lichess-org / kaladin Goto Github PK

Lichess usually goes with AGPLv3

Currently ratingDiff/date is causing more false positives than expected.

This is most likely due to stale date filters when selecting accounts for the initial data pull.

Need to add new accounts into the training data and re-train

180 day limit should begin with the most recent game played in the time control.

Hello,

I've been trying to use your Docker setup and scripts, but I'm encountering some issues. It seems that the scripts are not functioning as expected, and I'm having trouble understanding what input data the model uses.

I was wondering if it's possible to provide a pre-populated MongoDB database or some sample data to better understand what the model expects as input. This would greatly help me in getting started with your project and troubleshooting the problems I'm facing.

Any guidance or assistance you can provide would be much appreciated. Thank you!

(The repo is very well-written, and I would like to participate in potentially adapting it to more recent Python versions and PyTorch)

For each user, this warning is logged:

WARNING:tensorflow:11 out of the last 11 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7fde4224f840> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.

To map the kaladin docker UID to the system UID, it would be nice if the kaladin docker user was UID/GID 1002, since that's kaladin's UID on the server.

Could be simply RUN useradd kaladin -u 1002 in the Dockerfile, it should not break anything else.

Insights in cheat reports should have a different period for each player, based on the earliest game in the data considered by Kaladin.

Sometimes this will be less than 180 days (for very active players), other time it will be longer than 180 days (for players who haven't played in a TC for some time).