Number of calls to logpdf is not niterations * nsamples

[wouterwln]

See https://github.com/ReactiveBayes/ExponentialFamilyProjection.jl/tree/iterations-bug for a test showcasing this bug

[bvdmitri]

I doesn't seem like a bug because the number of calls can be fewer than niterations * nsamples when the projection converges more quickly and requires less iterations due to the tolerance setting. In your test, since you're using a simple function (return 0.0), it likely converges in just one (or few) iteration. To test this hypothesis, you could try setting tolerance = missing.