RuntimeError: Tune is not installed, so `get_tune_resources` is not supported. You can install Ray Tune via `pip install ray[tune]`.

[ksneab7]

Problem

When running the example code for ray.tune integration in the readme,
The script successfully exectures from ray import tune but fails on the line resources_per_trial=ray_params.get_tune_resources())
I think this is due to the TUNE_INSTALLED global set in the xgboost_ray/tune.py file.

It seems that line 15 in that script (from ray.tune import is_session_enabled) is failing due to ray.tune not having that attribute.

Steps to reproduce:

in this section of the readme

Error

Traceback (most recent call last):
  File "./testing_ray_data_load.py", line 46, in <module>
    resources_per_trial=ray_params.get_tune_resources())
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "./venv/lib/python3.11/site-packages/xgboost_ray/main.py", line 484, in get_tune_resources
    return _get_tune_resources(
           ^^^^^^^^^^^^^^^^^^^^
  File "./venv/lib/python3.11/site-packages/xgboost_ray/tune.py", line 180, in _get_tune_resources
    raise RuntimeError(
RuntimeError: Tune is not installed, so `get_tune_resources` is not supported. You can install Ray Tune via `pip install ray[tune]`.

[littlethumb123]

Came across the same issue and reinstall Ray didn't help.

[aditanase]

I am hitting this as well, trying to pre-install the ML libs in a docker image using pip / requirements.txt, on top of the default ray image.

Here's the minimal input that I am using:

pyarrow>=19.0.1
ray[data,serve,train,tune]==2.43.0
lightgbm-ray==0.1.9
xgboost-ray==0.1.19
scikit-learn==1.6.1

I confirmed that the dependencies made it in the image by running pip list.
Not sure what to try next.

[NinaHerrmann]

Same problem. I tried to trace back the problem but could not find a fast fix.

from ray.tune import is_session_enabled

(tune.py l.15) fails but even if this is outcommented, another import fails

from ray.tune.integration.lightgbm import (
    TuneReportCheckpointCallback as OrigTuneReportCheckpointCallback,
)

If there is no chance that this can get fixed soon I might take a look - but as I am not to experienced with python packages
(3.13)