如何借助Anaconda依赖安装PyPI包？安装pip包时优先复用conda依赖

Great question! Managing packages across conda and pip can feel tricky at first, but there are straightforward workflows to handle both your requests. Let’s break this down step by step.

First and foremost, always work within an activated conda environment to avoid cluttering your base setup. Here’s how to install PyPI packages properly:

• Activate your target environment:
  conda activate your_env_name
• Once activated, use the environment’s built-in pip (not your system pip!) to install the PyPI package:
  pip install your_desired_package
• To keep your environment reproducible, export it after installation—this will include both conda and pip-installed packages:
  conda env export > environment.yml

The key here is to prevent pip from overwriting or installing duplicate versions of dependencies that are already available via conda. Here are the most reliable methods:

1. Manual Dependency Management (Simplest Approach) #

• First, install all possible dependencies for your target PyPI package using conda. You can check if a dependency exists on conda with:
  conda search dependency_name
• Once all conda-compatible dependencies are installed, use pip with the --no-deps flag to install the PyPI package. This tells pip to skip installing dependencies entirely and rely on your existing conda-managed versions:
  pip install --no-deps your_desired_package

2. Automated Locked Environments (Best for Reproducibility) #

If you want a hands-off way to ensure conda dependencies are prioritized, use conda-lock to resolve and lock all dependencies (both conda and pip):

• Install conda-lock in your base environment or target environment:
  conda install -c conda-forge conda-lock
• Create an environment.yml file that lists your conda packages and includes the PyPI package under a pip section:

  name: your_env_name
  channels:
    - conda-forge
    - defaults
  dependencies:
    - python=3.10
    - numpy
    - pandas
    - pip:
      - your_desired_pypi_package
  

• Generate a lock file that resolves all dependencies with conda prioritized:
  conda-lock lock -f environment.yml
• Create or update your environment from the lock file:
  conda-lock install -n your_env_name conda-lock.yml

Critical Note #

Always make sure you’re using the pip that’s installed inside your conda environment (not your system-wide pip). Verify this with:
which pip
The output should point to a path inside your conda environment’s bin directory. If not, reactivate your environment or install pip via conda: conda install pip

内容的提问来源于stack exchange，提问作者Gere