search

kuz / DeepMind-Atari-Deep-Q-Learner

The original code from the DeepMind article + my tweaks
http://www.nature.com/nature/journal/v518/n7540/full/nature14236.html
1.83k stars 532 forks source link

How to resume training from snapshot #18

Open fairylrt opened 8 years ago

fairylrt commented 8 years ago

I have modified the run_gpu code and add argument network, and I'm sure the network is loaded from file (I add a print at loading place), but the result is just like from scratch. However, using the test_gpu, I'm sure the snapshot does have a trained network.

Does anyone have some suggestion?

J6767 commented 8 years ago

Please post how you have edited run_gpu, that may be the source of the error.

jtgebert commented 8 years ago

Has anyone figured out how to resume a network yet. I tried adding a parameter to run_gpu for a pretrained network, but when I put "-network $2" (it is the second parameter) into the args array it still runs the game from scratch

iostreamdoth commented 8 years ago

I would suggest saving the step count as soon as file is saved and start from that step count.