Multidimensional GuassianTreeMLP does not work. about golearn HOT 1 CLOSED

This has been fixed! The following issues were present:

1. In the multi-dimensional Gaussian implementation, the standard deviation instead of variance was used to calculate the log probability density (commit df92180).
2. The Online Experiment struct did not properly transfer actions between Agents and Environments. Instead of using a copy of the action, the pointer to the action selected by the Agent was sent to the Environment through the Step() function of the Environment. This cause the pointed to action to be clipped, causing the Agent to update with a clipped action instead of its originally selected action (commit 2a1d136).

from golearn.