As far as I understand, scipy 1.5 started using a different (more feature-rich) function to make numerical derivatives. This function is more strict on the output on SLSQP (the fortran minimize algorithm). SLSQP unexpectedly does not completely stay inside the parameter bounds you supply, giving errors / warnings from the new function.
This was only partially solved in scipy 1.5, and completely in 1.6. Trackpy unfortunately still suffers from the issue with 1.5.4, so I chose to emit a warning to recommend users to upgrade/downgrade scipy in this case. The unittests are skipped on scipy 1.5.
Closes #644
As far as I understand, scipy 1.5 started using a different (more feature-rich) function to make numerical derivatives. This function is more strict on the output on SLSQP (the fortran minimize algorithm). SLSQP unexpectedly does not completely stay inside the parameter bounds you supply, giving errors / warnings from the new function.
See related Scipy issue: https://github.com/scipy/scipy/pull/13009
This was only partially solved in scipy 1.5, and completely in 1.6. Trackpy unfortunately still suffers from the issue with 1.5.4, so I chose to emit a warning to recommend users to upgrade/downgrade scipy in this case. The unittests are skipped on scipy 1.5.