-
Notifications
You must be signed in to change notification settings - Fork 307
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
AIX360 conda version #145
Comments
@Marleen1 The current setup.py attempts to install all packages required by all algorithms, so that users can essentially run any algorithm after installation. However each explainability algorithm requires only a subset of packages. Try installing the toolkit via pip. For packages that you are unable to install, you can try two options: (a) comment them out of setup.py (b) try installing those packages via conda and then do a regular pip install. Moving forward, we hope to update setup.py with install_requires and extras_require options so that installation is lighter and users optionally install the heavier packages (tensorflow, pytorch, cvxpy, etc.) based on the algorithm they wish to use. Any contributions/pull requests in this regard are welcome. We also hope to actively maintain a conda version in future, but the current option to install the latest github code is via pip. |
I'm unable to recreate the installation errors mentioned in #141 or #139 on mac. On mac (Big Sur 11.6), these instructions work fine. I faced a known minor error with certifi, which was rectified via Pls let us know if you in case you face any errors while installing the latest git code. |
Which version of Python are you using? It's possible that it is installable on Python 3.6 but Python 3.6 has already reached end of life https://devguide.python.org/devcycle/#end-of-life-branches --it is not even getting security fixes since December 2021. The reality is that aix360 install is completely broken on all Python versions that are currently maintained. This is an important issue for me because I am building a library that depends on My solution has been to fork aix360, comment out the unneeded dependencies (including the killer one |
@mmdanziger Thanks for sharing your requirements. Which algorithms from the toolkit are using for your work? You can see: https://github.com/Trusted-AI/AIX360/blob/master/.github/workflows/Build.yml to track which systems we build and test aix360 on. |
We use I'd really like to push my new library to pypi with a dependency on aix360. Do you have any plans to make any subset of aix360 installable via pypi on any versions of python that are currently supported? |
Thanks, yes we plan to make the installation lighter with some algorithms by default and others as optional. This will increase the python version coverage as well to > 3.6. I can't promise a date at this point, but likely soon. |
In addition to 3.6 being dead, another reason to urgently upgrade the python version to >= 3.8 is that there is no support for conda envirionments for < 3.8 on arm64. Considering that all new macs are arm64, this is pretty serious. |
@vijay-arya The instruction to use |
regarding the conda command mentioned by @rootsmusic (conda create -n aix360_env -c https://public.dhe.ibm.com/ibmdl/export/pub/software/server/ibm-ai/conda/ -c defaults -c powerai -c conda-forge aix360). Could someone provide more information about the expected behavior of this command or any troubleshooting steps for conda installation issues? |
As others reported as well (see issue #141 and #139) I cannot install AIX360 through github, but I can install it through conda (see https://anaconda.org/powerai/aix360). This version however is from more than two years ago. Can the version of AIX360 in conda be updated?
The text was updated successfully, but these errors were encountered: