Comments (12)
It does not matter, just replace inner-product layer with 1*1 kernel convolutional layer!
Thank you!
from dqn-in-the-caffe.
Were you able to get pong or breakout to train up with the change to the model you made? I made a different change that doesn't work. I think I have a bit to learn in terms of how Caffe works.
from dqn-in-the-caffe.
I got this problem too.
How to fix this problem ?
from dqn-in-the-caffe.
I tried adding a Reshape layer to the dqn.prototxt. I didn't think that would affect the update because of how the target and filter arrays are being set but I'm getting enormous Q values once training starts. I don't have much experience with reinforcement learning but I haven't seen Q values above 5 or so in my other tasks. Any ideas?
from dqn-in-the-caffe.
I put together a version that works over at
https://github.com/watts4speed/fast-dqn-caffe
It has some instructions also about how to get it going. I made some speed improvements also.
from dqn-in-the-caffe.
Sorry for the problems. I haven't run my code for a while and don't know whether it works with the latest Caffe and ALE.
@watts4speed nice work!
from dqn-in-the-caffe.
Ya there's a couple things. The caffe guys changed their API for solver creation. See what I did over on my branch. Then somewhere maybe after September something changed in caffe that broke the training all together. I've never been able to figure out what it was yet. I put a commit number that I know works for caffe on the readme for the link above. It should also work for the verion here. Let me know if you figure out what the issue is with running with the head of caffe/master :-)
from dqn-in-the-caffe.
I worked through the code also and found what looks to be an error when resetting the filters and targets for each forward pass. The filters and targets are not the correct values after calling reset and doing one step. Is there a reason for this? Am I understanding this correctly?
After 50 iterations the computed targets for each of the 32 outputs (this looks reasonable):
0.173218 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0
0 0 0 0 0.175913 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.176915 0 0 0 0 0 0
0 0 0 0 0.176081 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0
0 0 0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0
0.173392 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0.176585 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0
0 0 0 0.173022 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.174964 0 0 0 0 0
0 0 0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0
0.174004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.174004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.173392 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0.17581 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.174467 0 0 0 0 0
0 0 0 0.174032 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0
0 0 0 0 0.173349 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0
0 0 0 0 0.175626 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0.177733 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.168742 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.175043 0 0 0 0 0
0 0.171759 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0
The corresponding filters (this also looks correct as the positions match):
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
After doing one step, though, the data in the filter and target blobs are different.
Targets post step:
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0 0 0
0 0 0.175913 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.176915 0 0 0 0 0 0 0 0
0 0 0.176081 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0 0 0
0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0 0.173392 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.176585
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0 0 0
0 0.173022 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.174964 0 0 0 0 0 0 0
0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0 0.174004 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.174004 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.173392 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0.17581 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.174467 0 0 0 0 0 0 0
0 0.174032 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0 0 0
0 0 0.173349 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0 0 0
0 0 0.175626 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0.177733 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.168742 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.175043 0 0 0 0 0 0 0.171759
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0 1 0
Filters post step:
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
If you look closely all the values are shifted two places to the left. This results in incorrect filtered q values since some actions might not even have a corresponding filter any more. Some rows are all zeros and some rows will have two targets.
Filtered q values:
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00185883 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0725774 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00195366 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0720046 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00185883 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0729674 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0723516 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0462732 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0.186157
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0137082 -0 -0 -0 -0 0 -0 0
0 0.0343856 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 -0 -0.0133068 -0 -0 -0 -0 0 -0 0
0 0 -0.0729674 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0423569 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0423569 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0462732 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0.0324558 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.014098 -0 -0 -0 -0 0 -0 0
0 0.0291945 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0137082 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0111855 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0137082 -0 -0 -0 -0 0 -0 0
0 0 -0.069723 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0111855 -0 -0 -0 -0 0 -0 0
0 0 -0.0663748 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0.0305264 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0723516 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 -0.00127689 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0107717 -0 -0 -0 -0 0 -0 0.180422
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00185883 -0 -0 -0 -0 -0 0 4.17635e-38 0
from dqn-in-the-caffe.
Hi Trevor,
Are you using the head of caffe/master?
For me things work with the version at the link above with a caffe/master
around Sept 2015. THe head of master doesn't would love to know why. See
the link above for a working version with caffe around Sept 2015. If you
start with that then get to the head of caffe main that would be really
helpful.
On Fri, Dec 18, 2015 at 10:20 AM, Trevor Barron [email protected]
wrote:
I worked through the code also and found what looks to be an error when
resetting the filters and targets for each forward pass. The filters and
targets are not the correct values after calling reset and doing one step.
