run in my_way_home about deep-reinforcement-learning-applied-to-doom

[Chaoukia]

Hello, sorry for the late answer, we were very busy lately. You're right about my_way_home, we've tried curiosity with an A3C backbone briefly but did not succeed in making the agent learn. If you can make it, we would be very Happy to learn How you did it. Thank you for your interest in our work. Best regards, Ayman Le ven. 3 mai 2019 15:03, WangC <[email protected]> a écrit :

…

Hi, it seems that in this code the agent cannot learn to navigation in my_way_home, does it? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEEEVHABXUAGEPGKJBUMCYDPTQZ3HANCNFSM4HKTYDRQ> .

[WangChen100]

Hi Ayman.
I find a successful repository and fork it from others'.

https://github.com/WangChen100/a2c_curiosity_vizdoom_tf

Compared with our implementation, the codes in this repository include lots of tricks, such as L2 regularizer, special calculation method on information entropy and so on.

To be honest, this repository deserve a mention and a review.

[Chaoukia]

Hi Wang, thank you for the repository, I'll take a loot at it.
In fact, our aim was to see how curiosity would behave when put in an environment requiring a lot of exploration like "my way home", but we did not investigate in depth why it didn't work and it is definitely something worth looking at.