Score tests #1847
Score tests #1847
Conversation
emma-nechamkin
requested review from
esfoobar-usds,
vim-usds and
mattbowen-usds
as code owners
| logger = get_module_logger(__name__) | ||
|
|
||
|
|
||
| # this is so slow? |
There was a problem hiding this comment.
This specific fixture takes a long time to load... any improvements?
There was a problem hiding this comment.
Once it's loaded once, will it change? If not, we could lru_cache it to at least only ever have to load it once per thread. The other option would be to write a faster to load file format, like feather --- but that has problems of its own, since it could mask problems with the CSV or create new ones.
There was a problem hiding this comment.
It does only get loaded a single time, so I think lru_cache will work. Do you happen to have an example of this handy? (if not, I am happy to google)
There was a problem hiding this comment.
I have a great example! See https://github.com/usds/justice40-tool/blob/emma-nechamkin/release/score-narwhal/data/data-pipeline/data_pipeline/etl/sources/geo_utils.py#L13 --- for a use like this, you can just import it from functools and then decorate the fixture and it should work. First load will be slow, but subsequent loads should come out of memory, unless pytest fixtures do dark magic.
| ) | ||
|
|
||
|
|
||
| # when should things be fixtures?? |
There was a problem hiding this comment.
General comment -- when should things be fixtures vs. constants? I've erred on the side of fixtures because that is how it looks in our codebase... but interested in comments and thoughts
There was a problem hiding this comment.
@esfoobar-usds I personally wouldn't make most of these fixtures --- I use fixtures for things I'm calculating (data or complicated mocks) that I want to reset with each test invocation, and I would put the rest just as constants I import to reduce the mental overhead with figuring out where a given name comes from (since fixtures have a little bit of magic to them in how they use DI). What's your-all's overall style though?
There was a problem hiding this comment.
This is SUPER helpful, @mattbowen-usds. TBH I have no idea on our structure, which is why (again, I am not an engineer!) I don't think I am the best choice for designing all of these tests. @esfoobar-usds could you please provide guidance?
The other thing I had in mind is ... are we doing the refactor? If we are, I was kind of working under the assumption that it would be easier to update the constants as fixtures... but this could very much be wrong!
| ] | ||
|
|
||
|
|
||
| def test_expected_pop_loss_percentile_columns( |
There was a problem hiding this comment.
this is really verbose, but the other option makes it much less modular. We also don't have a nice standardized way of going from pctile to threshold. Thoughts?
There was a problem hiding this comment.
I agree, this is no fun. Given the structural similarity and the broad scope of these, I'd be pretty okay with a single "test_percentile_columns" that just iterates over these columns, given the clarity of the assertions. I can draft up a version after lunch for folks to think over.
There was a problem hiding this comment.
That sounds really great to me!
| logger = get_module_logger(__name__) | ||
|
|
||
| ### TODO: we need to blow this out for all eight categories | ||
| def _helper_test_percentile_against_threshold( |
There was a problem hiding this comment.
this verbosely checks that all percentile comparisons are weakly greater.
|
** Score Deployed! **
|
|
** Score Deployed! **
|
|
** Score Deployed! **
|
|
This looks like a good set of tests! There are also some things we can think of more as "smell tests" (like The smell tests could be like -- there are no null values in a category for rows where we have data in all of the rows, etc. Or that if you |
TY @lucasmbrown-usds but also stop working. Yup -- a lot of those smell tests are included in these tickets: |
|
** Score Deployed! **
|
|
In another round of testing, we need to create these tests for the Islands |
| logger = get_module_logger(__name__) | ||
|
|
||
|
|
||
| # this is so slow? |
There was a problem hiding this comment.
Once it's loaded once, will it change? If not, we could lru_cache it to at least only ever have to load it once per thread. The other option would be to write a faster to load file format, like feather --- but that has problems of its own, since it could mask problems with the CSV or create new ones.
| ) | ||
|
|
||
|
|
||
| # when should things be fixtures?? |
There was a problem hiding this comment.
@esfoobar-usds I personally wouldn't make most of these fixtures --- I use fixtures for things I'm calculating (data or complicated mocks) that I want to reset with each test invocation, and I would put the rest just as constants I import to reduce the mental overhead with figuring out where a given name comes from (since fixtures have a little bit of magic to them in how they use DI). What's your-all's overall style though?
|
|
||
| @pytest.fixture | ||
| def donut_hole_adjacency_parameters(): | ||
| """Test that donut indicator uses the right threshold""" |
There was a problem hiding this comment.
I don't really understand this docstring --- this is not a test?
There was a problem hiding this comment.
Oh yikes, sorry -- the string should say "parameters to test"...
|
|
||
|
|
||
| @pytest.fixture | ||
| def final_score_df(): |
There was a problem hiding this comment.
I'm guessing this is slow too, and I think is duplicate below. This is a nit but we can probably factor some of these into a common fixtures.py
There was a problem hiding this comment.
yeah that sounds good to me!
| ] | ||
|
|
||
|
|
||
| def test_expected_pop_loss_percentile_columns( |
There was a problem hiding this comment.
I agree, this is no fun. Given the structural similarity and the broad scope of these, I'd be pretty okay with a single "test_percentile_columns" that just iterates over these columns, given the clarity of the assertions. I can draft up a version after lunch for folks to think over.
| donut_hole_income_threshold_and_column, | ||
| low_income_percentile, | ||
| ): | ||
| """We do it all! Check donut income and adjacency thresholds, and then that |
There was a problem hiding this comment.
This is my favorite doc string in the suite
|
** Score Deployed! **
|
|
** Score Deployed! **
|
|
** Score Deployed! **
|
| ) | ||
| non_natural_space = PercentileTestConfig( | ||
| field_names.TRACT_PERCENT_NON_NATURAL_FIELD_NAME, | ||
| field_names.NON_NATURAL_PCTILE_THRESHOLD, |
There was a problem hiding this comment.
these field names in particular make me so mad at myself lol
| unemployment, | ||
| low_median_income, | ||
| linguist_isolation, | ||
| poverty, |
There was a problem hiding this comment.
and then, just for my own accounting, we have the three boolean factors (aml, hrs, and fuds) and the composite fancy lead indicator. this is great!
|
** Score Deployed! **
|
|
I just need to fixup the build and whatever lint errors I missed. Should be able to do that first thing tomorrow. |
|
|
||
|
|
||
| def test_percentile_columns(final_score_df): | ||
| low_income = PercentileTestConfig( |
There was a problem hiding this comment.
So having done this, I started wanting to make it fancier --- like make the list happen automatically via some metaprogramming, and be able to impute the names of fields that follow a convention. But also I wanted to keep the test as obvious as possible, so I stuck with what I had. Honestly writing the test reinforced that the actual score code itself could use some refactoring to reduce duplication there, and then the test could rely on stronger conventions within tho score code to be further simplified.
There was a problem hiding this comment.
This seems right to me tbh. A lot of the score code, I think, was built early on and then just copy/pasted... I think a light-touch refactor could be really helpful.
|
** Score Deployed! **
|
mattbowen-usds
force-pushed
the
emma-nechamkin/1846-score-tests
branch
2 times, most recently
from
4c3a3ca
to
4eb7c2a
Compare
|
** Score Deployed! **
|
* Cleanup tests slightly before refactor (#1846) * Refactor score calculations tests * Feedback from review
This is annoying, but pylint in python3.8 was crashing parsing the named tuple. We weren't using any namedtuple-specific features, so I made the type a dataclass just to get pylint to behave.
mattbowen-usds
force-pushed
the
emma-nechamkin/1846-score-tests
branch
from
6f1c1bc
to
ef52014
Compare
|
We are explicitly punting the last two todos:
|
|
** Map Deployed! ** |
| - name: Run Smoketests | ||
| run: | | ||
| poetry run pytest data_pipeline/ -m smoketest |
| @@ -1,5 +1,5 @@ | |||
| import functools | |||
| from collections import namedtuple | |||
| from dataclasses import dataclass | |||
There was a problem hiding this comment.
YAY DATACLASS, DOWN WITH NAMEDTUPLE /cc @lucasmbrown-usds
| typename="ReversePercentile", | ||
| field_names=["field_name", "low_field_name"], | ||
| ) | ||
| @dataclass |
There was a problem hiding this comment.
