scikit-hep / iminuit / 34866917936

100%

push

github

fix: scipy() mutated options, misreported call limit, and stale covariance (#1203)

:robot: _AI text below_ :robot:

`Minuit.scipy()` had three related problems. It wrote `maxiter`,
`maxfev`, or `maxfun` into the `options` dict that the caller passed in,
so `opts = {"disp": False}; m.scipy(options=opts)` left `opts` with an
extra `maxiter` key. It passed the cumulative call count `self.nfcn` to
`FunctionMinimum`, so `m.migrad(); m.strategy = 0; m.scipy(ncall=100)`
reported `has_reached_call_limit` and `valid == False` although the
scipy run converged in about 75 calls. With `strategy == 0` and no exact
Hessian, it did not update `self._covariance`, so `m.covariance` still
showed the matrix from the previous `migrad()` call.

`options` is now copied before it is changed. `FunctionMinimum` receives
`r["nfev"]`, the number of calls in this run; `FMin.nfcn` still reports
the total, as for the other algorithms. `_make_covariance()` now runs
whenever `hesse()` does not.

Part of #1192

Co-authored-by: Iason Krommydas <iason.krom@gmail.com>

4 of 4 new or added lines in 1 file covered. (100.0%)

3961 of 3972 relevant lines covered (99.72%)

1.0 hits per line