Halve entropy

ml-agents/mlagents/trainers/tests/torch/test_distributions.py

         assert log_prob == pytest.approx(-0.919, abs=0.01)
 
     for ent in dist_instance.entropy().flatten():
-        # entropy of standard normal at 0
-        assert ent == pytest.approx(2.83, abs=0.01)
+        # entropy of standard normal at 0, based on 1/2 + ln(sqrt(2pi)sigma)
+        assert ent == pytest.approx(1.42, abs=0.01)
 
 
 def test_tanh_gaussian_dist_instance():

ml-agents/mlagents/trainers/torch/distributions.py

         return torch.exp(log_prob)
 
     def entropy(self):
-        return torch.log(2 * math.pi * math.e * self.std + EPSILON)
+        return 0.5 * torch.log(2 * math.pi * math.e * self.std + EPSILON)
 
 
 class TanhGaussianDistInstance(GaussianDistInstance):