Hi folks,
I posted this as a comment on #2172, but I wanted to note it here. Right now the holdup for merging yt-4 into master is the answer testing. We actually would not expect the answers to be the same between the two (at the very least because the ordering of values will be different) so we need to do some kind of manually inspection.
My plan for addressing answer testing differences, which I will start by doing manually:
- All grid and octree data should have either the same, or justifiably different answers. In my view, "justifiably different" includes unit changes, and also includes changes as a result of particles being selected on edges of grid boundaries. These will be documented. (See #2195 )
- Results for all data selections in SPH and particle datasets should be identical in count and values, although the order will likely be different. To address this, we will have a "sorting order" field that is the morton index of the particles, and the values will be sorted and compared. This should help to identify situations where different numbers of particles are being selected (typically they should not be, except for situations related to smoothing length.)
- Manual inspection of visualizations that avoid meshing in 4.0 and that utilize meshing in 3.x.
I believe this should cover most of the cases, and will take to the next step of verification. To conduct these tests, I'm going to work on a script that outputs the appropriate values into an HDF5 file, and then compare the results for both. This will be somewhat distinct from the answer testing and designed for ease of exploration.
Once I have a system prepared for this, I will post back here, and I will likely put the results online to view.
-Matt
This sounds like a good plan, thank you for being conscientious about changing answers.
On Mon, Oct 7, 2019 at 10:55 AM Matthew Turk matthewturk@gmail.com wrote:
Hi folks,
I posted this as a comment on #2172, but I wanted to note it here. Right now the holdup for merging yt-4 into master is the answer testing. We actually would not expect the answers to be the same between the two (at the very least because the ordering of values will be different) so we need to do some kind of manually inspection.
My plan for addressing answer testing differences, which I will start by doing manually:
- All grid and octree data should have either the same, or justifiably different answers. In my view, "justifiably different" includes unit changes, and also includes changes as a result of particles being selected on edges of grid boundaries. These will be documented. (See #2195 )
- Results for all data selections in SPH and particle datasets should be identical in count and values, although the order will likely be different. To address this, we will have a "sorting order" field that is the morton index of the particles, and the values will be sorted and compared. This should help to identify situations where different numbers of particles are being selected (typically they should not be, except for situations related to smoothing length.)
- Manual inspection of visualizations that avoid meshing in 4.0 and that utilize meshing in 3.x.
I believe this should cover most of the cases, and will take to the next step of verification. To conduct these tests, I'm going to work on a script that outputs the appropriate values into an HDF5 file, and then compare the results for both. This will be somewhat distinct from the answer testing and designed for ease of exploration.
Once I have a system prepared for this, I will post back here, and I will likely put the results online to view.
-Matt
yt-dev mailing list -- yt-dev@python.org To unsubscribe send an email to yt-dev-leave@python.org
-
Matthew Turk -
Nathan