Fixed bug with logging of std for full-cov
This commit is contained in:
		
							parent
							
								
									cb9ee4f302
								
							
						
					
					
						commit
						4854346f2d
					
				@ -366,7 +366,7 @@ class PPO(GaussianRolloutCollectorAuxclass, OnPolicyAlgorithm):
 | 
			
		||||
                    "train/std", th.mean(self.policy.chol).mean().item())
 | 
			
		||||
            else:
 | 
			
		||||
                self.logger.record(
 | 
			
		||||
                    "train/std", th.mean(th.diagonal(self.policy.chol, dim1=-2, dim2=-1)).mean().item())
 | 
			
		||||
                    "train/std", th.mean(th.sqrt(th.diagonal(self.policy.chol.T @ self.policy.chol, dim1=-2, dim2=-1))).mean().item())
 | 
			
		||||
 | 
			
		||||
        self.logger.record("train/n_updates",
 | 
			
		||||
                           self._n_updates, exclude="tensorboard")
 | 
			
		||||
 | 
			
		||||
		Loading…
	
		Reference in New Issue
	
	Block a user