mcinstall 0.3.0

Quick-install/provision a fresh Miniconda distribution from scratch.

A script to quickly make/provision a fresh Miniconda installation from scratch.

The goal of this script is to quickly install a fresh Miniconda across different operating systems and use it in a CI/CD context, too. It will download a Miniconda binary based on your operating system, unpack the binary and install it locally. It was tested on MacOS and Linux ok, has decent Windows support and also aims to support ARM6 and ARM7 on the Raspberry Pi via Berryconda (sadly no longer maintained).

This script has no external dependencies, but expects to be run with Python 3.6+.

N.B. This project is often updated online which is why its git history might look strange…

Installation

You can use pip (for Python 3) to install and uninstall this package like the following:

pip install mcinstall
pip uninstall mcinstall

More recent, not yet released, versions can be installed like the following without cloning the repo:

pip install -e git+https://github.com/deeplook/mcinstall#egg=mcinstall

Or you clone the repo locally and install the package like this:

git clone https://github.com/deeplook/mcinstall.git
cd mcinstall
pip install -e .

Sample Usage

If you run it like this:

mcinstall ~/Downloads/mc3

it will run a command like this on macOS (using some defaults and decent system introspection) to create a conda base installation:

bash Miniconda3-latest-MacOSX-x86_64.sh -b -f -p ~/Downloads/mc3

This can be activated then with a command like this:

source ~/Downloads/mc3/bin/activate

Windows Example:

On Windows if you execute this command:

mcinstall %USERPROFILE%\downloads\mc3

it will run a command like this:

start /wait "" Miniconda3-latest-Windows-x86_64.exe /InstallationType=JustMe /RegisterPython=0 /S /D=%USERPROFILE%\downloads\mc3

This conda installation can be activated then with a command like this:

%USERPROFILE%\mc3\condabin\activate

Suggested test:

mcinstall --verbose --pip-dependencies jupyter,torch ~/Downloads/torchy
source ~/Downloads/torchy/bin/activate
python -c "import torch; print('ok')"

Suggested test dependencies files:

$ more ~/Downloads/reqs.txt
asciinema
torch
torchvision

$ more ~/Downloads/env.yml
name: test
channels:
  - conda-forge
dependencies:
  - voila

Sample run (replaced home directory with ~ manually):

$ mcinstall --verbose --pip-dependencies jupyter,torch ~/Downloads/torchy
Making directory ~/Downloads/torchy.
Downloading https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh ...
Copying to ~/Downloads/Miniconda3-latest-MacOSX-x86_64.sh ...
Running command: bash ~/Downloads/Miniconda3-latest-MacOSX-x86_64.sh -b -f -p ~/Downloads/torchy
PREFIX=~/Downloads/torchy
Unpacking payload ...
Collecting package metadata (current_repodata.json): ...working... done
Solving environment: ...working... done

[...]

Preparing transaction: ...working... done
Executing transaction: ...working... done
installation finished.

Running command: ~/Downloads/torchy/bin/pip install jupyter
[...]

Running command: ~/Downloads/torchy/bin/pip install torch
[...]

Run this to start using your fresh Miniconda: "source ~/Downloads/torchy/bin/activate".

Sample log file (mcinstall.log):

wget https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
mv Miniconda3-latest-MacOSX-x86_64.sh ~/Downloads/Miniconda3-latest-MacOSX-x86_64.sh
bash ~/Downloads/Miniconda3-latest-MacOSX-x86_64.sh -b -f -p ~/Downloads/torchy
source ~/Downloads/torchy/bin/activate
~/Downloads/torchy/bin/pip install jupyter
~/Downloads/torchy/bin/pip install torch

TO DO

• improve logging executed commands to reproduce them as a shell script

• add self.clean_dest_path to MinicondaInstaller.__init__

• same with self.download_path

• consolidate Windows support

• improve configuration

• make a sample screencast with asciinema