search

rail-berkeley / rlkit

Collection of reinforcement learning algorithms
MIT License
2.45k stars 550 forks source link

How could I see the video in rig... #19

Closed cww97 closed 5 years ago

cww97 commented 5 years ago

I see

            save_video=True,
            save_video_period=100,

I can only see the png files in the folder data, I wanna know how or where can I find the video files.

vitchyr commented 5 years ago

It should be in the same folder inside of data as the pngs.

Note that you need to wait until the RL part of the algorithm is running. If you see output about "train/epoch", then it is still pretraining the VAE. Once you see the log outputting information about the "Policy Loss", then that means RL is running.

Lastly, save_video_period=100 means that videos will only be logged once every 100 epochs. If you want to save videos more frequently, reduce this number.

cww97 commented 5 years ago

It should be in the same folder inside of data as the pngs.

Note that you need to wait until the RL part of the algorithm is running. If you see output about "train/epoch", then it is still pretraining the VAE. Once you see the log outputting information about the "Policy Loss", then that means RL is running.

Lastly, save_video_period=100 means that videos will only be logged once every 100 epochs. If you want to save videos more frequently, reduce this number.

Thank you, sir. I can see the video for 100, 200, 300 now(due to my poor gpu, I need more time)

In reality I have two Dobot Robotic Arms, and some cameras. I wonder where can I change the Robotic Arms and cameras in the sample experiment. And how could I test in real World.

Sorry it seem that my question is a little stupid, I did not learn a lot about Reinforcement Learning and Pytorch(even Machine Learning or Deep Learning). In another word, a rookie. I read the paper 'RIG', it seems that I can almost get the idea of what it means. Could you give me some advice of understand the algorithm, or DRL better ?