Yohan Chatelain
Yohan Chatelain
I can disable the debug info, so BET should not complain. But it would be nice to check the return code and the .stderr attribute to avoid this kind of...
I just discovered that the [`decimal`](https://docs.python.org/3/library/decimal.html#module-decimal) package is part of the python standard library. That could be a good option to replace `bigfloat` without adding an extra dependency.
Hi @titeup, I did not work on it and actually it only offers arbitrary precision for Decimal representation so we'll have issue rounding numbers to binary precision. But we could...
Hi @effigies, Sure, I can do it. Which version should we compare 20.2.8 to? The previous version, 20.2.7?
Ok I see, so we need to check this patch against the 20.2.8 release version. I'll launch that.
Hi @effigies, the longest subject takes approximately 5 days to run with fuzzy, so I should have the results next week.
@effigies, I’ve finished the numerical stability test. [Here](https://github.com/yohanchatelain/fmriprep-reproducibility/blob/master/pull-929/plot.ipynb) is the notebook summarising the results. The reference was built from 30 Random Rounding repetitions computed with fMRIPrep 20.2.8, and the test...
The merge of #339 should solve the issues. We could reopen it if it's not the case.
FMA are now supported with #339 but not the reduce operations. I will open an issue for that.