[continuumio/anaconda3:latest] TensorFlow package installation consumes much more time now about docker-images HOT 2 CLOSED

@alexander-smolyakov you are mixing different channels here:

• the anaconda3 container has all the anaconda3 packages from defaults installed into the so called base env
• the miniconda3 container has only conda + its dependencies in the base env
• you try to install tensorflow / pytorch from conda-forge, which is an entire different channel compared to defaults!
• it is best practise to leave the base env either untouched (and instead create a new env with the packages from a different channel) OR install tensorflow / pytorch from defaults.... reason is that installing packages from different channels with many dependencies easily causes conflicts.... the more there already is in the base env (lots of stuff in the anaconda3 case) the more likely this happens.

in other words:

FROM continuumio/miniconda3

# lets create a new env
RUN conda create --name test-env -c conda-forge --yes \
    tensorflow \
    pytorch

OR

FROM continuumio/anaconda3

# lets install into base, but only packages from defaults (no extra --channel parameter)
RUN conda install --yes \
    tensorflow \
    pytorch

from docker-images.

@dbast, thanks for your reply. I tested the suggested options. Creating a new environment works fine.

OR install tensorflow / pytorch from defaults....

FROM continuumio/anaconda3

# lets install into base, but only packages from defaults (no extra --channel parameter)
RUN conda install --yes \
    tensorflow \
    pytorch

When trying installation from the default channel, the same issue appears, and conda tries to resolve some conflicts in the environment during installation which take a lot of time:

=> [2/2] RUN   echo "conda install tensorflow"   && conda install --yes tensorflow   && echo "conda install pytorch"   && conda install --yes pytorch             676.8s
 => => # Collecting package metadata (current_repodata.json): ...working... done
 => => # Solving environment: ...working... unsuccessful initial attempt using frozen solve. Retrying with flexible solve.
 => => # Solving environment: ...working... unsuccessful attempt using repodata from current_repodata.json, retrying with next repodata source.
 => => # Collecting package metadata (repodata.json): ...working... done
 => => # Solving environment: ...working... unsuccessful initial attempt using frozen solve. Retrying with flexible solve.
 => => # Solving environment: ...working...

I wander if something is broken in the base environment.

from docker-images.