For starters, I recommend to read first Getting started with conda
Table of contents
- What is conda
- Create new environment
- Export environment
- Activate or deactivate environment
- Update environment
- Install and uninstall package
- Remove environment
- Rename environment
- Uninstalling Anaconda distribution
Conda is a cross-platform environment and package management system. It allows you to create, delete, update, clone, import, export environments and install, uninstall, search, update packages while solving dependency hell. This repo is based on Anaconda Distribution, a popular conda distribution that contains 1,500 scientific packages, pip, Python and of course conda. You can apply the cheat sheet also on miniconda, a "smaller" version of Anaconda that contains a smaller amount of packages, pip, Python and conda.
Have a look at this comparison between Anaconda and miniconda
If you want to create an empty environment you can use the create command followed by the $ENVNAME and specifying that you don't want any package installed:
conda create --name $ENVNAME --no-default-packages
Of course you can create a new environment with a specified Python version or packages to install:
conda create --name $ENVNAME python=3.8
If you want to list all the available environments install, use the command env list:
conda env list
This command will put all the environments and location:
# conda environments:
base * C:\Users\Riccardo\anaconda3
env_1 C:\Users\Riccardo\anaconda3\envs\env_1
env_2 C:\Users\Riccardo\anaconda3\envs\env_2
env_3 C:\Users\Riccardo\anaconda3\envs\env_3
As you can see there is a * near the base environment, this will help you to understand the environment in use in that moment.
Sometimes you have already setup an environment used as baseline and want to add new packages, this procedure can be done by cloning a specified environment. You have to use the same create command followed by the environemtn to clone with --clone command:
conda create --name $ENVNAME --clone base
YAML is a human-readable data-serialization language mainly used as configuration file. In our case, .yml file contains all the required packages to create and reproduce an environment.
With standard command create by imposing the .yml file path we can create a new environment as described internally in the .yml configuration:
conda env create -n $ENVNAME --file $ENV.yml
If the $ENVNAME has been already specified inside the .yml file, we can just remove the -n $ENVNAME part:
conda env create -f environment.yml
Sometimes you need to export your environment with all the required packages, this is extremely useful we want to setup the same environment on another pc/server/whatever.
In this case just use the export command followed by destination path of the .yml configuration file:
conda env export > environment.yml
After setup many environments maybe you need a command to activate or deactivate a particular one. The commands are pretty straight forward, use activate command followed by the $ENVNAME to activate an environment:
conda activate $ENVNAME
Instead, if you want to deactivate the environment just type deactivate command followed by the $ENVNAME:
conda deactivate $ENVNAME
If you want to deactivate and environment and activate another one you don't really need to deactivate the environment first but you can actually activate the environment by just using the activate command, the previous environment will be deactivated.
Conda as package manager can update all the depencies easily with the update command:
conda update --all
Please note that conda will only update packages installed by conda and not the one installed by pip. To upgrade a package in pip please use the following command:
pip install --upgrade $PACKAGENAME
It is possible to update all the pip packages but it is not recommended due to the complexity to handle environment checks in pip.
Conda as package manager can be useful to install or uninstall packages. Inside a conda environment you can install packages with the standard package manager for Python (pip).
Let's have a look at pip and then conda:
Use the command install followed by the $PACKAGENAME to install it:
conda install $PACKAGENAME
Use the command remove followed by the $PACKAGENAME to uninstall it:
conda remove $PACKAGENAME
The same che be applied to pip package manager. Please use the install command to install a package:
pip install $PACKAGENAME
Use the uninstall command followed by the $PACKAGENAME to uninstall it:
pip uninstall $PACKAGENAME
To list all the packages installed in the environment just simply use:
conda list
The output will be a list of packages with: name, version, build and channel (where the repository come from):
Name Version Build Channel
pytorch 2.0.0 py3.10_cuda11.7_cudnn8_0 pytorch
pytorch-cuda 11.7 h16d0643_3 pytorch
...
tensorboard 2.12.2 pyhd8ed1ab_0 conda-forge
...
torchaudio 2.0.0 pypi_0 pypi
torchvision 0.15.0 pypi_0 pypi
Sometimes you need to look for an installed package inside a specified environment. This can be achieved by using the command list followed by the $PACKAGENAME:
conda list $PACKAGENAME
If you don't remember the $PACKAGENAME conda introduces some regex-like search. For example if you remember only the first letters of $PACKAGENAME you can use list command followed by hyphen and the first letters. For example you look for sci packages:
conda list ^sci
The output will be a list of packages with: name, version, build and channel (where the repository come from).
Name Version Build Channel
scikit-image 0.19.3 py39hd77b12b_1
scikit-learn 1.2.1 py39hd77b12b_0
scikit-learn-intelex 2023.0.2 py39haa95532_0
scipy 1.10.0 py39h321e85e_1
If you remember only part of packagename, does not matter if the beginning, the middle or the end of $PACKAGENAME, just type what you remember after list command.
The following command will show all the packages with name torch inside:
Name Version Build Channel
pytorch 1.13.1 py3.9_cpu_0 pytorch
pytorch-mutex 1.0 cpu pytorch
torchaudio 0.13.1 py39_cpu pytorch
torchvision 0.14.1 py39_cpu pytorch
Also, multiple $PACKAGENAME search can be applied by concatenating names or part of the name with a pipe |. The following command will show all packages that contains sci and torch:
conda list "(sci|num)"
The output will be a combination of the two search:
Name Version Build Channel
pytorch 1.13.1 py3.9_cpu_0 pytorch
pytorch-mutex 1.0 cpu pytorch
scikit-image 0.19.3 py39hd77b12b_1
scikit-learn 1.2.1 py39hd77b12b_0
scikit-learn-intelex 2023.0.2 py39haa95532_0
scipy 1.10.0 py39h321e85e_1
torchaudio 0.13.1 py39_cpu pytorch
torchvision 0.14.1 py39_cpu pytorch
To list all packages to update inside an environment please use the search command:
conda search --outdated
The same apples to search for all pip installed packages:
pip list --outdated
If you want to delete an environment you can use the remove command as the following:
conda remove -n $ENVNAME --all
By applying --all you will remove all the packages related to the environment.
To rename an environment just use the command rename followed by the environment you want to rename ($ENVNAME) and the name ($NEW_ENVNAME) to apply to it:
conda rename -n $ENVNAME $NEW_ENVNAME
To make a clean uninstallation of the Anaconda distribution you must have installed the anaconda-clean package inside the base environment:
conda install anaconda-clean
Then, run the command:
anaconda-clean --yes
This command will create a backup of all files and directories that might be removed in a folder named .anaconda_backup.
At the end of the process you can delete all the Anaconda-related folders, for envs (anaconda3\envs) and packages (anaconda3\pkgs).