chainer / chainerrl

ChainerRL is a deep reinforcement learning library built on top of Chainer.
MIT License
1.17k stars 224 forks source link

AdditiveGaussian Overflow #603

Closed mcmips closed 3 years ago

mcmips commented 4 years ago

I am trying to use AdditiveGaussian with DDQN and I get the following error during episode.. After this error occurs and even if I ignore the error, no further episode or iteration results in any reward and every iteration results in invalid action values... Warning (from warnings module): File "C:\Python37\lib\site-packages\chainer\functions\loss\huber_loss.py", line 37 square_part = utils.force_array(xp.square(linear_part), dtype) RuntimeWarning: overflow encountered in square episode iteration = 0

Any clue please.. thanks for looking into it..