Conda install of [neuralforecast] is broken

[twobitunicorn]

What happened + What you expected to happen

Doing a straight up install on a python=3.11.* conda environment gives the following error. This is from executing the code conda install -c conda-forge neuralforecast.

Same ticket just a bit older. https://github.com/Nixtla/neuralforecast/issues/440

Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): - WARNING conda.models.version:get_matcher(546): Using .* with relational operator is superfluous and deprecated and will be removed in a future version of conda. Your spec was 1.8.0.*, but conda is ignoring the .* and treating it as 1.8.0
WARNING conda.models.version:get_matcher(546): Using .* with relational operator is superfluous and deprecated and will be removed in a future version of conda. Your spec was 1.9.0.*, but conda is ignoring the .* and treating it as 1.9.0
done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: |
Found conflicts! Looking for incompatible packages.
This can take several minutes.  Press CTRL-C to abort.
failed

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

Output in format: Requested package -> Available versionsThe following specifications were found to be incompatible with your system:

  - feature:/osx-arm64::__osx==14.0=0
  - neuralforecast -> numba -> __osx[version='>=10.9|>=11.0']
  - python=3.11.6 -> ncurses[version='>=6.4,<7.0a0'] -> __osx[version='>=10.9']

Your installed version is: 14.0

Note that strict channel priority may have removed packages required for satisfiability.c

Versions / Dependencies

conda

Reproduction script

conda create --name nixtla python=3.11.6 conda install -c conda-forge neuralforecast

Issue Severity

Low: It annoys or frustrates me.

[jmoralez]

Hey. Can you try the following:

• Merging the installs: conda create -n nixtla -c conda-forge python=3.11.6 neuralforecast
• Adding other channels: conda create -n nixtla -c conda-forge -c pytorch -c defaults python=3.11.6 neuralforecast
• Using mamba: conda install -n base -y mamba && mamba create -n nixtla -c conda-forge python=3.11.6 neuralforecast

[twobitunicorn]

@jmoralez Yes. Ran ran them on my macbook pro (arm, m2). They each throw the same error. See attached. If I remove neuralforecast and install only python=3.11.6 (and/or pytorch) and then run pip install neuralforecast it works just fine.

out.log

[jmoralez]

Are you able to install just numba?

[twobitunicorn]

Yes. Running conda create -n nixtla -c conda-forge python=3.11.6 numba installs namba=0.58.1 no problems.

[jmoralez]

And installing neuralforecast after?

[twobitunicorn]

Same issue. I have been looking over the feedstock and it seems that the script for osx was taken out .scripts/run_osx_build.sh. This is still referenced in build-locally.py. The verify config only seems to support linux now as that is all it gets from the ci support directory .ci_support/*.yaml.

[jmoralez]

The file is noarch, so it has to be a problem in one of the dependencies.

[jmoralez]

Neither of us has an arm64 mac, so we can't debug. We appreciate you trying things out.

[jmoralez]

BTW in the logs above numba was the problem. Can you post the error you get after installing numba and trying to install neural?

[twobitunicorn]

your welcome, appreciate all the work you all have done!
Uploading out.log…

It is the same issue. Attached the output.

[twobitunicorn]

I played with --no-deps and --only-deps. Same issue with both of those.

For funsies I went through each of the deps (environment.yml) and did an individual conda/pip install on each one and had no issues.

[twobitunicorn]

Jumped on my other machine. Running mamba install -c conda-forge neuralforecast --only-deps gives the following error below. I can install pyppmd with either conda or pip. ray-tune is constrained by <3.11.0a0 on conda/mamba but pip install ray[tune] installs no problem. Executing pip install neuralforecast (with or without a pip install of pyppmd and ray-tune) installs just fine.

Could not solve for environment specs
The following packages are incompatible
└─ neuralforecast   is installable with the potential options
   ├─ neuralforecast [0.0.7|0.0.8|...|1.6.0] would require
   │  └─ pyppmd >=0.17.0,<0.18.0 , which does not exist (perhaps a missing channel);
   └─ neuralforecast [1.6.0|1.6.1|1.6.2|1.6.3|1.6.4] would require
      └─ ray-tune >=2.2.0  with the potential options
         ├─ ray-tune 2.3.0 would require
         │  └─ python >=3.10,<3.11.0a0 , which can be installed;
         ├─ ray-tune 2.3.0 would require
         │  └─ python >=3.8,<3.9.0a0 , which can be installed;
         └─ ray-tune 2.3.0 would require
            └─ python >=3.9,<3.10.0a0 , which can be installed.

[jmoralez]

For funsies I went through each of the deps (environment.yml) and did an individual conda/pip install on each one and had no issues.

Even ray-tune? It doesn't seem like they publish the conda package for osx-arm64 (files).

[twobitunicorn]

Apologies. That was the one I had to do the pip in conda/pip. pyppmd can be installed through conda.

[jmoralez]

After you install ray-tune from pip are you able to install neuralforecast from conda?

[twobitunicorn]

No. It doesn't seem to pick up the pip package.

[twobitunicorn]

I filed an issue with the conda feedstock maintainers for adding arm64 support for ray and ray-tune. https://github.com/conda-forge/ray-packages-feedstock/issues/134