingConfig: Change default of repeat_per_collect to 1 (safest option)

This commit is contained in:
Dominik Jain 2023-12-14 18:53:25 +01:00
parent 408d51f9de
commit 45a1a3f259

View File

@ -67,7 +67,7 @@ class SamplingConfig(ToStringMixin):
collected during training. collected during training.
""" """
repeat_per_collect: int | None = 10 repeat_per_collect: int | None = 1
""" """
controls, within one gradient update step of an on-policy algorithm, the number of times an controls, within one gradient update step of an on-policy algorithm, the number of times an
actual gradient update is applied using the full collected dataset, i.e. if the parameter is actual gradient update is applied using the full collected dataset, i.e. if the parameter is