search

IBM / rl-testbed-for-energyplus

Reinforcement Learning Testbed for Power Consumption Optimization using EnergyPlus
MIT License
177 stars 74 forks source link

Taking long time for training #58

Open revathij opened 3 years ago

revathij commented 3 years ago

Hi All,

I am new to the EnergyPlus and trying to set up the rl-testbed-for-energyplus for my project.

If I run the below command, it is taking long time to train. time python3 -m baselines_energyplus.trpo_mpi.run_energyplus --num-timesteps 1000000000,

Can you please let me know how long will it take to complete the training and the minimum hardware specification to run the program

biemann commented 3 years ago

Hi,

You used the same number of timesteps, that are used in the paper, but it is not really necessary to have so many if you want to have an acceptable policy. In my experiments with TRPO, you can have decent results after 10^6 timesteps. The program works on my laptop, although if you want to use other algorithms, a GPU can significantly accelerate training. But is not absolutely essential, as it is still possible to train a network in around 2 days on a CPU.

revathij commented 3 years ago

hi Thanks a lot. I have tried and it is taking a shorter time. Thanks for your help.