Wrong pandas version installed with modin-omnisci==0.10.2 from conda-forge

[anmyachev]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.04
• Modin version (modin.__version__): 0.10.2 (from conda-forge)
• Python version: 3.9.6
• Code we can use to reproduce:

conda install modin-omnisci -c conda-forge
python -c "import modin.pandas as pd"

Describe the problem

Wrong Pandas addiction in Modin receipt for conda.

Source code / logs

UserWarning: The pandas version installed 1.3.0 does not match the supported pandas version in Modin 1.3.2. This may cause undesired side effects!

[anmyachev]

@maksimbo1 can you take a look?

[maksimbo1]

Yeah, i've heard about this problem, i'll look into it.

[gshimansky]

This problem appears not just with modin-omnisci. Installing modin=0.10.2 from conda-forge leads to pandas-1.3.0 as well. Both modin and modin-omnisci get a wrong pandas version.

[gshimansky]

As a workaround to fix up broken conda environment with wrong pandas version it is possible to use this approach: caporaso-lab/pretrained-feature-classifiers#6 (comment)
Here is the contents of explicit file packages.txt that allows to install pandas=1.3.2 (for python 3.8.10 from conda-forge) into any environment with no dependencies checking:

@EXPLICIT
https://conda.anaconda.org/conda-forge/linux-64/pandas-1.3.2-py38h43a58ef_0.tar.bz2

Use at your own risk.

[gshimansky]

One more problem I found with current conda-forge recipe of modin=0.10.2 (not sure if it deserves a separate issue) is that modin no longer requres ray-core. This command conda create -n modin-0.10.2 -c conda-forge modin=0.10.2 creates an environment without any ray packages and modin defaults to using dask.

[maksimbo1]

Found problem with conda-forge recipe (thanks to @gshimansky and @aregm ), sended PR with fixed recipe: conda-forge/modin-feedstock#21.

about ray - package modin-ray still requires ray-core, which seems to be okay.

[h-vetinari]

This problem is easily solvable with conda install modin pandas=1.3.2. The current 0.10.2 builds are compatible (from the POV of conda-forge) with all pandas 1.3.

Whether modin should pin down to patch level at all is discussed in #3371

[gshimansky]

@h-vetinari this doesn't seem to work as you describe. Attempting to add pandas=1.3.2 leads to conflicts for me:

$ conda create -n gregory-modin-0.10.2 -c conda-forge modin-all=0.10.2 pandas=1.3.2
Collecting package metadata (current_repodata.json): done
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: -
Found conflicts! Looking for incompatible packages.                                                                   failed

UnsatisfiableError: The following specifications were found to be incompatible with each other:

Output in format: Requested package -> Available versions

Package pandas conflicts for:
pandas=1.3.2
modin-all=0.10.2 -> modin-core==0.10.2=py38h578d9bd_0 -> pandas==1.3

[h-vetinari]

Ah, this is annoying. Apparently conda-build and conda have different behaviours here (i.e. adding .*, resp. adding .0 apparently).

So yeah, the last build is then not working as I intended, sorry about that.

[devin-petersohn]

Can this be closed now? @anmyachev @maksimbo1

[maksimbo1]

@devin-petersohn looks like in conda-forge still old package, i reproduced this error right now.

[gshimansky]

Looks like new binaries have finally reached conda-forge repositories and modin-all=0.10.2 now requires pandas-1.3.2.

[devin-petersohn]

Thanks for checking @gshimansky! Closing as resolved. (Please reopen if necessary)