[Tune] Will `pip install xgboost-ray==0.1.0` work with ray 2.0.dev?

Ray Libraries (Data, Train, Tune, Serve) Ray Tune
1

to use xgboost-ray with ray 2.0 dev branch, can 0.1.0 be used, or should it be from head of dev in git:

pip install git+https://github.com/ray-project/xgboost_ray.git#xgboost_ray

2

The XGBoost-Ray CI makes sure that it always works with both the latest master and the latest release (currently 1.3.0) - so you should be fine either way!

3

Thanks, will give it a try!

4

i’m doing a trial-run of your xgb training example:
https://docs.ray.io/en/master/xgboost-ray.html#usage

it loads the cancer dataset but the training run fails with:

ConnectionError: Cannot send request due to data channel shutting down. Request: req_id: 3
put {
  data: "\200\005\225y\007\000\000\000\000\000\000\214\033ray.cloudpickle.cloudpickle\224\214\r_builtin_type\224\223\224\214\nLambdaType\224\205\224R\224(h\002\214\010CodeType\224\205\224R\224(K\000K\001K\005K\021K\037C\222t\000\203\000s\020\210\000|\001|\002\216\001S\000t\001j\002t\003\203\001}\003|\000d\000k\ts6t\004d\001|\001\233\
... more

I am using ray 2.0 build from yesterday, and running inside k8s, connecting to a ray head node via the new client/server connection (from a jupyter notebook). FWIW, I ran through a basic dask example earlier on the same environment, so the basic ray and ray+dask is operating properly.

5

I restarted my ray cluster and now the xgboost example runs fine. Unsure what caused the ConnectionError but the example itself works now.