1.7.0 Release Candidate

[trivialfis]

We are about to release version 1.7.0 of XGBoost. In the next week, we invite everyone to try out the release candidate (RC).

Roadmap: #8282
Release note: #8374

Feedback period: until the end of October 30, 2022. No new feature will be added to the release; only critical bug fixes will be backported.

@dmlc/xgboost-committer

Available packages:

• Python packages:

pip install xgboost==1.7.0rc1

• R packages:

R binary packages with CUDA enabled for testing:

Linux x86_64: Download
Windows x86_64: Download

sha256sum:

ebaa8517b673735ae87b3b27d2f744b77d61974d4411c161455eefd1cf59cc39  xgboost_r_gpu_linux_1.7.0-rc1.tar.gz
75f678ea669a82459f506c3c0cc613b5e8fa20a6898b4d883e6cf846f1a39c16  xgboost_r_gpu_win64_1.7.0-rc1.tar.gz

Install:

R CMD INSTALL ./xgboost_r_gpu_linux.tar.gz

• JVM packages

Show instructions (Maven/SBT)

Maven

<dependencies>
  ...
  <dependency>
      <groupId>ml.dmlc</groupId>
      <artifactId>xgboost4j_2.12</artifactId>
      <version>1.7.0-RC1</version>
  </dependency>
  <dependency>
      <groupId>ml.dmlc</groupId>
      <artifactId>xgboost4j-spark_2.12</artifactId>
      <version>1.7.0-RC1</version>
  </dependency>
</dependencies>

<repositories>
  <repository>
    <id>XGBoost4J Release Repo</id>
    <name>XGBoost4J Release Repo</name>
    <url>https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/</url>
  </repository>
</repositories>

SBT

libraryDependencies ++= Seq(
  "ml.dmlc" %% "xgboost4j" % "1.7.0-RC1",
  "ml.dmlc" %% "xgboost4j-spark" % "1.7.0-RC1"
)
resolvers += ("XGBoost4J Release Repo"
              at "https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/")

Starting from 1.2.0, XGBoost4J-Spark supports training with NVIDIA GPUs. To enable this capability, download artifacts suffixed with -gpu, as follows:

Show instructions (Maven/SBT)

Maven

<dependencies>
  ...
  <dependency>
      <groupId>ml.dmlc</groupId>
      <artifactId>xgboost4j-gpu_2.12</artifactId>
      <version>1.7.0-RC1</version>
  </dependency>
  <dependency>
      <groupId>ml.dmlc</groupId>
      <artifactId>xgboost4j-spark-gpu_2.12</artifactId>
      <version>1.7.0-RC1</version>
  </dependency>
</dependencies>

<repositories>
  <repository>
    <id>XGBoost4J Release Repo</id>
    <name>XGBoost4J Release Repo</name>
    <url>https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/</url>
  </repository>
</repositories>

SBT

libraryDependencies ++= Seq(
  "ml.dmlc" %% "xgboost4j-gpu" % "1.7.0-RC1",
  "ml.dmlc" %% "xgboost4j-spark-gpu" % "1.7.0-RC1"
)
resolvers += ("XGBoost4J Release Repo"
              at "https://s3-us-west-2.amazonaws.com/xgboost-maven-repo/release/")

PRs to be backported

• Use bst_ulong instead of size_t for CSR conversion #8369
• [doc] Cleanup outdated documents for GPU. [skip ci] #8378
• Fix CUDA async stream. #8380
• [pyspark] Improve tutorial on enabling GPU support. #8385
• Fix building XGBoost with libomp 15 #8384

[trivialfis]

All packages are uploaded.

[hcho3]

The 1.7.0 stable version is now available on PyPI.

[trivialfis]

Will upload R package to CRAN as well.

[trivialfis]

Submitted R package to cran, let's track https://cran.r-project.org/web/checks/check_results_xgboost.html .

[trivialfis]

@hcho3 Could you please help make the jvm release when you are available?

[trivialfis]

pre-tests failed with following notes:

Flavor: r-devel-linux-x86_64-debian-gcc, r-devel-windows-x86_64
Check: compilation flags in Makevars, Result: NOTE
  Package has both 'src/Makevars.in' and 'src/Makevars'.
  Installation with --no-configure' is unlikely to work.  If you intended
  'src/Makevars' to be used on Windows, rename it to 'src/Makevars.win'
  otherwise remove it.  If 'configure' created 'src/Makevars', you need a
  'cleanup' script.

Flavor: r-devel-linux-x86_64-debian-gcc
Check: examples, Result: NOTE
  Examples with CPU time > 2.5 times elapsed time
                       user system elapsed ratio
  cb.gblinear.history 1.454  0.017    0.49 3.002

Need to cleanup the automake files, not sure what to do with the CPU time.

[trivialfis]

Posted a question on the r devel list about the CPU time.

[trivialfis]

Running tests on winbuilder for CRAN submission again.

[hcho3]

JVM packages have been available on Maven Central since Nov 4

[trivialfis]

I made another attempt for R package today, still failing for the same error with cb.gblinear.history even after reducing the number of threads to 1. I haven't been able to figure out the exact cause.

[trivialfis]

@hcho3 @hetong007 @RAMitchell We might skip 1.7 for CRAN due to non-trivial changes required to pass the CRAN tests. At the moment, we are running into the error #8366 (comment) . I haven't been able to figure out the exact cause since I can't reproduce it on my devices (tried both 16 cores and 12 cores machines).
Also, there are known issues like #8455

[trivialfis]

I made another submission. Fingers crossed.

[hcho3]

@trivialfis We got another warning:

Found the following significant warnings:
  src/predictor/cpu_predictor.cc:167:27: warning: narrowing conversion of '(long long unsigned int)beg' from 'long long unsigned int' to 'size_t' {aka 'unsigned int'} inside { } [-Wnarrowing]
See https://www.r-project.org/nosvn/R.check/r-oldrel-windows-ix86+x86_64/xgboost-00install.html for details.

https://www.r-project.org/nosvn/R.check/r-oldrel-windows-ix86+x86_64/xgboost-00check.html

Should we consider setting up CI for 32-bit envs?

[trivialfis]

@hcho3 I don't think we will maintain a 32-bit package, its tests are failing anyway. Unless there's another message from the cran maintainer we might just leave it unsupported.

[hcho3]

All we have to do is to ensure that the code builds without warning using 32-bit system. We've already disabled all tests for 32-bit.

[trivialfis]

there's an error in examples, you can search "ERROR" in the link you shared.

[hcho3]

Let's go ahead and disable all tests for 32-bit

[trivialfis]

@hcho3

Should we consider setting up CI for 32-bit envs?

Would you like to help with this? Maybe an optional test like R nold?

[hcho3]

Sure, I'll do it

[trivialfis]

Thank you! I will come back to this tomorrow.