📦 1. Basic Installation & Upgrades

The bread and butter of setting up your experiment.

• pip install <package>: The standard install.

• pip install <pkg1> <pkg2>: Install multiple libraries (e.g., pip install torch torchvision).

• pip install -U pip: Upgrades pip itself (essential for support of newer .whl formats).

• pip install <package> --upgrade: Forces an upgrade of an existing library to the latest version.

• pip install <package>==1.13.1: Installs a specific version. Crucial for matching PyTorch versions to your CUDA drivers.

• pip install <package>~=2.1.0: Installs a version compatible with 2.1.0 (allows patches but not breaking changes).

🛠️ 2. Research & Development Mode

Commands for when you are building your own library or using bleeding-edge research code.

• pip install -e .: Editable mode. Links the current folder to your environment. If you change your code, the "installed" version updates instantly without re-installing.

• pip install git+https://github.com/user/repo.git: Installs a library directly from a GitHub repository (useful for papers that haven't hit PyPI yet).

• pip install "package[extra]": Installs optional dependencies (e.g., pip install "ray[tune]" or pip install "pandas[excel]").

• pip install --pre <package>: Allows installation of pre-release or "alpha" versions (e.g., trying a new Beta of TensorFlow).

📋 3. Environment Replication & Auditing

Critical for moving from your local machine to a high-performance computing (HPC) cluster.

• pip freeze > requirements.txt: Generates a "snapshot" of every library and version in your environment.

• pip install -r requirements.txt: Installs every library listed in a file—the standard way to share ML projects.

• pip list: Shows all installed packages and their versions.

• pip list --outdated: Identifies which of your ML libraries have newer versions available.

• pip show <package>: Displays metadata (location, dependencies, license). Use this to find exactly where your site-packages are stored.

• pip check: Verifies if installed packages have compatible dependencies (great for debugging "DLL not found" errors).

🧹 4. Cleaning & Management

Managing the clutter of failed experiments.

• pip uninstall <package> -y: Uninstalls a package without asking for "Yes/No" confirmation.

• pip cache purge: Clears the internal cache. If a download was corrupted (common with large torch binaries), run this.

• pip uninstall -r requirements.txt -y: Mass-uninstalls everything listed in your requirements file.

🏎️ 5. Advanced Configuration (The "Pro" Commands)

• pip install <package> --no-cache-dir: Forces a fresh download. Useful if you're low on disk space or suspect a corrupted cache.

• pip install <package> -i https://pypi.tuna.tsinghua.edu.cn/simple: Uses a mirror index. If the official PyPI is slow or blocked, this can be 10x faster.

• pip install --no-index --find-links=/path/to/wheels <package>: Installs from local files only. Essential for servers that do not have internet access for security reasons.

• python -m pip <command>: The safest way to run pip. It ensures you are using the pip associated with the exact Python version you are currently running.