Dependency Management with Conda

:::info The Conda dependency management introduced here involves creating a completely independent environment, which typically means you need an environment with a different Python version than the system's default. If you don't have this requirement, there's no need to create an independent environment. Using pip install --user can fully accomplish this. Please refer to the documentation Installing Additional Dependencies Under the Default Python Version. :::

Python in HyperAI is managed through Conda. The default installation environment path can be obtained with the following command:

$ conda env list

# conda environments:
#
base                  *  /usr/local

You can see that the default environment is in /usr/local. The dependencies in each environment can be obtained through conda list. A complete list of installed dependencies is also provided under "Runtime Environment" on the left side of the documentation.

Creating a New Environment with Conda

1. Create a new environment under /openbayes/home

conda create -p open-mmlab python=3.9 -y

After installation is complete, activate the new environment with conda activate /openbayes/home/open-mmlab.

:::note The key to being able to save the environment is to store the new environment's save path under /openbayes/home. Through "Continue Execution", this content can be bound again to a new execution. :::

2. Install other dependencies according to the documentation

conda install pytorch torchvision -c pytorch

git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection

# Continue installation according to official documentation
pip install mmcv
pip install -r requirements/build.txt
pip install "git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI"
pip install -v -e .  # or "python setup.py develop"

After completing the installation, mmdetection-related dependencies are installed in /openbayes/home/open-mmlab (which is also under /output/open-mmlab).

3. Open the original execution through continue execution

Bind the mmdetection environment prepared in the previous execution to a new execution through "Continue Execution". The previously configured environment can be activated again and used by running the command conda activate /openbayes/home/open-mmlab/.

Integrating the Newly Created Conda with Jupyter Workspace

Jupyter workspace can integrate with Conda to allow different notebooks to specify different Conda environments. Follow these steps to add a custom Conda environment to your Jupyter workspace.

conda activate /openbayes/home/open-mmlab/
(/openbayes/home/open-mmlab/)$ conda install ipykernel
(/openbayes/home/open-mmlab/)$ python -m ipykernel install --user --name=open-mmlab --display-name="Python (open-mmlab)"

After reopening the Jupyter workspace page, you can see an additional option: