Questions: Is it possible to run the reinforcement learning experiment without printing out the record?

[Ttopiac]

I am training the robotic model to solve a task with TRPO by using the following two libraries.

From garage.trainer import Trainer
from garage.torch.algos import TRPO

During the experiment, programs will print the logs. However, I want it not to do so in order to save more experiment time. Is it possible to only run the experiment without printing out the experiment logs?

I am talking about the logs like the following. Thank you for reading my questions.

Evaluation/AverageDiscountedReturn -3499.43
Evaluation/AverageReturn -19772.9
Evaluation/Iteration 901
Evaluation/MaxReturn -19419.8
Evaluation/MinReturn -20069.7
Evaluation/NumEpisodes 4
Evaluation/StdReturn 241.69
Evaluation/TerminationRate 1
GaussianMLPPolicy/Entropy 34.2077
GaussianMLPPolicy/KL 0.00994284
GaussianMLPPolicy/KLBefore 0
GaussianMLPPolicy/LossAfter 22.0052
GaussianMLPPolicy/LossBefore 35.2744
GaussianMLPPolicy/dLoss 13.2692
GaussianMLPValueFunction/LossAfter 8.19468
GaussianMLPValueFunction/LossBefore 8.19468
GaussianMLPValueFunction/dLoss 9.53674e-07
TotalEnvSteps 3.608e+06

[krzentner]

It's very unlikely that logging has any measurable affect on performance, but if you don't want to save logs, you can set trainer.enable_logging to false

[krzentner]

It's also worth considering that whatever application you have in mind, you most likely need to check the log anyways to make sure that the training didn't produce a broken policy. The only case where you don't is if you're training one of the easy gym or atari environments with known hyperparameters. Unfortunately deep reinforcement learning as a field is not yet at a point where it reliably produces working policies to new environments without human supervision.