Fix COBRA install: add scikit-learn dependency + revamp install docs (#391)

[marouenbg]

Fixes #391.

Problem

A user following the documented install (git clone → cd netZooPy → pip3 install -e .) could not run COBRA.

netZooPy/cobra/cobra.py imports from sklearn.linear_model import LinearRegression, Lasso, but setup.py did not list scikit-learn in install_requires (it was only in requirements.txt). Because netZooPy/__init__.py eagerly runs from netZooPy import cobra, a clean pip install -e . left import netZooPy itself broken with ModuleNotFoundError: No module named 'sklearn', not just COBRA.

Separately, the reporter's No module named 'netZooPy.cobra' / (unknown location) errors come from an environment pitfall: running Python from the clone's parent directory (or a Jupyter/VS Code kernel that isn't the install env) makes the cloned netZooPy folder resolve as an empty namespace package that shadows the installed one.

Changes

• setup.py: add scikit-learn to install_requires (the substantive fix).
• README.md: keep it minimal — only adds from netZooPy.cobra import cobra to the usage example.
• docs/install/index.md: revamp the install guide —
  • sync the dependency list with setup.py (was missing h5py, joblib, statsmodels, scikit-learn, click, tables, torch) and note pip installs them automatically; document CuPy as the optional GPU dependency;
  • recommend a virtual environment;
  • expand the Troubleshooting section to cover both failure modes above.

Troubleshooting content lives in the install guide (where users look) rather than the README.

Verification

• Clean python -m venv + pip install -e . now auto-installs scikit-learn; import netZooPy and from netZooPy.cobra import cobra both succeed; cobra() runs in nnls and MLE modes.
• tests/test_cobra.py passes.
• Audited every submodule's module-load-time imports: scikit-learn was the only undeclared hard dependency. cupy is correctly optional (lazy, GPU-only); patsy is guaranteed via statsmodels.

🤖 Generated with Claude Code