Thank you! I will try another time with a branch off of Narwahl to see if the pylint errors resolved after you merge this
| from data_pipeline.score import field_names | ||
|
|
||
|
|
||
| @pytest.fixture(scope="session") |
|
** Map Deployed! ** |
|
** Map Deployed! ** |
|
this is great! ty for the catch! |
|
** Score Deployed! **
|
|
** Map Deployed! ** |
* Create deploy_be_staging.yml (#1575) * Imputing income using geographic neighbors (#1559) Imputes income field with a light refactor. Needs more refactor and more tests (I spotchecked). Next ticket will check and address but a lot of "narwhal" architecture is here. * Adding HOLC indicator (#1579) Added HOLC indicator (Historic Redlining Score) from NCRC work; included 3.25 cutoff and low income as part of the housing burden category. * Update backend for Puerto Rico (#1686) * Update PR threshold count to 10 We now show 10 indicators for PR. See the discussion on the github issue for more info: #1621 * Do not use linguistic iso for Puerto Rico Closes 1350. Co-authored-by: Shelby Switzer <[email protected]> * updating * Do not drop Guam and USVI from ETL (#1681) * Remove code that drops Guam and USVI from ETL * Add back code for dropping rows by FIPS code We may want this functionality, so let's keep it and just make the constant currently be an empty array. Co-authored-by: Shelby Switzer <[email protected]> * Emma nechamkin/holc patch (#1742) Removing HOLC calculation from score narwhal. * updating ejscreen data, try two (#1747) * Rescaling linguistic isolation (#1750) Rescales linguistic isolation to drop puerto rico * adds UST indicator (#1786) adds leaky underground storage tanks * Changing LHE in tiles to a boolean (#1767) also includes merging / clean up of the release * added indoor plumbing to chas * added indoor plumbing to score housing burden * added indoor plumbing to score housing burden * first run through * Refactor DOE Energy Burden and COI to use YAML (#1796) * added tribalId for Supplemental dataset (#1804) * Setting zoom levels for tribal map (#1810) * NRI dataset and initial score YAML configuration (#1534) * update be staging gha * NRI dataset and initial score YAML configuration * checkpoint * adding data checks for release branch * passing tests * adding INPUT_EXTRACTED_FILE_NAME to base class * lint * columns to keep and tests * update be staging gha * checkpoint * update be staging gha * NRI dataset and initial score YAML configuration * checkpoint * adding data checks for release branch * passing tests * adding INPUT_EXTRACTED_FILE_NAME to base class * lint * columns to keep and tests * checkpoint * PR Review * renoving source url * tests * stop execution of ETL if there's a YAML schema issue * update be staging gha * adding source url as class var again * clean up * force cache bust * gha cache bust * dynamically set score vars from YAML * docsctrings * removing last updated year - optional reverse percentile * passing tests * sort order * column ordening * PR review * class level vars * Updating DatasetsConfig * fix pylint errors * moving metadata hint back to code Co-authored-by: lucasmbrown-usds <[email protected]> * Correct copy typo (#1809) * Add basic test suite for COI (#1518) * Update COI to use new yaml (#1518) * Add tests for DOE energy budren (1518 * Add dataset config for energy budren (1518) * Refactor ETL to use datasets.yml (#1518) * Add fake GEOIDs to COI tests (#1518) * Refactor _setup_etl_instance_and_run_extract to base (#1518) For the three classes we've done so far, a generic _setup_etl_instance_and_run_extract will work fine, for the moment we can reuse the same setup method until we decide future classes need more flexibility --- but they can also always subclass so... * Add output-path tests (#1518) * Update YAML to match constant (#1518) * Don't blindly set float format (#1518) * Add defaults for extract (#1518) * Run YAML load on all subclasses (#1518) * Update description fields (#1518) * Update YAML per final format (#1518) * Update fixture tract IDs (#1518) * Update base class refactor (#1518) Now that NRI is final I needed to make a small number of updates to my refactored code. * Remove old comment (#1518) * Fix type signature and return (#1518) * Update per code review (#1518) Co-authored-by: Jorge Escobar <[email protected]> Co-authored-by: lucasmbrown-usds <[email protected]> Co-authored-by: Vim <[email protected]> * Update etl_score_geo.py Yikes! Fixing merge messup! * Create deploy_be_staging.yml (#1575) * Imputing income using geographic neighbors (#1559) Imputes income field with a light refactor. Needs more refactor and more tests (I spotchecked). Next ticket will check and address but a lot of "narwhal" architecture is here. * Adding HOLC indicator (#1579) Added HOLC indicator (Historic Redlining Score) from NCRC work; included 3.25 cutoff and low income as part of the housing burden category. * Update backend for Puerto Rico (#1686) * Update PR threshold count to 10 We now show 10 indicators for PR. See the discussion on the github issue for more info: #1621 * Do not use linguistic iso for Puerto Rico Closes 1350. Co-authored-by: Shelby Switzer <[email protected]> * updating * Do not drop Guam and USVI from ETL (#1681) * Remove code that drops Guam and USVI from ETL * Add back code for dropping rows by FIPS code We may want this functionality, so let's keep it and just make the constant currently be an empty array. Co-authored-by: Shelby Switzer <[email protected]> * Emma nechamkin/holc patch (#1742) Removing HOLC calculation from score narwhal. * updating ejscreen data, try two (#1747) * Rescaling linguistic isolation (#1750) Rescales linguistic isolation to drop puerto rico * adds UST indicator (#1786) adds leaky underground storage tanks * Changing LHE in tiles to a boolean (#1767) also includes merging / clean up of the release * added indoor plumbing to chas * added indoor plumbing to score housing burden * added indoor plumbing to score housing burden * first run through * Refactor DOE Energy Burden and COI to use YAML (#1796) * added tribalId for Supplemental dataset (#1804) * Setting zoom levels for tribal map (#1810) * NRI dataset and initial score YAML configuration (#1534) * update be staging gha * NRI dataset and initial score YAML configuration * checkpoint * adding data checks for release branch * passing tests * adding INPUT_EXTRACTED_FILE_NAME to base class * lint * columns to keep and tests * update be staging gha * checkpoint * update be staging gha * NRI dataset and initial score YAML configuration * checkpoint * adding data checks for release branch * passing tests * adding INPUT_EXTRACTED_FILE_NAME to base class * lint * columns to keep and tests * checkpoint * PR Review * renoving source url * tests * stop execution of ETL if there's a YAML schema issue * update be staging gha * adding source url as class var again * clean up * force cache bust * gha cache bust * dynamically set score vars from YAML * docsctrings * removing last updated year - optional reverse percentile * passing tests * sort order * column ordening * PR review * class level vars * Updating DatasetsConfig * fix pylint errors * moving metadata hint back to code Co-authored-by: lucasmbrown-usds <[email protected]> * Correct copy typo (#1809) * Add basic test suite for COI (#1518) * Update COI to use new yaml (#1518) * Add tests for DOE energy budren (1518 * Add dataset config for energy budren (1518) * Refactor ETL to use datasets.yml (#1518) * Add fake GEOIDs to COI tests (#1518) * Refactor _setup_etl_instance_and_run_extract to base (#1518) For the three classes we've done so far, a generic _setup_etl_instance_and_run_extract will work fine, for the moment we can reuse the same setup method until we decide future classes need more flexibility --- but they can also always subclass so... * Add output-path tests (#1518) * Update YAML to match constant (#1518) * Don't blindly set float format (#1518) * Add defaults for extract (#1518) * Run YAML load on all subclasses (#1518) * Update description fields (#1518) * Update YAML per final format (#1518) * Update fixture tract IDs (#1518) * Update base class refactor (#1518) Now that NRI is final I needed to make a small number of updates to my refactored code. * Remove old comment (#1518) * Fix type signature and return (#1518) * Update per code review (#1518) Co-authored-by: Jorge Escobar <[email protected]> Co-authored-by: lucasmbrown-usds <[email protected]> Co-authored-by: Vim <[email protected]> * Update etl_score_geo.py Yikes! Fixing merge messup! * updated to fix linting errors (#1818) Cleans and updates base branch * Adding back MapComparison video * Add FUDS ETL (#1817) * Add spatial join method (#1871) Since we'll need to figure out the tracts for a large number of points in future tickets, add a utility to handle grabbing the tract geometries and adding tract data to a point dataset. * Add FUDS, also jupyter lab (#1871) * Add YAML configs for FUDS (#1871) * Allow input geoid to be optional (#1871) * Add FUDS ETL, tests, test-datae noteobook (#1871) This adds the ETL class for Formerly Used Defense Sites (FUDS). This is different from most other ETLs since these FUDS are not provided by tract, but instead by geographic point, so we need to assign FUDS to tracts and then do calculations from there. * Floats -> Ints, as I intended (#1871) * Floats -> Ints, as I intended (#1871) * Formatting fixes (#1871) * Add test false positive GEOIDs (#1871) * Add gdal binaries (#1871) * Refactor pandas code to be more idiomatic (#1871) Per Emma, the more pandas-y way of doing my counts is using np.where to add the