Question about gaussian distribution internals.

[bolshoytoster]

I'm trying to port ppo2 to rust, and I've managed to mostly do this. I have, however come across something I don't understand.

In baselines/common/distributions.py, when creating the DiagGaussianDistribution, the tensor mean is multiplied by 0, then added to logstd:

baselines/baselines/common/distributions.py

Line 105 in ea25b9e

pdparam = tf.concat([mean, mean * 0.0 + logstd], axis=1)

Can anyone explain why this isn't just this?

pdparam = tf.concat([mean, logstd], axis=1)