search

arviz-devs / arviz

Exploratory analysis of Bayesian models with Python
https://python.arviz.org
Apache License 2.0
1.56k stars 388 forks source link

Tests `test_plot_ppc_discrete...` fail with `kde`. #2282

Closed GaetanLepage closed 7 months ago

GaetanLepage commented 9 months ago

Describe the bug The following tests fail on both the latest verion (0.16.1) and master.

To Reproduce Steps to reproduce the behavior. Ideally a self-contained snippet of code, or link to a notebook or external code. Please include screenshots/images produced with ArviZ here, or the stacktrace including arviz code to help.

git clone  git@github.com:arviz-devs/arviz.git
cd arviz
git checkout v0.16.1

# Setup the environment
python -m venv venv
source venv/bin/activate
pip install .
pip install ".[test]"
pip install pytest
pip install numba torch jax

pytest --strict -rsf --durations=20 -p no:warnings "arviz/tests/base_tests/test_plots_matplotlib.py::test_plot_ppc_discrete[True-kde]"
pytest --strict -rsf --durations=20 -p no:warnings "arviz/tests/base_tests/test_plots_matplotlib.py::test_plot_ppc_discrete_save_animation[kde]"

Expected behavior The tests pass.

Additional context Error logs:

``` e = TypingError('Failed in nopython mode pipeline (step: nopython frontend)\nNo implementation of function Function(\n """\n hist, bin_edges = np.histogram(data, bins=bins, range=range_hist)\n ^\n') issue_type = 'typing' def error_rewrite(e, issue_type): """ Rewrite and raise Exception `e` with help supplied based on the specified issue_type. """ if config.SHOW_HELP: help_msg = errors.error_extras[issue_type] e.patch_message('\n'.join((str(e).rstrip(), help_msg))) if config.FULL_TRACEBACKS: raise e else: > raise e.with_traceback(None) E numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend) E No implementation of function Function() found for signature: E E >>> histogram(array(int64, 1d, C), bins=unicode_type, range=none) E E There are 2 candidate implementations: E - Of which 2 did not match due to: E Overload in function 'np_histogram': File: numba/np/arraymath.py: Line 3912. E With argument(s): '(array(int64, 1d, C), bins=unicode_type, range=none)': E Rejected as the implementation raised a specific error: E TypingError: Failed in nopython mode pipeline (step: nopython frontend) E No implementation of function Function() found for signature: E E >>> le(unicode_type, int64) E E There are 22 candidate implementations: E - Of which 18 did not match due to: E Overload of function 'le': File: : Line N/A. E With argument(s): '(unicode_type, int64)': E No match. E - Of which 2 did not match due to: E Operator Overload in function 'le': File: unknown: Line unknown. E With argument(s): '(unicode_type, int64)': E No match for registered cases: E * (bool, bool) -> bool E * (int8, int8) -> bool E * (int16, int16) -> bool E * (int32, int32) -> bool E * (int64, int64) -> bool E * (uint8, uint8) -> bool E * (uint16, uint16) -> bool E * (uint32, uint32) -> bool E * (uint64, uint64) -> bool E * (float32, float32) -> bool E * (float64, float64) -> bool E - Of which 2 did not match due to: E Overload in function 'set_issubset': File: numba/cpython/setobj.py: Line 1623. E With argument(s): '(unicode_type, int64)': E Rejected as the implementation raised a specific error: E TypingError: All arguments must be Sets, got (unicode_type, int64) E raised from /scratch/alya/galepage/temp/arviz/venv/lib/python3.10/site-packages/numba/cpython/setobj.py:108 E E During: typing of intrinsic-call at /scratch/alya/galepage/temp/arviz/venv/lib/python3.10/site-packages/numba/np/arraymath.py (3974) E E File "venv/lib/python3.10/site-packages/numba/np/arraymath.py", line 3974: E def histogram_impl(a, bins=10, range=None): E E v = view.item() E if not bin_min <= v <= bin_max: E ^ E E raised from /scratch/alya/galepage/temp/arviz/venv/lib/python3.10/site-packages/numba/core/typeinfer.py:1086 E E During: resolving callee type: Function() E During: typing of call at /scratch/alya/galepage/temp/arviz/arviz/stats/density_utils.py (979) E E E File "arviz/stats/density_utils.py", line 979: E def histogram(data, bins, range_hist=None): E E """ E hist, bin_edges = np.histogram(data, bins=bins, range=range_hist) E ^ venv/lib/python3.10/site-packages/numba/core/dispatcher.py:409: TypingError ```

Versions and environmens:

``` aiohttp==3.8.6 aiosignal==1.3.1 appdirs==1.4.4 -e git+ssh://git@github.com/arviz-devs/arviz.git@4a8eb07c287d3efd3852309fc12ae49d747dda8c#egg=arviz asciitree==0.3.3 astor==0.8.1 async-timeout==4.0.3 attrs==23.1.0 beanmachine==0.2.0 black==22.3.0 botorch==0.6.6 charset-normalizer==3.3.0 click==8.1.7 clikit==0.6.2 cloudpickle==2.2.1 cmake==3.27.6 cmdstanpy==1.2.0 contourpy==1.1.1 crashtest==0.3.1 cycler==0.12.1 emcee==3.1.4 exceptiongroup==1.1.3 fasteners==0.19 ffmpeg==1.4 filelock==3.12.4 fonttools==4.43.1 frozenlist==1.4.0 fsspec==2023.9.2 functorch==2.0.0 gpytorch==1.8.1 graphviz==0.20.1 h5netcdf==1.2.0 h5py==3.9.0 httpstan==4.10.1 idna==3.4 iniconfig==2.0.0 jax==0.4.18 jaxlib==0.4.18 Jinja2==3.1.2 joblib==1.3.2 kiwisolver==1.4.5 lit==17.0.2 llvmlite==0.41.0 MarkupSafe==2.1.3 marshmallow==3.20.1 matplotlib==3.8.0 ml-dtypes==0.3.1 mpmath==1.3.0 multidict==6.0.4 multipledispatch==1.0.0 mypy-extensions==1.0.0 networkx==3.1 numba==0.58.0 numcodecs==0.12.0 numpy==1.25.2 numpyro==0.13.2 nvidia-cublas-cu11==11.10.3.66 nvidia-cublas-cu12==12.1.3.1 nvidia-cuda-cupti-cu11==11.7.101 nvidia-cuda-cupti-cu12==12.1.105 nvidia-cuda-nvrtc-cu11==11.7.99 nvidia-cuda-nvrtc-cu12==12.1.105 nvidia-cuda-runtime-cu11==11.7.99 nvidia-cuda-runtime-cu12==12.1.105 nvidia-cudnn-cu11==8.5.0.96 nvidia-cudnn-cu12==8.9.2.26 nvidia-cufft-cu11==10.9.0.58 nvidia-cufft-cu12==11.0.2.54 nvidia-curand-cu11==10.2.10.91 nvidia-curand-cu12==10.3.2.106 nvidia-cusolver-cu11==11.4.0.1 nvidia-cusolver-cu12==11.4.5.107 nvidia-cusparse-cu11==11.7.4.91 nvidia-cusparse-cu12==12.1.0.106 nvidia-nccl-cu11==2.14.3 nvidia-nccl-cu12==2.18.1 nvidia-nvjitlink-cu12==12.2.140 nvidia-nvtx-cu11==11.7.91 nvidia-nvtx-cu12==12.1.105 opt-einsum==3.3.0 packaging==23.2 pandas==2.1.1 pastel==0.2.1 pathspec==0.11.2 patsy==0.5.3 Pillow==10.0.1 platformdirs==3.11.0 plotly==5.17.0 pluggy==1.3.0 pylev==1.4.0 pyparsing==3.1.1 pyro-api==0.1.2 pyro-ppl==1.8.6 Pyro5==5.14 pysimdjson==5.0.2 pystan==3.7.0 pytest==7.4.2 python-dateutil==2.8.2 pytz==2023.3.post1 scikit-learn==1.3.1 scipy==1.11.3 serpent==1.41 six==1.16.0 stanio==0.3.0 statsmodels==0.14.0 sympy==1.12 tenacity==8.2.3 threadpoolctl==3.2.0 tomli==2.0.1 torch==2.0.1 tqdm==4.66.1 triton==2.0.0 typing_extensions==4.8.0 tzdata==2023.3 webargs==8.3.0 xarray==2023.9.0 xarray-datatree==0.0.12 xarray-einstats==0.6.0 yarl==1.9.2 zarr==2.16.1 ```
GaetanLepage commented 9 months ago

Note: the tests don't fail when numba is not installed.

OriolAbril commented 9 months ago

I think I know the reason, will try to fix it asap (might be next week or so).

We have a "conditional_jit" version of np.histogram, which uses numba's one if numba is installed. Here https://github.com/arviz-devs/arviz/blob/main/arviz/plots/backends/matplotlib/ppcplot.py#L417 we use "auto" as bins which is valid in numpy but not supported in numba (looks like by looking at the source code). My guess is that this used to silenly fall back to python mode and use numpy, but now the behaviour is more strict and it fails.

We could probably use our own "get_bins" function within https://github.com/arviz-devs/arviz/blob/main/arviz/stats/density_utils.py#L958 if bins is "auto" before passing the arguments down to numba

GaetanLepage commented 9 months ago

Thank you for the explanation ! In the meantime, we will disable those tests on our side as they are not indicating a critical bug of arviz.