The goal of this issue is to identify the reason why the FiscalSim-US test is different from the OpenFisca-US test and to fix it so that the FiscalSim-US test gives the same output as the OpenFisca-US test.
The Problem
If you do the following steps:
fork and clone the fiscalsim-us repository
create the fiscalsim-us-dev conda environment: conda env create -f environment.yml
install the fiscalsim_us package into that conda environment: pip install -e .
You should be able to run the full set of tests on the repository using the make test command. When I do these same steps with the OpenFisca-US repository (fork, clone, create openfisca-us-test conda env, install openfisca_us package), I get everything passing (plus 2 appropriately skipped tests of test_acs.py), except for the 5 tests of test_against_taxsim.py. These fail for a known reason that the current tests only work on the Linux operating system.
The problem seems to be something with the hdf5 data from the CPS and the interface with the `openfisca-tools` package, which is a dependency of this project (for now). I get the following error message in each case.
/opt/anaconda3/envs/fiscalsim-us-dev/lib/python3.7/site-packages/openfisca_tools/microsimulation.py: ...
...
E KeyError: "Unable to open object (object 'person_id' doesn't exist)"
Another error seems to be in the interface between the `openfisca-tools/data/dataset.py` file and the `fiscalsim_us/data/datasets/cps/cps.py` file and its hdf5 implementation. The error traceback is the following.
/opt/anaconda3/envs/fiscalsim-us-dev/lib/python3.7/site-packages/openfisca_tools/data/dataset.py:69: in wrapper
...
fiscalsim_us/data/datasets/cps/cps.py:36: in generate
cps = h5py.File(self.file(year), mode="w")
raw_data = <HDF5 file "raw_cps_2021.h5" (mode r)>
self = <fiscalsim_us.data.datasets.cps.cps.CPS object at 0x7f9c2de63650>
...
/opt/anaconda3/envs/fiscalsim-us-dev/lib/python3.7/site-packages/h5py/_hl/files.py:533: in init
...
E OSError: Unable to create file (unable to truncate a file which is already open)
## What has been changed in FiscalSim-US
`FiscalSim-US` has had [5 PRs](https://github.com/TheCGO/fiscalsim-us/pulls?q=is%3Apr+is%3Aclosed) up to now that mainly change values of `openfisca-us` to `fiscalsim-us` in hundreds of files. I suspect that something in the process messed up the interface between the `openfisca-tools` package, the data requests in `cps.py` and the `h5py` package.
cc: @austinperryfrancis @ss7886
The goal of this issue is to identify the reason why the
FiscalSim-UStest is different from theOpenFisca-UStest and to fix it so that theFiscalSim-UStest gives the same output as theOpenFisca-UStest.The Problem
If you do the following steps:
fiscalsim-usrepositoryfiscalsim-us-devconda environment:conda env create -f environment.ymlfiscalsim_uspackage into that conda environment:pip install -e .You should be able to run the full set of tests on the repository using the
make testcommand. When I do these same steps with the OpenFisca-US repository (fork, clone, createopenfisca-us-testconda env, installopenfisca_uspackage), I get everything passing (plus 2 appropriately skipped tests oftest_acs.py), except for the 5 tests oftest_against_taxsim.py. These fail for a known reason that the current tests only work on the Linux operating system.When I run the tests in the
fiscalsim-us-devenvironment, I get the following output. The differences are that:test_against_taxsim.pytests are now errors "E" instead of failures "F"test_microsim.pytest failstest_cps.pytest failstest_cps.pytest gets skippedfiscalsim_us/tests/test_variables.py ................................... [ 3%] ........................................................................ [ 10%] ........................................................................ [ 16%] ........................................................................ [ 23%] ........................................................................ [ 30%] ........................................................................ [ 37%] ........................................................................ [ 44%] ........................................................................ [ 50%] ........................................................................ [ 57%] ........................................................................ [ 64%] ........................................................................ [ 71%] ........................................................................ [ 78%] ........................................................................ [ 84%] ........................................................................ [ 91%] ........................................................................ [ 98%] .. [ 98%] fiscalsim_us/tests/code_health/parameters.py . [ 98%] fiscalsim_us/tests/code_health/variable_names.py . [ 98%] fiscalsim_us/tests/microsimulation/test_against_taxsim.py EEEEE [ 99%] fiscalsim_us/tests/microsimulation/test_microsim.py F [ 99%] fiscalsim_us/tests/microsimulation/data/test_imports.py .. [ 99%] fiscalsim_us/tests/microsimulation/data/acs/test_acs.py ss [ 99%] fiscalsim_us/tests/microsimulation/data/cps/test_cps.py Fs [100%]
/opt/anaconda3/envs/fiscalsim-us-dev/lib/python3.7/site-packages/openfisca_tools/microsimulation.py: ... ... E KeyError: "Unable to open object (object 'person_id' doesn't exist)"
/opt/anaconda3/envs/fiscalsim-us-dev/lib/python3.7/site-packages/openfisca_tools/data/dataset.py:69: in wrapper ... fiscalsim_us/data/datasets/cps/cps.py:36: in generate cps = h5py.File(self.file(year), mode="w") raw_data = <HDF5 file "raw_cps_2021.h5" (mode r)> self = <fiscalsim_us.data.datasets.cps.cps.CPS object at 0x7f9c2de63650> ... /opt/anaconda3/envs/fiscalsim-us-dev/lib/python3.7/site-packages/h5py/_hl/files.py:533: in init ... E OSError: Unable to create file (unable to truncate a file which is already open)