You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We currently have a script, courtesy of @lichensongs, to visualize the MCTS search process in a web-browser. The script works by loading a JSON search log, which can be generated by configuring an mcts::Manager appropriately.
Currently, the only spot where we do such a configuration is in cpp/src/mcts/main/UnitTests.cpp (here). The context is for testing against a golden file for unit-testing purposes, using a mock neural-network evaluation service.
It would be generally useful to expose a way to generate such a search log outside of this context. Perhaps something like:
This would create a blank game, apply the given actions, and then do an MCTS search on the resultant state using the passed-in model. See the test_search() function in cpp/src/mcts/main/UnitTests.cpp.
Notes:
We probably want a new c++ binary for this purpose. If we match the existing structure of our cmake files, this would mean making repetitive boilerplate changes in each cpp/src/games/*/CMakeLists.txt.
The current visualizer script (py/mcts_viz/web_viz.py) assumes that the json log file you want to visualize lives in sample_search_logs/mcts_tests/. This assumption becomes inappropriate once we add a tool like the above.
The text was updated successfully, but these errors were encountered:
We currently have a script, courtesy of @lichensongs, to visualize the MCTS search process in a web-browser. The script works by loading a JSON search log, which can be generated by configuring an
mcts::Manager
appropriately.Currently, the only spot where we do such a configuration is in
cpp/src/mcts/main/UnitTests.cpp
(here). The context is for testing against a golden file for unit-testing purposes, using a mock neural-network evaluation service.It would be generally useful to expose a way to generate such a search log outside of this context. Perhaps something like:
This would create a blank game, apply the given actions, and then do an MCTS search on the resultant state using the passed-in model. See the
test_search()
function incpp/src/mcts/main/UnitTests.cpp
.Notes:
cpp/src/games/*/CMakeLists.txt
.py/mcts_viz/web_viz.py
) assumes that the json log file you want to visualize lives insample_search_logs/mcts_tests/
. This assumption becomes inappropriate once we add a tool like the above.The text was updated successfully, but these errors were encountered: