Fingerprint Mismatch Question/Flag

[mpkeightley]

I just regenerated the TMD using WSL2 (see PR #526 for context) and TC 6.7.1 and have a question about the fingerprint mismatch results copied below: are these big differences to be expected/normal? I regenerated twice just to make sure.

I understand that the fingerprint was generated under 6.7.0 and there is some version drift to be expected. But some of the mismatches are very large and I don't know if that is expected/normal, so I am flagging it.

Statistic	Reference	My build	Rel. diff
tip_income.sum	3.84e8	5.32e8	38.5%
tip_income.std	60,701	134,938	122%
tip_income.max	9.70e6	3.77e7	288%

overtime_income and pencon_p/pencon_s show the same pattern:

Statistic	Reference	My build	Rel. diff
overtime_income.max	3.58e7	4.15e7	160%
overtime_income.std	200,449	242,678	211%
pencon_p.std	6,386.92	5,492.54	140%
pencon_s.std	4,875.49	4,375.30	103%

Please let me know if I can provide more information.

[martinholmer]

@mpkeightley said in TMD issue #532:

I just regenerated the TMD using WSL2 (see PR #526 for context) and TC 6.7.1 and have a question about the fingerprint mismatch results copied below:

Thanks for using Microsoft's WSL2: using this Linux environment should make it easier for you to use the TMD repository and it certainly will make it easier for us to help when you have questions.

In order to better help you, I need you to send me (either in a new comment in this issue or in a private email) the complete screen output generated by the make data command.

[mpkeightley]

@martinholmer - It looks like PR #533 resolved the issue. I get clean regenerations in WSL2 (and in Windows).

[martinholmer]

@mpkeightley said in issue #532:

It looks like PR #533 resolved the issue. I get clean regenerations in WSL2 (and in Windows).

Great! I'm glad to hear this.

The scikit-learn package (which TMD uses when imputing the pencon_* and tip/overtime/autoloan variables) was "upgraded" last month in ways that make it impossible to exactly reproduce the TMD input data files when using the new package version. So, we'll stick with the older version until we can more fully assess our options for using the new version.

If anything else comes up, please open a new issue.