NVIDIA / NVIDIA/cudf

Remove redundant tests

Open
#19,695 0 comments 0 reactions 0 assignees View on GitHub
improvement Python tests
Dominant language
C++
Stars
9.8k
Forks
1.1k
Avg merge
3d 6m
Merged PRs (30d)
278

Description

The cudf classic test suite is largely built on comparing results to pandas. Now that we have cudf.pandas, we are effectively doubling this work by running the entire pandas test suite with cudf.pandas enabled. We should consider how many of these tests we could conceivably either remove altogether or upstream to pandas itself. For the foreseeable future we intend to maintain cudf has having the ability to run separately and with pandas compatibility mode turned off to allow for more efficient execution in some scenarios (e.g. not sorting join results), so we will likely need to retain some subset of our tests. Removing duplicates and contributing others upstream is likely beneficial to everyone involved though since our test suite exercises many edge cases that the pandas test suite does not. The big challenge here will be that since current runs of the pandas test suite are significantly longer than running the cudf test suite (~50 min vs ~20 min) we risk dramatically increasing our CI times if we use the pandas test suite as our primary one. We could consider doing some parsimonious subselection of tests on different CI jobs if we wished to avoid that, but it's also likely premature to be overthinking this since #18659 and #19693 will probably have a significant impact on the total runtime of the pandas test suite with cudf.pandas (the former will increase it while the latter will reduce it). We can revisit this topic once we've made more progress there. In the interim, though, removing redundant tests and upstreaming others is a worthwhile endeavor for us to start now.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.