Bug/pd np version

[jamesETsmith]

Description

I noticed in some local testing that the following tests were failing:

• test_OnDiskDataset_heterogeneous[csv-True-True]
• test_OnDiskDataset_heterogeneous[csv-True-False]
• test_OnDiskDataset_heterogeneous[csv-False-True]
• test_OnDiskDataset_heterogeneous[csv-False-False]
• test_OnDiskDataset_preprocess_graph_with_single_type
• test_record_stream_graph_negative

Details

The OnDiskDataset issues are caused by behavior in the pandas > 3 releases (which explains why they didn't show up in the last release of DGL. You can't modify values for certain things in place, so we just have to tweak two lines:

# Before (in-place, fails on read-only arrays):
src += node_type_offset[node_type_to_id[src_type]]
dst += node_type_offset[node_type_to_id[dst_type]]

# After (out-of-place, works regardless of read-only flag):
src = src + node_type_offset[node_type_to_id[src_type]]
dst = dst + node_type_offset[node_type_to_id[dst_type]]

The test_record_stream_graph_negative test, I'm less sure about and it seems like it's relying on the underlying allocator to reuse blocks of memory, which doesn't seem to happen for us in my local testing. I'm testing with the same dependencies we supported in the last release so I'm not sure that anything changed on the rocm side. It seems like a potentially flaky test, and I'm not sure it's worth our time to dig into. @GMNGeoffrey what do you think?

Test results

Quick summary of the test results:

571:[DGL-ROCm-CI/Build+Test]   | ============================= test session starts ==============================
580:[DGL-ROCm-CI/Build+Test]   | ============================ slowest 100 durations =============================
584:[DGL-ROCm-CI/Build+Test]   | ========================= 1 passed, 1 warning in 3.09s =========================
585:[DGL-ROCm-CI/Build+Test]   | ============================= test session starts ==============================
677:[DGL-ROCm-CI/Build+Test]   | ============================ slowest 100 durations =============================
778:[DGL-ROCm-CI/Build+Test]   | ================ 2729 passed, 205 skipped, 7 warnings in 37.17s ================
779:[DGL-ROCm-CI/Build+Test]   | ============================= test session starts ==============================
960:[DGL-ROCm-CI/Build+Test]   | ============================ slowest 100 durations =============================
1061:[DGL-ROCm-CI/Build+Test]   | ========= 7452 passed, 653 skipped, 5477 warnings in 817.97s (0:13:37) =========

Checklist

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [$CATEGORY] (such as [NN], [Model], [Doc], [Feature]])
• I've leverage the tools to beautify the python and c++ code.
• The PR is complete and small, read the Google eng practice (CL equals to PR) to understand more about small PR. In DGL, we consider PRs with less than 200 lines of core code change are small (example, test and documentation could be exempted).
• All changes have test coverage
• Code is well-documented
• To the best of my knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
• Related issue is referred in this PR
• If the PR is for a new model/paper, I've updated the example index here.

[jamesETsmith]

Just a nudge on this @GMNGeoffrey and @gcapodagAMD

[GMNGeoffrey]

Sorry, I missed this before

[GMNGeoffrey]

Is this related?

[jamesETsmith]

Strangely enough, git was missing and installing the graphbolt deps would crash because hipcollections would pull in it's own deps like libhipcxx with git and crash. I'm not sure why this wasn't popping up before. Maybe I didn't catch it with #16?

[GMNGeoffrey]

Yeah this looks like it's verifying that a bug exists or something... I agree it's probably not worth trying to replicate it, although it's not clear why it wasn't failing before...

[jamesETsmith]

Agreed about it not being clear. Something smells fishy, but I think it's small enough that we can skip