search

tBuLi / symfit

Symbolic Fitting; fitting as it should be.
http://symfit.readthedocs.org
MIT License
233 stars 17 forks source link

scipy requirements #277

Closed Jhsmit closed 4 years ago

Jhsmit commented 4 years ago

The current scipy requirement for symfit 0.5.1 is scipy >= 1.0, <1.2, which could cause conda to install symfit 0.4.6 when the user has scipy >= 1.2 installed since it has more lenient requirements of only scipy >= 1.0.

Is the current version of symfit really incompatible with say scipy 1.3? Checking / fixing this would allow conda to distribute more the most recent version and thereby prevent bugs. See also: https://stackoverflow.com/questions/58341121/global-fitting-with-symfit-typestructure-of-dataset/

Also, personally I like to use the new workers kwargs for DifferentialEvolution which is available from scipy v1.2.

pckroon commented 4 years ago

On master with scipy==1.3.1 all tests pass. I have no clue about conda though. Maybe you could open a PR to fix this? :)

Jhsmit commented 4 years ago

yeah, I can have look at some point. What I remember is that there where (for my applications) some (temporal) performance issues with scipy >1.2

ocefpaf commented 4 years ago

I have no clue about conda though. Maybe you could open a PR to fix this? :)

conda and other downstream packagers are only following what is specified in https://github.com/tBuLi/symfit/blob/master/requirements.txt#L2, and that is scipy >= 1.0, <1.2.

Jhsmit commented 4 years ago

Scipy <1.2 is not compatible with python 3.8. I'll do some testing next week but since #283 runs without a problem on python 3.8 without this constraint I dont think there is any issue. I remember that in my applications the scipy >1.2 versions had some performance issues but I'll see if that is still the case

Jhsmit commented 4 years ago

I've tested my applications of symfit with scipy == 1.1.0 and scipy == 1.3.1 and didnt find any noticable difference.