values i need, then groupby and size. It is definitely more compact, and also I think more correct! * Update configs per Emma suggestions (#1871) * Type fixed! (#1871) * Remove spurious import from vscode (#1871) * Snapshot update after changing col name (#1871) * Move up GDAL (#1871) * Adjust geojson strategy (#1871) * Try running census separately first (#1871) * Fix import order (#1871) * Cleanup cache strategy (#1871) * Download census data from S3 instead of re-calculating (#1871) * Clarify pandas code per Emma (#1871) * Disable markdown check for link * Adding DOT composite to travel score (#1820) This adds the DOT dataset to the ETL and to the score. Note that currently we take a percentile of an average of percentiles. * Adding first street foundation data (#1823) Adding FSF flood and wildfire risk datasets to the score. * first run -- adding NCLD data to the ETL, but not yet to the score * Add abandoned mine lands data (#1824) * Add notebook to generate test data (#1780) * Add Abandoned Mine Land data (#1780) Using a similar structure but simpler apporach compared to FUDs, add an indicator for whether a tract has an abandonded mine. * Adding some detail to dataset readmes Just a thought! * Apply feedback from revieiw (#1780) * Fixup bad string that broke test (#1780) * Update a string that I should have renamed (#1780) * Reduce number of threads to reduce memory pressure (#1780) * Try not running geo data (#1780) * Run the high-memory sets separately (#1780) * Actually deduplicate (#1780) * Add flag for memory intensive ETLs (#1780) * Document new flag for datasets (#1780) * Add flag for new datasets fro rebase (#1780) Co-authored-by: Emma Nechamkin <[email protected]> * Adding NLCD data (#1826) Adding NLCD's natural space indicator end to end to the score. * Add donut hole calculation to score (#1828) Adds adjacency index to the pipeline. Requires thorough QA * Adding eamlis and fuds data to legacy pollution in score (#1832) Update to add EAMLIS and FUDS data to score * Update to use new FSF files (#1838) backend is partially done! * Quick fix to kitchen or plumbing indicator Yikes! I think I messed something up and dropped the pctile field suffix from when the KP score gets calculated. Fixing right quick. * Fast flag update (#1844) Added additional flags for the front end based on our conversation in stand up this morning. * Tiles fix (#1845) Fixes score-geo and adds flags * Update etl_score_geo.py * Issue 1827: Add demographics to tiles and download files (#1833) * Adding demographics for use in sidebar and download files * Updates backend constants to N (#1854) * updated to show T/F/null vs T/F for AML and FUDS (#1866) * fix markdown * just testing that the boolean is preserved on gha * checking drop tracts works * OOPS! Old changes persisted * adding a check to the agvalue calculation for nri * updated with error messages * updated error message * tuple type * Score tests (#1847) * update Python version on README; tuple typing fix * Alaska tribal points fix (#1821) * Bump mistune from 0.8.4 to 2.0.3 in /data/data-pipeline (#1777) Bumps [mistune](https://github.com/lepture/mistune) from 0.8.4 to 2.0.3. - [Release notes](https://github.com/lepture/mistune/releases) - [Changelog](https://github.com/lepture/mistune/blob/master/docs/changes.rst) - [Commits](lepture/mistune@v0.8.4...v2.0.3) --- updated-dependencies: - dependency-name: mistune dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * poetry update * initial pass of score tests * add threshold tests * added ses threshold (not donut, not island) * testing suite -- stopping for the day * added test for lead proxy indicator * Refactor score tests to make them less verbose and more direct (#1865) * Cleanup tests slightly before refactor (#1846) * Refactor score calculations tests * Feedback from review * Refactor output tests like calculatoin tests (#1846) (#1870) * Reorganize files (#1846) * Switch from lru_cache to fixture scorpes (#1846) * Add tests for all factors (#1846) * Mark smoketests and run as part of be deply (#1846) * Update renamed var (#1846) * Switch from named tuple to dataclass (#1846) This is annoying, but pylint in python3.8 was crashing parsing the named tuple. We weren't using any namedtuple-specific features, so I made the type a dataclass just to get pylint to behave. * Add default timout to requests (#1846) * Fix type (#1846) * Fix merge mistake on poetry.lock (#1846) Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: Jorge Escobar <[email protected]> Co-authored-by: Jorge Escobar <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Matt Bowen <[email protected]> Co-authored-by: matt bowen <[email protected]> * just testing that the boolean is preserved on gha (#1867) * updated with hopefully a fix; coercing aml, fuds, hrs to booleans for the raw value to preserve null character. * Adding tests to ensure proper calculations (#1871) * just testing that the boolean is preserved on gha * checking drop tracts works * adding a check to the agvalue calculation for nri * updated with error messages * tribal tiles fix (#1874) * Alaska tribal points fix (#1821) * tribal tiles fix * disabling child opportunity * lint * removing COI * removing commented out code * Pipeline tile tests (#1864) * temp update * updating with fips check * adding check on pfs * updating with pfs test * Update test_tiles_smoketests.py * Fix lint errors (#1848) * Add column names test (#1848) * Mark tests as smoketests (#1848) * Move to other score-related tests (#1848) * Recast Total threshold criteria exceeded to int (#1848) In writing tests to verify the output of the tiles csv matches the final score CSV, I noticed TC/Total threshold criteria exceeded was getting cast from an int64 to a float64 in the process of PostScoreETL. I tracked it down to the line where we merge the score dataframe with constants.DATA_CENSUS_CSV_FILE_PATH --- there where > 100 tracts in the national census CSV that don't exist in the score, so those ended up with a Total threshhold count of np.nan, which is a float, and thereby cast those columns to float. For the moment I just cast it back. * No need for low memeory (#1848) * Add additional tests of tiles.csv (#1848) * Drop pre-2010 rows before computing score (#1848) Note this is probably NOT the optimal place for this change; it might make more sense for each source to filter its own tracts down to the acceptable tract list. However, that would be a pretty invasive change, where this is central and plenty of other things are happening in score transform that could be moved to sources, so for today, here's where the change will live. * Fix typo (#1848) * Switch from filter to inner join (#1848) * Remove no-op lines from tiles (#1848) * Apply feedback from review, linter (#1848) * Check the values oeverything in the frame (#1848) * Refactor checker class (#1848) * Add test for state names (#1848) * cleanup from reviewing my own code (#1848) * Fix lint error (#1858) * Apply Emma's feedback from review (#1848) * Remove refs to national_df (#1848) * Account for new, fake nullable bools in tiles (#1848) To handle a geojson limitation, Emma converted some nullable boolean colunms to float64 in the tiles export with the values {0.0, 1.0, nan}, giving us the same expressiveness. Sadly, this broke my assumption that all columns between the score and tiles csvs would have the same dtypes, so I need to account for these new, fake bools in my test. * Use equals instead of my worse version (#1848) * Missed a spot where we called _create_score_data (#1848) * Update per safety (#1848) Co-authored-by: matt bowen <[email protected]> * Add tests to make sure each source makes it to the score correctly (#1878) * Remove unused persistent poverty from score (#1835) * Test a few datasets for overlap in the final score (#1835) * Add remaining data sources (#1853) * Apply code-review feedback (#1835) * Rearrange a little for readabililty (#1835) * Add tract test (#1835) * Add test for score values (#1835) * Check for unmatched source tracts (#1835) * Cleanup numeric code to plaintext (#1835) * Make import more obvious (#1835) * Updating traffic barriers to include low pop threshold (#1889) Changing the traffic barriers to only be included for places with recorded population * Remove no land tracts from map (#1894) remove from map * Issue 1831: missing life expectancy data from Maine and Wisconsin (#1887) * Fixing missing states and adding tests for states to all classes * Removing low pop tracts from FEMA population loss (#1898) dropping 0 population from FEMA * 1831 Follow up (#1902) This code causes no functional change to the code. It does two things: 1. Uses difference instead of - to improve code style for working with sets. 2. Removes the line EXPECTED_MISSING_STATES = ["02", "15"], which is now redundant because of the line I added (in a previous pull request) of ALASKA_AND_HAWAII_EXPECTED_IN_DATA = False. * Add tests for all non-census sources (#1899) * Refactor CDC life-expectancy (1554) * Update to new tract list (#1554) * Adjust for tests (#1848) * Add tests for cdc_places (#1848) * Add EJScreen tests (#1848) * Add tests for HUD housing (#1848) * Add tests for GeoCorr (#1848) * Add persistent poverty tests (#1848) * Update for sources without zips, for new validation (#1848) * Update tests for new multi-CSV but (#1848) Lucas updated the CDC life expectancy data to handle a bug where two states are missing from the US Overall download. Since virtually none of our other ETL classes download multiple CSVs directly like this, it required a pretty invasive new mocking strategy. * Add basic