Reproduce results for LSTM-DQN

[Isminoula]

Hello,

It seems that running the run_cpu for the Home World, without changing the parameters does not produce the same result as in the paper. Most likely I am doing something wrong... Could you please help me out reproduce your results?

Thanks!

run_cpu.txt
plot_questcompletion.pdf
plot_reward.pdf

[karthikncode]

Hi,
Could you check if you setup the environment correctly following this? (i.e. could you check the game build, rewards, etc. by manually playing?)
I can't see anything wrong with the run_cpu file you have attached.

[Isminoula]

Hello,

Thank you for the reply.
The environment seems to be working, I played manually and the rewards are as expected (eating apple before going to bed).
Is the state dimension supposed to be 100? (state_dim=100)

Thank you again for the help.

[karthikncode]

We want the state_dim to be >= the largest input state description in number of words. So, you could try a lower state_dim (say 30) for the Home world.