We use the existing tests and run them against the same server as the R tests.
Since this series of tests runs with every check-in, I have removed a few redundant and extremely slow tests. A few tests were originally there to try installing popular ML packages, but most of these ML packages are extremely large and have lots of dependencies. For CI, it is better to test the functionality, so I have trimmed to basic tests.
There are also 2 bug fixes in this PR:
pip moved away from holding pep425tags in their codebase in pip version 20+, so we just import from wheel instead.
The way we resolved requirement specs was not the most robust. We can just use pkg_resources's Requirement class functionality instead of writing our own. In particular, our original implementation did not work with "~=".
This PR adds Python to the Travis-CI pipeline.
We use the existing tests and run them against the same server as the R tests.
Since this series of tests runs with every check-in, I have removed a few redundant and extremely slow tests. A few tests were originally there to try installing popular ML packages, but most of these ML packages are extremely large and have lots of dependencies. For CI, it is better to test the functionality, so I have trimmed to basic tests.
There are also 2 bug fixes in this PR: