ERWR positive_adv set to True?

rlworkgroup / garage

A toolkit for reproducible reinforcement learning research.

MIT License

1.88k stars 310 forks source link

ERWR positive_adv set to True? #673

Closed zequnyu closed 5 years ago

zequnyu commented 5 years ago

I'm trying to remove kwargs out of all the algorithms. In ERWR the default positive_adv is set to True. When it passing to VPG, then NPO, the self.positive_adv is never set, so line 279 in npo.py is never reached. Not sure if this is the intention. The test won't pass if self.positive_adv in NPO is set. Would you please check it? @ahtsan

CatherineSue commented 5 years ago

I think it is set in tf.algos.BatchPolopt here?

zequnyu commented 5 years ago

Thank you @CatherineSue .