It was recently announced that IBM Watson Machine Learning Community Edition now includes two auxiliary conda channels – the Early Access Channel and the Supplementary Channel. Here we will provide step-by-step instructions and best practice recommendations to use each of these conda channels.
Early Access Channel
With the WML CE focus on providing a consumable distribution for enterprise level machine learning tasks, there is often a delay between when the upstream community releases a new version of a framework or tool and when it is included in an official WML CE release. The WML CE Early Access Channel provides packages ahead of an official WML CE release. Packages in this channel have been built and preliminarily tested by IBM, but are not fully supported.
Using the Early Access Channel
The recommended way to use the early access conda channel is to add it to specific environments that need access to the experimental packages, instead of adding it to the global conda channel list. This ensures that only the environments intended to work with experimental packages have access to them and production environments are not compromised. Since the early access packages have the same names as the released packages, it is also important to have the early access channel come before the main channel:
conda create -n wml_ce_test conda activate wml_ce_test conda config --env --prepend channels https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda-early-access/
The early access channel will now be at the top of
channels: - https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda-early-access/ - defaults
Early access packages can now be installed in this environment using the normal commands. E.g.:
conda activate wml_ce_test conda install tensorflow-gpu=2.1 Collecting package metadata (repodata.json): done Solving environment: done ... The following packages will be downloaded: package | build ---------------------------|----------------- ... tensorflow-2.1.0 |gpu_py37_914.g4f6e601 3 KB https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda-early-access tensorflow-base-2.1.0 |gpu_py37_e5bf8de_72632.gbc9303f 597.3 MB https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda-early-access tensorflow-estimator-2.1.0 |py37_7ec4e5d_1461.g4f6e601 603 KB https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda-early-access tensorflow-gpu-2.1.0 | 914.g4f6e601 2 KB https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda-early-access ... ------------------------------------------------------------ ... The following NEW packages will be INSTALLED: ... tensorflow ibmdl/export/pub/software/server/ibm-ai/conda-early-access/linux-ppc64le::tensorflow-2.1.0-gpu_py37_914.g4f6e601 tensorflow-base ibmdl/export/pub/software/server/ibm-ai/conda-early-access/linux-ppc64le::tensorflow-base-2.1.0-gpu_py37_e5bf8de_72632.gbc9303f tensorflow-estima~ ibmdl/export/pub/software/server/ibm-ai/conda-early-access/linux-ppc64le::tensorflow-estimator-2.1.0-py37_7ec4e5d_1461.g4f6e601 tensorflow-gpu ibmdl/export/pub/software/server/ibm-ai/conda-early-access/linux-ppc64le::tensorflow-gpu-2.1.0-914.g4f6e601 ...
Note: Confirm that the packages are the correct version and are coming from
Packages in the WML CE Supplementary Channel are curated and tested, but they are not supported with fixes and security updates like the packages in the main channel. The conda recipes for the packages in this channel are hosted on our Github.com repository and are open for contributions or suggestions.
Using the Supplementary Channel
The supplementary channel does not provide newer copies of packages in the main channel, so adding it to the global channel list shouldn’t cause a problem, and order is not as important, although it should still probably come before the default channels.
conda config --add channels https://anaconda.org/powerai
~/.condarc file may look something like:
channels: - https://anaconda.org/powerai - defaults
Alternative usage of the Supplementary Channel
Since the supplementary channel is hosted at Anaconda.org, it is fairly straightforward to install packages directly from the channel, instead of modifying
conda create -n wml_ce_test conda activate wml_ce_test conda install <package> -c powerai