Which installer to use for Miniconda with Python 3.6?

Question

I’ve planned to use Miniconda (MC) and TensirFlow (TF) together, and since TF with Python 3 requires Python 3.4, 3.5 or 3.6, I need to install an older version of MC with Python 3.

However, the installer naming convention on the archive page is a bit unclear. What I want is Python 3.6 x64 for Windows, but for this there seems to be two installers: Miniconda-3.6.0-Windows-x86_64.exe and Miniconda3-3.6.0-Windows-x86_64.exe. What’s the difference between these?

Also, while both of these intuitively seem to indicate Python 3.6, what’s up with other installers such as Miniconda3-4.5.12-Windows-x86_64.exe; for sure, this cannot indicator Python 4.5? What is going on here?

Answer 1

Find older version of conda from here: https://repo.anaconda.com/miniconda/

https://repo.anaconda.com/miniconda/Miniconda3-4.3.11-Linux-x86_64.sh is python 3.6.0

Answer 2

Stick with the miniconda latest, so in your case Miniconda3-latest-Windows-x86_64.exe. Unlike pip, Conda is a very intelligent package manager and will try to match dependency versions so that your installed packages will all work in a given virtual environment. My suggestion is to create a conda virtual environment, and then install TensorFlow inside. This is what I do at work.

Also, FYI, the conda version of TensorFlow has been optimized to run faster than the pip version in many situations see this post from Anaconda.

Example:

    conda create -n tf
    conda activate tf
    conda install tensorflow
    ## alternatively if you have a compatible nvidia gpu:
    conda install tensorflow-gpu