Introduces a new feature, gaussian mixture modeling. The algorithm takes some initial labeling and uses a mixture of gaussians to approximate each classes distribution in the feature space. Each unlabled element is then asigned a label based on which classes distibution it fits best.

This is implemented as a c++/cuda extention, but currently it is only implemented in cuda. It may be run with any spatial dimension, channel count, class count, and mixture component count (number of gaussians in each classes mixture).

This PR also introduces JIT compilation for c++/cuda extentions. The source code for this extention is not compiled when monai is installed. When an end user comes to create a GMM python object for the first time, the code will be compiled. This allows us to take parameters passed to the GMM python object and use these as compile time constants in the build allowing the compiler to provide more optimized binaries. A new build is created and stored for each configuration the user asks for. Once a configuration has been built, the JIT system will provide this binary for all subsequent requests.

see #1601

Status

Ready

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.