In on_policy_learning.py, current_learning_iteration is only being incremented at the end of the learning run, so all intermediate saves have the incorrect iteration. This is fixed, so if you resume with an intermediate checkpoint, it starts counting from the correct place (mainly just a nice-to-have for the tensorboard visualizations).
In on_policy_learning.py,
current_learning_iteration
is only being incremented at the end of the learning run, so all intermediate saves have the incorrect iteration. This is fixed, so if you resume with an intermediate checkpoint, it starts counting from the correct place (mainly just a nice-to-have for the tensorboard visualizations).