provide pip compatible installation

[lcrmorin]

Description

Currently the documentation only provides a conda command for installation. The pip command is mentioned but the link doesn't lead to a pip command. As is it is not possible to test Khiops in non-conda environnements (typically, most of the online environments such as collab, kaggle, hf...).

[lcrmorin]

Nevermind, the windows command appears to work too. Maybe the pip commands should be default. Nowadays Anaconda is getting a bit out of fashion.

[lcrmorin]

Getting this error from the pip install (after pip installing 10.2.4):

KhiopsEnvironmentError: The 'khiops_env' script not found for the current 'binary+pip' installation method. Make sure you have installed khiops >= 10.2.3. Go to https://khiops.org/ for more information.

Not sure this is linked to above problem, but I reopen as this might be a general issue about installing and testing in collab like environment.

[lucaurelien]

Thank you for bringing this up and for your interest in Khiops! We truly appreciate feedbacks and are here to help clarify any points.

To provide some context:

1. We do provide a pip version to ensure that everyone can use Khiops, regardless of their preferred installation method or environment. However, pip is not the default or recommended method because it requires installing the Khiops binary (a C++ library requiring OS-level installation) and its MPI dependency separately. This approach is better suited for advanced users familiar with managing system-level dependencies. Additionally, it may not work on certain environments that restrict admin rights for standard users;
2. We recommend using conda to install Khiops, as it handles everything seamlessly, including the Khiops binary and its MPI dependency. This minimizes configuration issues and ensures that your environment is fully compatible. For optimal results, we also recommend creating and activating a dedicated conda environment before installation;
3. Google Colab environments are fully compatible with Khiops when using Conda;
4. Kaggle no longer supports Conda natively, but you can install Khiops in a Kaggle environment using the following steps:

# Install khiops-core apt package
! wget -O "khiops.deb" "https://github.com/KhiopsML/khiops/releases/download/10.2.4/khiops-core-openmpi_10.2.4-1-jammy.amd64.deb" 
! sudo dpkg -i "khiops.deb" || sudo apt-get -f -y install 

# Install the khiops python package
! pip install "https://github.com/KhiopsML/khiops-python/releases/download/10.2.4.0/khiops-10.2.4.0.tar.gz"

Thank you again for reaching out! We hope this clears things up, but please don’t hesitate to let us know if you have any further questions or need more assistance.

NB: installation via pip is not available on macOS yet (only conda).

[lcrmorin]

Thanks for clarifying all this. As you might have guessed I have both personal and professional use cases in mind. I can confirm this is working on Kaggle. For my professional environment (where not having admin rights is the norm), I'll try the Anaconda approach.