Is there a reason for this? Am I understanding this correctly?After 50 iterations the computed targets for each of the 32 outputs (this
looks reasonable):
0.173218 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0
0 0 0 0 0.175913 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.176915 0 0 0 0 0 0
0 0 0 0 0.176081 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0
0 0 0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0
0.173392 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0.176585 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0
0 0 0 0.173022 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.174964 0 0 0 0 0
0 0 0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0
0.174004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.174004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.173392 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0.17581 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.174467 0 0 0 0 0
0 0 0 0.174032 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0
0 0 0 0 0.173349 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0
0 0 0 0 0.175626 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0.177733 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.168742 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0.175043 0 0 0 0 0
0 0.171759 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0The corresponding filters (this also looks correct as the positions match):
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0After doing one step, though, the data in the filter and target blobs are
different.
Targets post step:
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0 0 0
0 0 0.175913 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.176915 0 0 0 0 0 0 0 0
0 0 0.176081 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0 0 0
0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0 0.173392 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.176585
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0 0 0
0 0.173022 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.174964 0 0 0 0 0 0 0
0 0 0.17177 0 0 0 0 0 0 0 0 0 0 0 0 0 0.174004 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.174004 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.173392 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0.17581 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.174467 0 0 0 0 0 0 0
0 0.174032 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.171213 0 0 0 0 0 0 0
0 0 0.173349 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.172245 0 0 0 0 0 0 0
0 0 0.175626 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0.177733 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0.173809 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.168742 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0.175043 0 0 0 0 0 0 0.171759
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0.177312 0 0 0 0 0 0 1 0Filters post step:
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0If you look closely all the values are shifted two places to the left.
This results in incorrect filtered q values since some actions might not
even have a corresponding filter any more. Some rows are all zeros and some
rows will have two targets.Filtered q values:
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00185883 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0725774 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00195366 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0720046 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00185883 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0729674 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0723516 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0462732 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0.186157
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0137082 -0 -0 -0 -0 0 -0 0
0 0.0343856 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 -0 -0.0133068 -0 -0 -0 -0 0 -0 0
0 0 -0.0729674 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0423569 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0423569 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0.0462732 0
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0.0324558 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.014098 -0 -0 -0 -0 0 -0 0
0 0.0291945 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0137082 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0111855 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0137082 -0 -0 -0 -0 0 -0 0
0 0 -0.069723 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0111855 -0 -0 -0 -0 0 -0 0
0 0 -0.0663748 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0.0305264 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0.0723516 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 -0.00127689 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0 -0.0107717 -0 -0 -0 -0 0 -0 0.180422
0 0 -0 0 0 -0 0 -0 0 0 -0 -0 -0 -0 -0 0 -0 0
0 0 -0 0 0 -0 0 -0 0 0.00185883 -0 -0 -0 -0 -0 0 4.17635e-38 0—
Reply to this email directly or view it on GitHub
#12 (comment)
.
from dqn-in-the-caffe.
How to save the model?
from dqn-in-the-caffe.
Hi,
In the file models/fast_dqn_solver.prototxt there are the lines:
snapshot intermediate results
snapshot: 1000000
snapshot_prefix: "model/dqn"
Currently every 1M steps the model is saved. You can modify this to save
it at some other interval.
-Peter
On Mon, Mar 14, 2016 at 8:43 AM, jiutiandiwang [email protected]
wrote:
How to save the model?
—
Reply to this email directly or view it on GitHub
#12 (comment)
.
from dqn-in-the-caffe.
I have got the same problem. heck failed: bottom[0]->shape() == bottom[i]->shape() bottom[0]: 1 1 100 20 20 (40000), bottom[1]: 1 1 20 20 20 (8000)
What do you mean by inner-product layer ? how to solve it? @omgteam
from dqn-in-the-caffe.
Related Issues (20)
- Parameters used to learn Pong? HOT 2
- Linking problems
- video file
- undefined references to `DisplayScreen::display_screen(MediaSource const&) HOT 4
- Source for pong.bin
- how to build and get_start?
- Is there a python implementation or interface for the same?
- where can download the Arcade Learning Environment? HOT 1
- Screen display requires directive __USE_SDL to be defined
- cannot find #include "caffe/proto/caffe.pb.h"
- which player is controlled by dqn
- Use One network or Two network?
- fatal error: caffe/proto/caffe.pb.h: No such file or directory #include "caffe/proto/caffe.pb.h" HOT 9
- Problem running with GPU? HOT 6
- A doubt at line 350 in dqn.cpp HOT 5
- no good results HOT 27
- Running Problem HOT 3
- How to Resume training? HOT 1
- problem with installing dqn
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from dqn-in-the-caffe.