Version 1.6 Availability On Conda Forge

Hi,

I see that XGBoost has release a stable 1.6 version, however is doesn’t seem to be available in the conda-forge channel, though there are previous versions available, such as 1.5.1.

I’m particular interested in version 1.6 as there is a huber_slope parameter now available for the pseudo huber loss.

What would the process be for XGBoost version 1.6 to reach the conda-forge channel?

I have noticed that there is the possibility of installing version 1.6 manually, but I am concerned about dependencies concerning the other packages which I am using in Python. Conda has so far been successful in maintaining the dependency requirements between the various packages I have installed.

Is there perhaps another channel that may host XGBoost 1.6 or has it not reached the conda-forge channel for whatever reason?

Thanks in advance

Vaughan

I submitted https://github.com/conda-forge/xgboost-feedstock/pull/88 to upload XGBoost 1.6.0 on Conda-forge.

Hi hcho3

Thanks for your assistance.

Hi hcho3,

Out of interest, what is the process to push version 1.6 of XGBoost to Conda-Forge, how long would it take?

Kind regards
Vaughan

https://github.com/conda-forge/xgboost-feedstock/pull/88 is currently stalled due to an unknown build failure. Not sure when I’ll get to fix it

I just want to bump this as well.

There’s an issue with the way xgboost interacts with pandas.

In this forum, the developers said it would be fixed in version 1.6.1 (See here for thread).

You can also see people asking about it here over at StackOverflow.

I’m running all this out of a virtual environment via Anaconda, and when I go to the xgboost documentation website, they seem to imply the following command should get me to 1.6.1 (or perhaps the latest version supported by Anaconda.

conda install -c conda-forge py-xgboost

But this does not seem to have support in the stable Anaconda release channel, if that’s the right word.

Just like @vaughan said, it looks like there is a way to install it manually on the system outside of Anaconda, but we’d rather do it safely within Anaconda, so as not to disturb the ecosystem of interdependencies.

So I’m bumping this thread now to see if there are any updates to this, or we all stuck on 1.5.1 of xgboost in Anaconda (which is not the end of the world).