Installing Anaconda

Follow these steps to install and set up your system with Anaconda.

A number of the deep learning frameworks require Anaconda. It is a platform-agnostic data science distribution with a collection of over 1,000 open source packages with free community support.

Anaconda terminology

Anaconda Distribution (AD)
Referred to in this documentation as Anaconda. This is an installable set of Python packages and a Python runtime binary.

WML CE supports Anaconda3. The Anaconda3 default environment is Python 3.7, but it can also create Python 3.6 environments.

The version of Anaconda tested and supported for this release of WML CE is 2019.10.

miniconda
Miniconda is the same as AD, but does not include the complete set, so it is much smaller. It only includes Python, some tools, and certificates. The version of Miniconda tested and supported for this release of WML CE is 4.7.12.1.
conda
This is one of the tools that comes with both miniconda and AD. It is basically the package manager and dependency solver. Anaconda 2019.10 and the latest Miniconda installer comes with 4.7.12, but conda 4.8.1 is the latest. If you are using conda 4.7 or later to upgrade an existing environment, review Notes when using conda 4.7.
Anaconda metapackage
The Anaconda metapackage has no actual content, but lists all of the AD Python packages as prerequisites so they all get installed when you install Anaconda.

Which version of Anaconda should I use?

  • Anaconda 4.6: If you are upgrading an existing WML CE 1.6.1 environment, it is recommended that you continue your existing Anaconda 4.6 installation instead of upgrading the conda tool to 4.7.

    For more information about upgrading WML CE, see Upgrading WML CE.

  • Anaconda 4.7: For a new WML CE installation, it is recommended that you use AD 2019.10 or the latest miniconda version, which is 4.7.12.1 at the time of publication.

Using a fresh Anaconda environment vs. the preinstalled base WML CE Anaconda environment

A successful installation of Anaconda Distribution results in the AD package set installed into the default conda environment. This environment is called the Base environment.

It is important to note that the AD packages have been installed as dependencies of the Anaconda metapackage. It is not recommended to install WML CE in Anaconda's base environment. Instead, it is recommended to create a new conda environment to install WML CE into.

For more information about managing conda environments, review Managing environments in the conda User guide.

Anaconda download locations

Anaconda distribution
https://repo.anaconda.com/archive/
Miniconda distribution
https://repo.anaconda.com/miniconda/

Installing Anaconda

Follow these steps to install Anaconda:
  1. Run the appropriate command:
    IBM® Power®
    bash Anaconda3-2019.10-Linux-ppc64le.sh
    x86_64 
    bash Anaconda3-2019.10-Linux-x86_64.sh
  2. Accept the license agreement
  3. Specify an installation location (default is $HOME/anaconda3)
  4. For setups that have a single Anaconda instance for multiple users, such as Watson Machine Learning Accelerator, reply no to update the .bashrc file. After the installation is complete, export the path with this command:
    export PATH=/opt/anaconda3/bin:$PATH
  5. For other WML CE users, reply yes to allow the installer to update the .bashrc file. In this case, if multiple users are using the same system, each user should install Anaconda individually.
  6. Update your terminal with changes to .bashrc:
    source ~/.bashrc

To uninstall Anaconda, follow these steps: https://docs.anaconda.com/anaconda/install/uninstall/.

Installing Miniconda

Follow these steps to install Miniconda:
  1. Run the appropriate command:
    • IBM Power:
      bash Miniconda3-latest-Linux-ppc64le.sh
    • x86:
      bash Miniconda3-latest-Linux-x86_64.sh
  2. Accept the license agreement.
  3. Specify an installation location (the default is $HOME/miniconda3).
  4. For setups that have a single Anaconda instance for multiple users, such as WML Accelerator, reply no to update the .bashrc file. After the installation is complete, export the path with this command:
    export PATH=$HOME/miniconda3/bin:$PATH
  5. For other WML CE users, reply yes to allow the installer to update the .bashrc file. In this case, if multiple users are using the same system, each user should install Miniconda individually.
  6. Update your terminal with changes to .bashrc:
    source ~/.bashrc

Setting strict channel priority (required for x86, recommended for ppc64le)

Set conda to use strict channel_priority to ensure that the WML CE channel is always used when packages of the same name exist in other channels.
conda config --set channel_priority strict
Tip:

In some instances, an Anaconda 4.6 environment can incorrectly select a package from the general free channel instead of the higher priority WML CE channel, potentially causing an undesired package variant to be installed. You can avoid this action by overriding the Anaconda default_channel by running the following config commands:

conda config --add default_channels https://repo.anaconda.com/pkgs/r
conda config --add default_channels https://repo.anaconda.com/pkgs/main

Notes when using conda 4.7

The 4.7.x series of the conda tool has brought upgrades to both the package indexing function as well as the dependency solver. Since this new function relies on new package metadata, it may not properly upgrade environments that were created using the 4.6.x series of conda, which lack the new metadata entries.

For more information on conda 4.7 and upgrading WML CE, see Upgrading WML CE. In addition to the upgrades in the indexing and solver, conda 4.7 has also removed default access to the deprecated Anaconda Free Channel. The changes in the solver might make conda 4.7 more reluctant to change or remove packages during upgrades or new package installations. In the events of a dependency conflict, installing WML CE into a clean conda environment is recommended.