tests for nature deprived (#1848) * Add wildfire tests (#1848) * Add flood risk tests (#1848) * Add DOT travel tests (#1848) * Add historic redlining tests (#1848) * Add tests for ME and WI (#1848) * Update now that validation exists (#1848) * Adjust for validation (#1848) * Add health insurance back to cdc places (#1848) Ooops * Update tests with new field (#1848) * Test for blank tract removal (#1848) * Add tracts for clipping behavior * Test clipping and zfill behavior (#1848) * Fix bad test assumption (#1848) * Simplify class, add test for tract padding (#1848) * Fix percentage inversion, update tests (#1848) Looking through the transformations, I noticed that we were subtracting a percentage that is usually between 0-100 from 1 instead of 100, and so were endind up with some surprising results. Confirmed with lucasmbrown-usds * Add note about first street data (#1848) * Issue 1900: Tribal overlap with Census tracts (#1903) * working notebook * updating notebook * wip * fixing broken tests * adding tribal overlap files * WIP * WIP * WIP, calculated count and names * working * partial cleanup * partial cleanup * updating field names * fixing bug * removing pyogrio * removing unused imports * updating test fixtures to be more realistic * cleaning up notebook * fixing black * fixing flake8 errors * adding tox instructions * updating etl_score * suppressing warning * Use projected CRSes, ignore geom types (#1900) I looked into this a bit, and in general the geometry type mismatch changes very little about the calculation; we have a mix of multipolygons and polygons. The fastest thing to do is just not keep geom type; I did some runs with it set to both True and False, and they're the same within 9 digits of precision. Logically we just want to overlaps, regardless of how the actual geometries are encoded between the frames, so we can in this case ignore the geom types and feel OKAY. I also moved to projected CRSes, since we are actually trying to do area calculations and so like, we should. Again, the change is small in magnitude but logically more sound. * Readd CDC dataset config (#1900) * adding comments to fips code * delete unnecessary loggers Co-authored-by: matt bowen <[email protected]> * Improve score test documentation based on Lucas's feedback (#1835) (#1914) * Better document base on Lucas's feedback (#1835) * Fix typo (#1835) * Add test to verify GEOJSON matches tiles (#1835) * Remove NOOP line (#1835) * Move GEOJSON generation up for new smoketest (#1835) * Fixup code format (#1835) * Update readme for new somketest (#1835) * Cleanup source tests (#1912) * Move test to base for broader coverage (#1848) * Remove duplicate line (#1848) * FUDS needed an extra mock (#1848) * Add tribal count notebook (#1917) (#1919) * Add tribal count notebook (#1917) * test without caching * added comment Co-authored-by: lucasmbrown-usds <[email protected]> * Add tribal overlap to downloads (#1907) * Add tribal data to downloads (#1904) * Update test pickle with current cols (#1904) * Remove text of tribe names from GeoJSON (#1904) * Update test data (#1904) * Add tribal overlap to smoketests (#1904) * Issue 1910: Do not impute income for 0 population tracts (#1918) * should be working, has unnecessary loggers * removing loggers and cleaning up * updating ejscreen tests * adding tests and responding to PR feedback * fixing broken smoke test * delete smoketest docs * updating click * updating click * Bump just jupyterlab (#1930) * Fixing link checker (#1929) * Update deps safety says are vulnerable (#1937) (#1938) Co-authored-by: matt bowen <[email protected]> * Add demos for island areas (#1932) * Backfill population in island areas (#1882) * Update smoketest to account for backfills (#1882) As I wrote in the commend: We backfill island areas with data from the 2010 census, so if THOSE tracts have data beyond the data source, that's to be expected and is fine to pass. If some other state or territory does though, this should fail This ends up being a nice way of documenting that behavior i guess! * Fixup lint issues (#1882) * Add in race demos to 2010 census pull (#1851) * Add backfill data to score (#1851) * Change column name (#1851) * Fill demos after the score (#1851) * Add income back, adjust test (#1882) * Apply code-review feedback (#1851) * Add test for island area backfill (#1851) * Fix bad rename (#1851) * Reorder download fields, add plumbing back (#1942) * Add back lack of plumbing fields (#1920) * Reorder fields for excel (#1921) * Reorder excel fields (#1921) * Fix formating, lint errors, pickes (#1921) * Add missing plumbing col, fix order again (#1921) * Update that pickle (#1921) * refactoring tribal (#1960) * updated with scoring comparison * updated for narhwal -- leaving commented code in for now * pydantic upgrade * produce a string for the front end to ingest (#1963) * wip * i believe this works -- let's see the pipeline * updated fixtures * Adding ADJLI_ET (#1976) * updated tile data * ensuring adjli_et in * Add back income percentile (#1977) * Add missing field to download (#1964) * Remove pydantic since it's unused (#1964) * Add percentile to CSV (#1964) * Update downloadable pickle (#1964) * Issue 105: Configure and run `black` and other pre-commit hooks (clean branch) (#1962) * Configure and run `black` and other pre-commit hooks Co-authored-by: matt bowen <[email protected]> * Removing fixed python version for black (#1985) * Fixup TA_COUNT and TA_PERC (#1991) * Change TA_PERC, change TA_COUNT (#1988, #1989) - Make TA_PERC_STR back into a nullable float following the rules requestsed in #1989 - Move TA_COUNT to be TA_COUNT_AK, also add a null TA_COUNT_C for CONUS that we can fill in later. * Fix typo comment (#1988) * Issue 1992: Do not impute income for null population tracts (#1993) * Hotfix for DOT data source DNS issue (#1999) * Make tribal overlap set score N (#2004) * Add "Is a Tribal DAC" field (#1998) * Add tribal DACs to score N final (#1998) * Add new fields to downloads (#1998) * Make a int a float (#1998) * Update field names, apply feedback (#1998) * Add assertions around codebook (#2014) * Add assertion around codebook (#1505) * Assert csv and excel have same cols (#1505) * Remove suffixes from tribal lands (#1974) (#2008) * Data source location (#2015) * data source location * toml * cdc_places * cdc_svi_index * url updates * child oppy and dot travel * up to hud_recap * completed ticket * cache bust * hud_recap * us_army_fuds * Remove vars the frontend doesn't use (#2020) (#2022) I did a pretty rough and simple analysis of the variables we put in the tiles and grepped the frontend code to see if (1) they're ever accessed and (2) if they're used, even if they're read once. I removed everything I noticed was not accessed. * Disable file size limits on tiles (#2031) * Disable file size limits on tiles * Remove print debugs I know. * Update file name pattern (#2037) (#2038) * Update file name pattern (#2037) * Remove ETL from generation (2037) I looked more carefully, and this ETL step isn't used in the score, so there's no need to run it every time. Per previous steps, I removed it from constants so the code is there it won't run by default. * Round ALL the float fields for the tiles (#2040) * Round ALL the float fields for the tiles (#2033) * Floor in a simpler way (#2033) Emma pointed out that all teh stuff we're doing in floor_series is probably unnecessary for this case, so just use the built-in floor. * Update pickle I missed (#2033) * Clean commit of just aggregate burden notebook (#1819) added a burden notebook * Update the dockerfile (#2045) * Update so the image builds (#2026) * Fix bad dict (2026) * Rename census tract field in downloads (#2068) * Change tract ID field name (2060) * Update lockfile (#2061) * Bump safety, jupyter, wheel (#2061) * DOn't depend directly on wheel (2061) * Bring narwhal reqs in line with main * Update tribal area counts (#2071) * Rename tribal area field (2062) * Add missing file (#2062) * Add checks to create version (#2047) (#2052) * Fix failing safety (#2114) * Ignore vuln that doesn't affect us 2113 https://nvd.nist.gov/vuln/detail/CVE-2022-42969 landed recently and there's no fix in py (which is maintenance mode). From my analysis, that CVE cannot hurt us (famous last words), so we'll ignore the vuln for now. * 2113 Update our gdal ppa * that didn't work (2113) * Don't add the PPA, the package exists (#2113) * Fix type (#2113) * Force an update of wheel 2113 * Also remove PPA line from create-score-versions * Drop 3.8 because of wheel 2113 * Put back 3.8, use newer actions * Try another way of upgrading wheel 2113 * Upgrade wheel in tox too 2113 * Typo fix 2113 Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: Emma Nechamkin <[email protected]> Co-authored-by: Shelby Switzer <[email protected]> Co-authored-by: Shelby Switzer <[email protected]> Co-authored-by: Emma Nechamkin <[email protected]> Co-authored-by: Matt Bowen <[email protected]> Co-authored-by: Jorge Escobar <[email protected]> Co-authored-by: lucasmbrown-usds <[email protected]> Co-authored-by: Jorge Escobar <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: matt bowen <[email protected]> Co-authored-by: matt bowen <[email protected]>
#1846
The goal of these tests are to ensure that the specific score calculations are correct. We use the true score output here and validate to make sure no calculation was obviously wrong. I think this is what our dear @mattbowen-usds would call "smoke tests".
Some outstanding tests
-
[ ] Test that HRS uses the right >= 3.25 threshold- [ ] Tests for the Island AreasLower priority