Build further examples for `everwillow`

[MoAly98]

Idea

We should try to build more examples to add to this project, demonstrating capabilities of everwillow beyond the current simple example. Examples we may want to try:

• A binned template fit (multiple bins)
• An unbinned model
• A combine model through pyhs3
• ?

[felixzinn]

Hi,

thanks for this library, looks really good
I was testing out the evermore example: https://github.com/MoAly98/everwillow/blob/2f1cd003718bd11244db992701a6ecda6cbe6c46/docs/python/test_evermore_example.py

I would like to perform a likelihood scan and was wondering what the best way for this would be in everwillow. what I am using now is this

@jax.jit
def fit(mu: float) -> float:
    result = ew.fit(
        partial(loss, args=(graphdef, static, hists)),
        params=sl.State.from_pytree(dynamic),
        observation=observation,
        fixed=sl.State.from_pytree({"mu": evm.Parameter(value=mu)}),
    )
    return result.nll

mu_values = jnp.linspace(0.0, 5.0, 21)
nll_values = jax.vmap(fit)(mu_values)

(using vmap for vectorization of fits #4 )

is this the way to do it?
is there an inbuilt method already?

would be nice to have an example of that as well, I think

[MoAly98]

hi @felixzinn, currently yes this is the way to vectorise for a likelihood scan. However for sure a an ew.scan() method can be provided to simplify this.