2

I'm trying to run H2o's automl and I want to see the results of XGboost in automl. When I'm trying to run this code:

aml1 <- h2o.automl(y = y, x = x, training_frame = train, keep_cross_validation_models = F, seed = 123)

I'm getting this message:

  |                            |   0% 16:25:37.796: AutoML: XGBoost is not available; skipping it. Job $03017f00000132d4ffffffff$_8c51b0759fd2e77fd8940b41f83340c2 was cancelled.

I'm using H2o 3.32.1.1 version on a Centos 7 linux server. I'm using R 3.6.0 version. I installed H2o using:

install.packages("h2o", type="source", repos=(c("http://h2o-release.s3.amazonaws.com/h2o/latest_stable_R"))) as given in: https://docs.h2o.ai/h2o/latest-stable/h2o-docs/downloading.html

Please help. I don't know why xgboost is not running in automl. And, how it could be fixed. I couldn't find any solution online yet to solve this issue.

EDIT

Issue got resolved by installing xgboost package in R. Thanks guys for your comments and directing me towards installing xgboost in R.

sm925
  • 2,648
  • 1
  • 16
  • 28
  • From the docs: "XGBoost is used only if it is available globally and if it hasn’t been explicitly disabled. You can check if XGBoost is available by using the h2o.xgboost.available() in R"; so it looks like you need to install xgboost globally (https://xgboost.readthedocs.io/en/latest/build.html#building-on-linux-distributions) – jared_mamrot Apr 20 '21 at 23:19
  • 1
    @jared_mamrot That's not correct, you don't need to install the xgboost R package to use XGBoost in H2O. They are totally separate things. H2O comes bundled with XGBoost unless you by-hand, specifically build a version w/o XGB support. The only place XGBoost doesn't work via H2O is in Windows, so this should work on CentOS. Can you clarify how you resolved the issue? Installing the xgboost R package would not fix anything with H2O, so I am not sure what that means. – Erin LeDell Apr 27 '21 at 22:36
  • I understand that H2O comes bundled with xgboost, and that installing xgboost through R should not be necessary, but my advice is based on my own experience with a similar package (not H2O) where xgboost couldn't locate 'something' and installing xgboost globally fixed my issue. It also appears to have worked in this case (from the question edit above): "Issue got resolved by installing xgboost package in R. Thanks guys for your comments and directing me towards installing xgboost in R." but I cannot say _why_ this workaround would solve the problem sorry – jared_mamrot Apr 28 '21 at 00:29
  • @jared_mamrot I think installing xgboost in R brought in some dependencies that were missing. That is some great out-of-the-box thinking, I wouldn't thought of that - as would probably most of the other users. We need to get better at reporting why XGBoost is not available. I would like to reproduce your issue - can you share details of your Centos 7 installation? What version exactly, what kind of install - eg. "minimal install" / "compute node"/...? – Michal Kurka Apr 28 '21 at 14:49
  • @ErinLeDell I have posted in my edit that installing xgboost package in R solved the issue. Erin, it did fix the issue. After installing xgboost in R; h2o's automl started building XGBoost models as well (earlier it was giving a warning that XGBoost not available; so skipping it). jared_mamrot I think you're right installing xgboost fixed something globally because of which H2o's automl started building XGBoost models. Michal Kurka sure I'll share my Centos 7 installation details – sm925 Apr 28 '21 at 23:38
  • @jared_mamrot Thanks for helping - i am surprised that this would affect it, but I am willing to believe it did, even though I don't quite yet understand how. Usually I just assume there's some context missing when something doesn't make sense and I was hoping to get more info... sorry for making assumptions. – Erin LeDell Apr 29 '21 at 07:05

0 Answers0