Call limit reached in hypotest fucntion

[ssaha1234]

Dear pyhf users,
I am running a hypotest with several signal hypotheses as shown below -

test_mus = np.linspace(0.2, 3, 60)
results = [
    pyhf.infer.hypotest(
        test_mu, data, model, return_expected_set=True, return_tail_probs=True
    )
    for test_mu in test_mus
]

Unfortunately, the Call Limit is being reached in the fit. I tried adding maxiter=1000000 inside the function, but I get an error saying this is not a valid argument. Is there a way, I can add maximum iterations in the hypotest ? Let me know if you would need more details about the likelihood, or the errors.
Thanks a lot !

Best,
Shreya

[matthewfeickert]

@ssaha1234 maxiter is controlled by the optimizer (c.f. docs for the different optimizers). So you'll need to set that attribute before you perform inference.

Probably the most straightforward way to do this (in pyhf v0.6.3) is during the backend setup. Here's an example using NumPy and the SciPy optimizer

import pyhf
print(f"pyhf version: v{pyhf.__version__}")  # pyhf version: v0.6.3
pyhf.set_backend("numpy", pyhf.optimize.scipy_optimizer(maxiter=2000000))
backend, optimizer = pyhf.get_backend()
print(f"{optimizer.maxiter=}")  # optimizer.maxiter=2000000

Though I'll note that if your fit is struggling here bumping the maxiter probably won't help, as iteration time isn't going to be the underlying problem.

[kratsg]

Here's a working example of what you intended to do. Note that a hypotest call will instantiate multiple optimizers (because it's not a single fit, but at least 5 separate fits using the specified optimizer).

import pyhf

new_optimizer = pyhf.optimize.scipy_optimizer(maxiter=1000000)
pyhf.set_backend(pyhf.tensorlib, new_optimizer)

model = pyhf.simplemodels.uncorrelated_background([5], [10], [2.5])
data = [15] + model.config.auxdata
test_mu = 1.0

result = pyhf.infer.hypotest(
    test_mu, data, model, return_expected_set=True, return_tail_probs=True,
)

[ssaha1234]

Hi @matthewfeickert and @kratsg,
Thank you very much for the information and code snippets, I will implement it and check if I can make the fit work.
@matthewfeickert I have seen a similar error before for a mle fit (but that is a single fit...)- #1751, maxiter did solve the issue, which is why I thought it would work out here as well :)). I will try it out and let you know !
Thanks !
Best,
Shreya

[ssaha1234]

Hi @matthewfeickert and @kratsg,
My fits worked when I increased the iterations ! Thanks a lot for all the help :))
Best,
Shreya