Pandas 2.0 update

[jeffkinnison]

Updates to make Ludwig compatible with Pandas 2.0. The current changes include:

• Remove experiment_impact_tracker as a dependency
• Update LudwigProfiler to replicate experiment_impact_tracker CPU/GPU attribute collection
• df.iteritems() -> df.items()
• df.append(df2) -> df_engine.concat([df, df2])
• Support for writing xls files was removed. This only seems to impact one integration test.

[github-actions]

Unit Test Results

  6 files  +  2    6 suites  +2   1h 1m 56s ⏱️ + 22m 14s
33 tests ±  0  30 ✔️ ±  0  3 💤 ±0  0 ❌ ±0 
99 runs  +33  90 ✔️ +30  9 💤 +3  0 ❌ ±0 

Results for commit b04c261. ± Comparison against base commit d68d96e.

♻️ This comment has been updated with latest results.

[justinxzhao]

Sweet! Just one comment, otherwise LGTM.

[justinxzhao]

nit: Remove empty docs.

If this is copied from the other repo, please reference it as a comment i.e.
Adapted from <link>, with additional adjustments for compatibility with Pandas 2.0.

[justinxzhao]

Nice cleanup.

[jeffkinnison]

The changes listed above should cover the 2.0 update, and as far as I can tell we should still be able to support 1.x because df.items and pd.concat have been in the API since at least 1.0.

requirements.txt does allow for versions all the way back to 0.25, though, and concat doesn't exist that far back. Do we want to bump the minimum version to 1.0?

@justinxzhao

[justinxzhao]

The changes listed above should cover the 2.0 update, and as far as I can tell we should still be able to support 1.x because df.items and pd.concat have been in the API since at least 1.0.

requirements.txt does allow for versions all the way back to 0.25, though, and concat doesn't exist that far back. Do we want to bump the minimum version to 1.0?

@justinxzhao

Bumping the min version to 1.0 SGTM!