How can I make anaconda environment file which could be use on other computers?
I exported my anaconda python environment to YML using conda env export > environment.yml
. The exported environment.yml
contains this line prefix: /home/superdev/miniconda3/envs/juicyenv
which maps to my anaconda's location which will be different on other's pcs.
conda env create -f environment.yml
Just as an aside, in my experience this is not going to work across platforms, because conda env will list many dependencies such as ` vs2015_runtime` if you are on Windows. But of course that is not available on linux.
environment.yml
file by hand, so you could leave it out - see stackoverflow.com/questions/39280638/…
I can't find anything in the conda
specs which allows you to export an environment file without the prefix: ...
line. However, like Alex pointed out in the comments, conda doesn't seem to care about the prefix line when creating an environment from the file.
With that in mind, if you want the other user to have no knowledge of your default install path, you can remove the prefix line with grep
before writing to environment.yml
.
conda env export | grep -v "^prefix: " > environment.yml
Either way, the other user then runs:
conda env create -f environment.yml
and the environment will get installed in their default conda environment path.
If you want to specify a different install path than the default for your system (not related to 'prefix' in the environment.yml), just use the -p
flag followed by the required path.
conda env create -f environment.yml -p /home/user/anaconda3/envs/env_name
Note that Conda recommends creating the environment.yml
by hand, which is especially important if you are wanting to share your environment across platforms (Windows/Linux/Mac). In this case, you can just leave out the prefix
line.
The easiest way to save the packages from an environment to be installed in another computer is:
$ conda list -e > req.txt
then you can install the environment using
$ conda create -n <environment-name> --file req.txt
if you use pip
, please use the following commands: reference https://pip.pypa.io/en/stable/reference/pip_freeze/
$ env1/bin/pip freeze > requirements.txt
$ env2/bin/pip install -r requirements.txt
Linux conda env export --no-builds | grep -v "prefix" > environment.yml
Windows conda env export --no-builds | findstr -v "prefix" > environment.yml
Rationale: By default, conda env export
includes the build information:
$ conda env export
...
dependencies:
- backcall=0.1.0=py37_0
- blas=1.0=mkl
- boto=2.49.0=py_0
...
You can instead export your environment without build info:
$ conda env export --no-builds
...
dependencies:
- backcall=0.1.0
- blas=1.0
- boto=2.49.0
...
Which unties the environment from the Python version and OS.
First activate your Conda environment (Below, myenv is the supposed name of the environment).
conda activate myenv
Then you just need to run this command
conda env export > environment.yml
Note that you could replace environment.yml with any other filename of your choice.
First activate your conda environment (the one u want to export/backup)
conda activate myEnv
Export all packages to a file (myEnvBkp.txt)
conda list --explicit > myEnvBkp.txt
Restore/import the environment:
conda create --name myEnvRestored --file myEnvBkp.txt
I find exporting the packages in string format only is more portable than exporting the whole conda
environment. As the previous answer already suggested:
$ conda list -e > requirements.txt
However, this requirements.txt
contains build numbers which are not portable between operating systems, e.g. between Mac
and Ubuntu
. In conda env export
we have the option --no-builds
but not with conda list -e
, so we can remove the build number by issuing the following command:
$ sed -i -E "s/^(.*\=.*)(\=.*)/\1/" requirements.txt
And recreate the environment on another computer:
conda create -n recreated_env --file requirements.txt
sed
command worked pretty nice. I had to delete a patch version. So major.minor.patch
to major.minor
and it worked. Turns out, the lower patch number had been dropped by the main repos.
For me the procedure given by Conda worked:
But only if executed in an anaconda prompt/ console.
VSCode terminal only generated a yml-file with a path to my environment, but no explicit list of packages in the dependencies.
--from-history
option, pip
packages are not included. So if you have some of those you should add them manually (could be copied from the full export).
For Windows Users
Open CMD Conda Activate SpidersEnv conda env export > SpidersEnv.yml Then browse the user folder and search for SpidersEnv.yml
this is also another way to backup (export) your environment
Success story sharing
prefix
helps create new environments faster when used on the same machine, but I don't have much evidence for this claim. I once ran without theprefix
and it had to download some extra packages that it didn't need to do before. And yet, that only happened once, so maybe after that they were cached on a system level(?) So, not needed for sharing, but maybe desirable when just using it for yourself.