-
## Trying out Squeeze and Excitation.
Looks really good.
Code: https://github.com/tensorflow/minigo/pull/673 (Brian would be nice for you to take a look)
Inspirations:
* https://arxiv.org/abs/…
-
when I use python gtp.py for 000990-pallas, it outputs following error.
How to solve it?thanks.
I0521 23:15:21.790073 140320694859520 saver.py:1284] Restoring parameters from /home/gzd/Oth…
-
the progress is very slow now , 40 blocks may reach the limit.
so, why not explore 50 or 60 blocks from now on?
-
all of the networks after lz226 have winrates less than 0.45
-
policy.py第110行:
for name, thing in locals().items():
应该改为:
for name, thing in list(locals().items()):
否则会出现异常RuntimeError: dictionary changed size during iteration
-
I tested the strength of the LZ225, playing in the equivalent time,
Use 2 x Playout with ElfV0 and ElfV2.
Use 3 x Playout with LZ_157 and 15b_224.
LZ_225 is the strongest in the equivalent time, …
-
Hi
I'm further reading the source code, wanted to clarify my understanding of concurrency/parallelism implementation is correct.
_Inference_ options:
* `FakeDualNet` - returns fixed policy/va…
-
LZ training server switched to -m 30 on 10-Jan-2019, IIRC. 'All' networks up to LZ200 were trained only with selfplay games ran with -m 999 ('all' = the last 9 months at least; I didn't retrieved the …
-
Of course that 20 block net has the fixed limit.
But I don't think that they moved to 40 block due to the limit of network.
(This opinion became popular after Hideki Kato which is developer of DeepZ…
-
I train the 15b weight on 40b self-games from time to time. I will update the weight here when it beats the previous best in my test.
2019/2/3
[edb61bc2](https://www.dropbox.com/s/u55b8qg9uw6tgn5/…