Is there any way to get OpenBLAS-based aesara through conda?

[VViCaiM]

We can get OpenBLAS-based numpy through conda install conda-forge::numpy "blas=*=openblas". But this doesn't work in aesara.

Additionally, will pip install aesara automatically install and uses OpenBLAS? I think that's the thing happen when pip install numpy. After installing aesara (pip) and TDM GCC, the error when import is

WARN: Could not locate executable g77
WARN: Could not locate executable f77
WARN: Could not locate executable ifort
WARN: Could not locate executable ifl
WARN: Could not locate executable f90
WARN: Could not locate executable DF
WARN: Could not locate executable efl
Output exceeds the [size limit](command:workbench.action.openSettings?[). Open the full output data [in a text editor](command:workbench.action.openLargeOutput?2067f149-bce9-45bf-b6ce-2e4155c13d65)
---------------------------------------------------------------------------
NoSectionError                            Traceback (most recent call last)
File d:\Python Kernel\lib\site-packages\aesara\configparser.py:237, in AesaraConfigParser.fetch_val_for_key(self, key, delete_key)
    236 try:
--> 237     return self._aesara_cfg.get(section, option)
    238 except InterpolationError:

File d:\Python Kernel\lib\configparser.py:782, in RawConfigParser.get(self, section, option, raw, vars, fallback)
    781 try:
--> 782     d = self._unify_values(section, vars)
    783 except NoSectionError:

File d:\Python Kernel\lib\configparser.py:1153, in RawConfigParser._unify_values(self, section, vars)
   1152     if section != self.default_section:
-> 1153         raise NoSectionError(section) from None
   1154 # Update with the entry specific variables

NoSectionError: No section: 'blas'

During handling of the above exception, another exception occurred:

KeyError                                  Traceback (most recent call last)
File d:\Python Kernel\lib\site-packages\aesara\configparser.py:353, in ConfigParam.__get__(self, cls, type_, delete_key)
    352 try:
--> 353     val_str = cls.fetch_val_for_key(self.name, delete_key=delete_key)
...
-> 2103 lines = (l.decode() for l in lines)
   2104 if parse:
   2105     selected_lines = []

UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd3 in position 2: invalid continuation byte

Sometimes, the error is different:

ValueError                                Traceback (most recent call last)
d:\Python Project\Final\test.ipynb Cell 3 in <cell line: 1>()
----> [1](vscode-notebook-cell:/d%3A/Python%20Project/Final/test.ipynb#ch0000002?line=0) import aesara

File d:\Python Kernel\lib\site-packages\aesara\__init__.py:126, in <module>
    122     return as_tensor_variable(x, **kwargs)
    125 # isort: off
--> 126 from aesara import scalar, tensor
    127 from aesara.compile import (
    128     In,
    129     Mode,
   (...)
    135     shared,
    136 )
    137 from aesara.compile.function import function, function_dump

File d:\Python Kernel\lib\site-packages\aesara\tensor\__init__.py:98, in <module>
     96 # adds shared-variable constructors
     97 from aesara.tensor import sharedvar  # noqa
---> 98 from aesara.tensor import (  # noqa
     99     basic_opt,
    100     blas,
    101     blas_c,
    102     blas_scipy,
...
---> 71     raise ValueError(f"The tag '{name}' is already present in the database.")
     73 if use_db_name_as_tag:
     74     if self.name is not None:

ValueError: The tag 'local_inplace_sparse_block_gemv' is already present in the database.

I think if using conda's m2w64-toolchain would be so easy. But conda seems force to use MKL. I try to use OpenBLAS instead of MKL because MKL performs worse than OpenBLAS on an AMD CPU.

Additional infomation:
The np.config.show() result:

openblas64__info:
    library_dirs = ['D:\\a\\numpy\\numpy\\build\\openblas64__info']
    libraries = ['openblas64__info']
    language = f77
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None)]
blas_ilp64_opt_info:
    library_dirs = ['D:\\a\\numpy\\numpy\\build\\openblas64__info']
    libraries = ['openblas64__info']
    language = f77
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None)]
openblas64__lapack_info:
    library_dirs = ['D:\\a\\numpy\\numpy\\build\\openblas64__lapack_info']
    libraries = ['openblas64__lapack_info']
    language = f77
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None), ('HAVE_LAPACKE', None)]
lapack_ilp64_opt_info:
    library_dirs = ['D:\\a\\numpy\\numpy\\build\\openblas64__lapack_info']
    libraries = ['openblas64__lapack_info']
    language = f77
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None), ('HAVE_LAPACKE', None)]
Supported SIMD extensions in this NumPy install:
    baseline = SSE,SSE2,SSE3
    found = SSSE3,SSE41,POPCNT,SSE42,AVX,F16C,FMA3,AVX2
    not found = AVX512F,AVX512CD,AVX512_SKX,AVX512_CLX,AVX512_CNL,AVX512_ICL

aesara version: 2.7.8
numpy version: 1.23.1
python version: 3.10.5

[twiecki]

You can use aesara-base for that purpose, it doesn't specify a compiler or blas implementation.

[VViCaiM]

You can use aesara-base for that purpose, it doesn't specify a compiler or blas implementation.

Thank you so much.