Giacomo Petrillo

I modified `Lmer.fit` to pass `**kwargs` to `lmer.glmer`. It works since `nAGQ` is just an integer.

I sometimes would like at least to have a different error variance parameter on different groups.

Ok

Ok, sorry. The minimal example is: ```python @jax.vmap def f(x): a = x[0:6].reshape(2, 3) b = x[6:9].reshape(3) result = jax.ShapeDtypeStruct((2, 3), x.dtype) return jax.pure_callback(jnp.add, result, a, b, vectorized=True) f(jnp.ones((5,9))) ```...

On second thought, the behavior I expect requires `pure_callback` to treat `callback` as an ufunc (vs. a generalized ufunc). This is not stated in the documentation. However, since with `vectorize=False`...

If someone encounters this same issue, in the meantime here is the workaround I'm going to use: ```python def pure_callback_ufunc(callback, dtype, *args, **kwargs): shape = jnp.broadcast_shapes(*(arg.shape for arg in args))...

I know this is the current behavior as defined, but I disagree it is "expected" by the user (me, in particular). It is surprising. It surely breaks ufuncs, which is...

I'd like this feature because I copy images a lot. I searched for code to do that online, found this: [https://stackoverflow.com/a/59461882/3942284](https://stackoverflow.com/a/59461882/3942284) ```js function copyElementToClipboard(element) { window.getSelection().removeAllRanges(); let range = document.createRange();...

I found an alternative to surfingkeys with this feature: https://github.com/gdh1995/vimium-c

Ok, I'll open a pull request then