search

MatheusMRFM / A3C-LSTM-with-Tensorflow

An implementation of the A3C deep reinforcement learning method using a LSTM layer. Created with Tensorflow.
29 stars 11 forks source link

About time step #4

Closed StephanieDU closed 6 years ago

StephanieDU commented 6 years ago

Thanks for your work. After reading your code, I am confused by the use of frames_stacked, since LSTM can already remember the sequence info among frames, what is the meaning of using 4 frames as input? Looking forward to your reply.

MatheusMRFM commented 6 years ago

Hello there! Your comment makes a lot of sense. Basically, I just copied the structure from the original A3C paper, where it is said that A3C uses a similar network structure as the DQN. Maybe a misunderstood something, but I left the input with the last four frames stacked. However, if I'm not mistaken, I have also tested it with only the last frame as input (when using the LSTM) and it still managed to work fine.

StephanieDU commented 6 years ago

Thanks so much for your prompt reply! I have learned a lot from your code. Many thanks.