If you're running Mac Big Sur and/or your Mac is running on an Apple Silicon chip), you must follow the instructions to install for Conda-forge. The instructions for Mac with Intel chip will not work.
For Conda-forge (Mac with Apple Silicon chip)
Koe can run on any operating system, as well as running as a Docker image. So to maximise compatibility, I
created different requirements.txt
for different environments.
requirements.txt
: contains all the libraries that are necessary to compile and run Koe, except the machine-learning partrequirements-basic.txt
: contains the extra libraries necessary for the machine-learning part of Koe to run on a tensorflow image (CPU-only)requirements-tensorflow.txt
: contains the extra libraries necessary for the machine-learning part of Koe oto run on a tensorflow image (GPU-enabled)requirements-conda-forge.txt
: contains the libraries that can be installed by pip in a conda environmentrequirements-dev.txt
: contains the extra libraries that are necessary to for development environment, mostly to run lint and profilerrequirements-production.txt
: contains the extra libraries that are necessary to deploy Koe as a webapp (using uWSGI)
Many python libraries cannot be installed using pip in conda-forge environment, they must be installed using conda install
.
The remaining libraries can be installed normally with pip. So for conda-forge the steps to install all python libraries are:
- First install all the special libraries compiled specifically for conda-forge using
conda install
command - Then, install the rest with
pip install requirements-conda-forge.txt
- Finally, install the dev tools with
pip install requirements-dev.txt
- First, install all necessary library with
pip install requirements.txt
- Finally, install the dev tools with
pip install requirements-dev.txt
This section is quite advanced and difficult to get right. Also might not be necessary as I'm maintaining an official Koe's Docker image on Docker Hub (https://hub.docker.com/repository/docker/crazyfffan/koe) So you shouldn't worry about this