DeepHealth Conda Builds

Conda recipes for DeepHealth software.

Installation

DeepHealth Conda packages come in three flavors:

• *-cpu: CPU-only
• *-gpu: GPU-enabled
• *-cudnn: GPU-enabled, with cuDNN support

The GPU-enabled packages support CUDA 11.3.

Note that ECVL/PyECVL does not actually offer cuDNN support. The cudnn tag in this case simply means that the package pulls the corresponding eddl-cudnn and/or pyeddl-cudnn dependency.

Configuring channels

Before installing, run the following configuration commands (you can omit the bioconda channel if you only want to install EDDL/PyEDDL):

conda config --add channels dhealth
conda config --add channels bioconda
conda config --add channels conda-forge
conda config --set channel_priority strict

Make sure you add the channels in the order shown above. Since --add adds the channel to the beginning of the list, the channels section in your configuration file (conda config --show) should now look like this:

channel_priority: strict
channels:
  - conda-forge
  - bioconda
  - dhealth
  - defaults

Package dependency

The DeepHealth Toolkit consists of two main C++ libraries: EDDL and ECVL. Python bindings are also available for both libraries: PyEDDL and PyECVL. The dependency graph is shown below:

      +--------+
      | PyECVL |
      +--------+
       ^      ^
       |      |
+------+-+  +-+------+
|  ECVL  |  | PyEDDL |
+--------+  +--------+
        ^    ^
        |    |
      +-+----+-+
      |  EDDL  |
      +--------+

For instance, if you install PyEDDL, you will also pull EDDL as a dependency, while if you install PyECVL you will install all four.

The Conda packages, available from the dhealth channel, are named according to a simple <library>-<target> scheme (e.g., pyeddl-gpu). Additionally, most of them (all except eddl) are compiled for a specific Python version. Currently, packages are available for Python 3.6, 3.7 and 3.8. Which one will be pulled depends on the Python version installed in your environment.

Example: install Python 3.7 and PyECVL compiled for GPU

conda config --add channels dhealth
conda config --add channels bioconda
conda config --add channels conda-forge
conda config --set channel_priority strict
conda create -y -n dh_toolkit
conda activate dh_toolkit
conda install -y python=3.7 pyecvl-gpu

Speeding up the installation

When popular channels such as conda-forge and bioconda are involved, installing packages can take a considerable amount of time. One of the easiest way to get a huge speedup is to use Mamba. The mamba command can be used as a faster drop-in replacement for conda. For instance:

mamba install -y python=3.7 eddl-cpu

Note on version tags

In some cases, the upstream version tag has been slightly altered to comply with the PEP 440 scheme. For instance, the Conda package for EDDL v0.8.3a has a version tag of 0.8.3a0.