Installation

Instructions are for Linux and Mac. Windows coming soon.

Todo

Make package installable via pip or conda to make the process simpler.

  1. Clone repository with the command.

using https:

$ git clone https://github.com/CPOMUCL/GPSat.git

or ssh:

$ git clone git@github.com:CPOMUCL/GPSat.git

  1. Create a virtual environment.

change directory, create virtual environment, activate virtual environment.

$ cd GPSat
$ python -m venv venv
$ source venv/bin/activate

  1. Install Packages.

$ pip install -r requirements.txt

  1. (Optional) Install GPSat in editable model

Changes made to source code will be reflected immediately, useful for development.

$ pip install -e ./

  1. (Mac Specific) Install HDF5

Requires homebrew.

$ brew install hdf5

Export paths. To make permanent add to ~/.zshrc or ~/.bashrc file.:

Apple Silicon

$ export HDF5_DIR=/opt/homebrew/opt/hdf5

Intel

$ export HDF5_DIR=/usr/local

Note

If using Mac with M1 chip, need to install appropriate tensorflow version (see tensorflow-metal). This worked: SYSTEM_VERSION_COMPAT=0 python -m pip install tensorflow-macos. Also found incompatibility of numba and cartopy. Resolved by install cartopy first and then numba. Perhaps reverse order of installation? In addition, there seems to be a conflict with mac tensorflow and cartopy. How to resolve this? Solution: conda install matplotlib==3.2.2.