-
-
This is still pending my completion of the python->c++ port of the MCTS logic, of course. This is taking me longer than one might expect because I am building in all the parallelization niceties that …
-
Hello, sorry to bother you. Not sure if that is even Katago's problem or something else. Have been using kata under Linux for years using OpenCL with an AMD GPU (RX570). Recently I switched to a RX677…
-
When I use the "analyse" sub-command to analyse the current board state, I find that the whiteOwnerMap looks like the result that AI has make a next move. However, what I want is the whiteOwnerMap of …
-
So a had an idea that perhaps adding auxiliary outputs/targets for previous the previous n moves and/or states. I think doing so might make the use of historical board states as inputs redundant as th…
-
I'm trying to use KataGo to get the dead stone information of a certain position. I saw the `final_status_list dead` command. How can I implement it in Python? Does this command return the captured st…
-
*Stuff might be edited in and out of this comment as we go... note that these are ideas and suggestions, not necessarily things I actually want or will implement...*
* [x] Deltas graph (not sure if…
-
Hi,
I presume the setting for score mean with/without komi is reversed. When I set it to "score mean" without komi, I expect a score estimate of around 6.5 on the empty board. When set to "score me…
w3333 updated
3 years ago
-
ROCm?
-
It is possible to make the neural network predict not only a real number but also uncertainty around it.
This is done by replacing typical L2 loss (i.e. log of a normal distribution with a constant v…