[ISSUE 534]: Deprecate conda environment.yml / move to pip

[HundredBillion]

This PR resolves Issue 534.

Since PyTorch is no longer shipping conda packages, this PR removes the conda environment files and switches the installation docs to a standard Python pip/venv-based setup.

For GPU installs, PyTorch's commands depend on the user's OS and CUDA version, so the docs point to the official PyTorch install page rather than hardcoding a command:

https://pytorch.org/get-started/locally/

For CPU-only and Apple Silicon users, torch is already declared in [project.dependencies], so it is installed automatically by pip install neural-amp-modeler — no separate PyTorch install step is needed.

Summary of changes

• Removed environments/environment_cpu_apple.yml and environments/environment_gpu.yml
• Rewrote docs/source/installation.rst to use a 4-step pip/venv install flow with OS-specific venv activation commands (macOS/Linux, Windows cmd.exe, Windows PowerShell)
• Added a [dev] optional-dependencies group to pyproject.toml (black, flake8, pre-commit) so the full dev environment installs with a single command: pip install -e ".[dev,test]"
• Removed the redundant non-GPU pip install torch step from the docs, since torch is pulled in transitively by pip install neural-amp-modeler
• Removed dead conda code from tests/test_graceful_shutdown.py (--conda-env CLI flag and related branching logic that was never exercised by the pytest path)

[sdatkinson]

Thanks for the PR, @HundredBillion!

This doesn't resolve #534 because installing still uses conda .yml files.

I'll give comments inline presently

[HundredBillion]

@sdatkinson - I noticed your PR to change the url for installing Pytorch.

I updated this PR in case you still want to use update the docs to install via requirements-gpu.txt

[sdatkinson]

I'm sorry, but this still isn't what I want. There are yml files still in the repo. Moving to pip should see them gone.

[HundredBillion]

I'm sorry, but this still isn't what I want. There are yml files still in the repo. Moving to pip should see them gone.

Got it — I misunderstood the issue. I treated it as “use pip inside the conda env files,” but what you want is to remove the conda env approach entirely. I’ll update the PR to delete the environment .yml files and switch the installation docs to a pip/venv-based flow.

[Copilot]

Pull request overview

This PR updates the project’s installation guidance to move away from conda-based environment setup and toward pip/venv-based installation, addressing Issue #534 about PyTorch no longer being available via conda.

Changes:

• Removed conda environment definitions for CPU (Apple) and GPU setups.
• Rewrote the installation docs to use Python + venv + python -m pip.
• Updated local development setup instructions to use editable installs and manual dev tooling installation.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
environments/environment_gpu.yml	Removes the GPU conda environment definition.
environments/environment_cpu_apple.yml	Removes the Apple/CPU conda environment definition.
docs/source/installation.rst	Updates installation and development instructions to use venv + pip and points users to PyTorch’s install selector.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rhaist]

Hi @HundredBillion — random passer-by here, also interested in seeing #534 land. A few suggestions if useful, take or leave:

1. [dev] extra in pyproject.toml — I think you mentioned wanting to pick up Copilot's nit about this. Defining the dev tools as an extra keeps the docs from drifting from the project's declared deps, and the local-dev install collapses to one line:

   [project.optional-dependencies]
   dev = [
       "black",
       "flake8",
       "pre-commit",
   ]

   $ pip install -e ".[dev,test]"
   

2. The non-GPU python -m pip install torch step might be removable. torch is already in dependencies in pyproject.toml, so pip install neural-amp-modeler pulls in the CPU build transitively. That'd make the default-case install genuinely one command, with the PyTorch-site link doing the work for GPU users (which matches what @sdatkinson asked for earlier).

3. One stray conda reference outside the env files: tests/test_graceful_shutdown.py has a --conda-env CLI flag and a conda_env kwarg on run_training_with_interrupt. The pytest path already hardcodes None; only the standalone runner uses it. Might be worth removing in the same PR for consistency, but totally a judgement call.

Tried all three locally on top of main (pytest run; same set of unrelated pre-existing failures as baseline, nothing new). Happy to open a follow-up PR if any of this is easier that way.

[HundredBillion]

@rhaist - Thanks for your review. I applied all 3 changes you suggested because I think they are all in scope for this PR.
Check out my changes and let me know if it lines up with what you did.