Open rfl-urbaniak opened 4 months ago
Have been running into this issue apparently.
still issues with what isort does between CI and locally, despite the version numbers being the same. Other people have faced this, I'm looking for a solution, but slowly considering using black w/o isort at least till switching to isort 6.0
All csv data are now in two
.db
files for the two levels of analysis (counties and msa). Prior to deployment of dbs to the polis server, these live locally. As the db files are now too large to store on GitHub, the user needs to runcsv_to_db_pipeline.py
before the first use to generate the db locally.The original
DataGrabber
classes have been refactored, renamed to...CSV
and decoupled from use dowstream.The new
DataGrabberDB
class has been introduced and passed on to function as genericDataGrabber
andMSADataGrabber
.Additional tests for
DataGrabberDB
have been introduced intest_data_grabber_sql
. Additionally,DataGrabberDB
under the generic alias passes all the tests that the originalDataGrabber
did.generate_sql.ipynb
(docs/experimental
) contains performance tests for both approaches. At least in the current setting the orignal method is faster. The main culprit seems to be:This is not too surprising, after some reflection, as illustrated by this comment from ChatGPT:
As the ultimate tests of the switch to DB would involve data updates and model retraining, I leave the original
.csv
files and classes until those events. Keep in mind they are now not needed for queries to work correctly (they are needed to generate the.db
files and for some of the tests).The new
pytest
release leads to incompatibilities that might be worth investigating later. For now, fixed thepytest
version to be7.4.3.
insetup.py
.Some cleaning scripts have been moved to a subfolder, which required a small refactoring of import statements in generic data cleaning pipeline scripts.
Incorrect indentation in a
DataGrabber
test has been fixed.It turns out that
isort
with --profile black on the runner still works as if without this profile. Checked versions between local install and the one on the runner, the version numbers are the same. More people have similar issues, I suspended isort and decided to trust black, at least till stableisort 6.0
gets out.Inference tests succeed with
pyro-ppl==1.8.5
fail withpyro-ppl==1.9
. For now fixed the version number insetup.py
, but will think about investigating this deeper.