Skip to content

Latest commit

 

History

History
129 lines (95 loc) · 3.8 KB

README.md

File metadata and controls

129 lines (95 loc) · 3.8 KB
Raw

  Meta-packages

Overview

These packages provide one-line installs for RAPIDS as well as environment setups for RAPIDS users and the RAPIDS containers.

Meta-packages

Package Availability

These meta-packages are available in two channels:

Channel Name Purpose
rapidsai Release versions of the packages; tied to a stable release of RAPIDS
rapidsai-nightly Nightly versions of the packages; allows for install of WIP nightly versions of RAPIDS

Install Packages

The install meta-packages are for RAPIDS installation and version pinning of core libraries to a RAPIDS release:

Package Name Purpose
rapids Provide a one package install for all RAPIDS libraries, version matched to a RAPIDS release
rapids-xgboost Defines the version of xgboost used for a RAPIDS release

Managing Versions

Packages without version restrictions do not need to use the following process and can be simply added as a conda package name to the recipe. For all other packages, follow this process to add/update versions used across all meta-packages:

  1. Examine the meta.yaml recipe to be modified
  2. Check if there is a pre-existing version definition like
cupy {{ cupy_version }}
  1. If so, skip to the section Updating Versions
  2. If not, continue with the section Adding Versions

Adding Versions

For new packages or those that do not have defined versions they need to be added.

Modifying Recipes

To add a package with versioning to the recipe we need the PACKAGE_NAME and the VERSIONING_NAME added to the file.

  • PACKAGE_NAME - is the conda package name
  • VERSIONING_NAME - is the conda package name with - replaced with _ and a suffix of _version added
    • For example
      • cupy would become cupy_version
      • scikit-learn would become scikit_learn_version

Once the PACKAGE_NAME and VERSIONING_NAME are ready, we can add them to the meta.yml as follows:

PACKAGE_NAME {{ VERSIONING_NAME }}
  • NOTE: The VERSIONING_NAME must be surrounded by the {{ }} for the substitution to work.

Using our examples of cupy and scikit-learn we would have these entries in the meta.yaml:

cupy {{ cupy_version }}

scikit-learn {{ scikit_learn_version }}

Modifying Versions File

In conda/recipes is versions.yaml - These are versions used by CI for testing in PRs and conda builds.

In this file we specify the version for the newly created VERSIONING_NAME.

For each VERSIONING_NAME we need a VERSION_SPEC. This can be any of the standard conda version specifiers:

>=1.8.0
>=0.48,<0.49
>=7.0,<8.0.0a0
=2.5
TIP - Correct version specs

NOTE: =2.5.* is not a valid version spec. Please use =2.5 instead which will be interpreted as =2.5.*. Otherwise conda build throws a warning message with the definition of .*. For example:

WARNING conda.models.version:get_matcher(531): Using .* with relational operator
is superfluous and deprecated and will be removed in a future version of conda.
Your spec was 0.23.*, but conda is ignoring the .* and treating it as 0.23

Combined together each of the versions files would add the following for each VERSIONING_NAME:

VERSIONING_NAME:
  - 'VERSION_SPEC'

Using our examples of cupy and scikit-learn we would have these entries in the meta.yaml:

cupy_version:
  - '>=7.0,<8.0.0a0'

scikit_learn_version:
  - '=0.21.3'

Updating Versions

Edit the versions.yaml file in conda/recipes and update the VERSION_SPEC as desired. If there is no defined version spec, see Modifying Versions Files for information on how to add one.